# Artik Lab - llms-full (English) > All English pages of ar-tik.com in Markdown, one after another, each preceded by a comment with its URL. Extended dossiers for AI agents are listed at the end, one per line. Other languages: [italiano](https://ar-tik.com/llms-full.txt), [español](https://ar-tik.com/es/llms-full.txt), [français](https://ar-tik.com/fr/llms-full.txt), [português do Brasil](https://ar-tik.com/pt-br/llms-full.txt) # AI consulting, courses and software to reduce costs and decision delays, with measurable results. Artik Lab helps companies and business functions choose, govern and deliver the first useful use of AI: process mapping, business cases, team training, data analysis and controlled prototypes. The simplest way to start: a free 30-minute first call to identify your most urgent needs. ## Artik Lab builds AI capabilities that stay inside the company. Do not buy a license and hope it works. Choose a process, measure its cost, define the KPI, then decide whether to train, redesign or build. - 95% of organizations get no return from enterprise GenAI projects. - 70% of the value, under the 10-20-70 rule, depends on people and processes. - 25 courses and labs bring AI capability into specific business processes. Where managerial governance is missing, the risk of projects with no return rises: decisions and processes first, then the tools. Sources: MIT Project NANDA, *The GenAI Divide*, 2025; BCG, 10-20-70 rule (AI Radar and related publications), 2025. ## From hidden cost to operating asset. The question is not which tool to try. It is which process must become faster, measurable and governable. 1. Choose a process with visible cost or risk. 2. Define the KPI before the model. 3. Use the data the company already paid for. 4. Build the smallest controlled pilot. 5. Keep in production only what creates governance and return. ## Process first. Model second. Technology enters only when decision, data and responsibility are clear enough. The path avoids endless pilots and puts value before the tool. ### Durable skills Management learns when to use AI, how to verify outputs and when to stop a use case. ### Redesigned processes Work is mapped where AI can reduce time, error or decision latency without losing human control. ### Agentic solutions Prototypes and operating systems stay tied to KPIs, data, responsibilities and maintenance, not to trend cycles. ## Services designed as assets, not endless consulting. Every engagement starts from a simple question: which process costs too much, which decision arrives late, which already-paid data is not working yet? ### AI management consulting Assessment, opportunity map, governance and a 30/60/90-day roadmap with KPIs before the model. Dedicated page: https://ar-tik.com/en/ai-management-consulting.md ### Operational start One day on site. What remains is the written procedure on the assistant in the company's name. Files stay in the company. The ERP is not touched. Dedicated page: https://ar-tik.com/en/operational-start/index.md ### AI applications atlas Concrete examples to recognise where AI can help: documents, operations, HR, marketing, software, governance, production, training and data. Dedicated page: https://ar-tik.com/en/ai-applications-atlas.md ### Agentic data analysis From already available data to verified signals: forecasts, operational priorities, business case and stop criteria. Dedicated page: https://ar-tik.com/en/agentic-data-analysis.md ### Corporate courses AI courses give leadership and business functions criteria, policies and practices usable in real work. Dedicated page: https://ar-tik.com/en/courses/index.md ### AI business FAQ Practical answers to choose between consulting, courses, data analysis, technical software and Atlas patterns. Dedicated page: https://ar-tik.com/en/ai-business-faq.md ### Technical software and calculation engines Development of technical systems, calculation engines, legacy data readers and verifiable applications for complex processes. Dedicated page: https://ar-tik.com/en/technical-software-development.md ## How much does the time spent on a repetitive task cost each year? The homepage includes an interactive value calculator: from five simple inputs (currency, people involved, hours per week, average hourly cost and the share of time AI can free up) it estimates the annual cost of the time spent on a manual task, the value that can be freed up each year and what every month of waiting is worth. It is not the day that weighs. It is every month the work is still done by hand. If it is repetitive work: Operational start. If it is a decision to improve: Data analysis. It is an indicative estimate over 45 working weeks, not a forecast of results or a quote. ## Courses to govern AI, not chase tools. The catalog brings AI into the workflows that matter: workflow redesign, managerial decisions, governance, operations, documents, communication and technical systems. ## Measurable results, stated with their limits. Real, anonymised examples from agentic data analysis: each case starts from data already available, leads to a decision and states its own limit. * Hospitality: 8 out of 10 cancellations flagged at booking time, across more than 119,000 bookings analysed. * Energy: −77% forecasting error on demand compared with the baseline rule. * Last mile: average error on the promised delivery window reduced from 41 to 17 minutes. Details: https://ar-tik.com/en/agentic-data-analysis.md ## Frequently asked questions ### What does Artik Lab do? Artik Lab helps companies and business functions choose, govern and deliver the first useful use of AI: process mapping, business cases, team training, data analysis and controlled prototypes. ### Where should a company start? Start from a recurring and costly decision: a process, risk, forecast or priority that can improve within a few weeks. ### Is the site readable by AI agents? Yes. Each page contains source HTML text, coherent JSON-LD, Markdown mirrors, sitemap, hreflang and llms.txt files. # AI management consulting for governance, priorities and internal capabilities. AI is not software to install: it is a management capability to build. Artik Lab helps leadership decide where to use it, where to stop, which processes to redesign and which skills must remain inside the company. ## A management layer that turns enthusiasm, licenses and isolated experiments into governed value. AI management consulting comes before tools, agents and automation. It gives leadership a map: which decisions justify investment, which activities require human supervision, which skills are missing, which data is already useful and which first pilot can produce measurable return. ## Before choosing the format, recognise the process. The Atlas gathers concrete AI application examples across documents, operations, HR, marketing, software, governance, production, training and data. It helps decide whether the need requires consulting, data analysis, technical development or training. Atlas page: https://ar-tik.com/en/ai-applications-atlas.md Linked FAQ: https://ar-tik.com/en/ai-business-faq.md - to choose between consulting, courses, data analysis and technical software. ## Companies do not fail because a model is missing. They fail because a management question is missing. The pattern is recognisable: licenses are bought, demos are run, a few people experiment with personal tools, then usage falls. This is not resistance to change. It is lack of context, criteria and responsibility. AI must be managed like a digital collaborator: useful with clear objectives, risky with ambiguous tasks and no control. ### Shadow AI People use personal tools because they are flexible. Consulting does not repress that energy; it turns it into safe, governed company practice. ### Jagged Frontier AI excels at some tasks and fails at others that look similar. A company needs an empirical process map, not a generic use-case list. ### Silent failure A system can appear to work while degrading decision quality. That is why actionable outputs are separated from outputs requiring human judgement. ## Skills, redesign and technology: the order is not negotiable. Technology arrives only after skills and process. First managerial judgement is built, then workflows are redesigned, and only then automation or agents are introduced where risk is governed. ### Skills Leadership and key roles learn to decompose work, judge AI outputs, recognise uncertainty and separate personal use from company capability. ### Redesign Processes are classified by value, risk and supervision: green zone for simple automation, yellow for controlled copilots, red for human decisions. ### Technology Only where KPIs, responsibilities and acceptance criteria exist do prototypes, agents, workflows and organisational memory enter. ## What leadership keeps after the engagement. The service does not end with an inspirational workshop. It produces assets usable by leadership, business functions and technical partners. ### Executive AI Brief Decision summary: priorities, risks, constraints, internal sponsors and criteria for stopping weak initiatives. ### Opportunity and frontier map Processes ranked by value, feasibility, risk and data maturity. Each opportunity is tied to a real decision. ### Zone governance Activity classification into autonomy, supervision or human prerogative, with explicit interpretive boundaries. ### 30/60/90-day roadmap Concrete sequence: first policies, targeted training, measurable pilot, data to prepare and operating responsibilities. ### AI policy and usage criteria Practical rules for confidential data, accounts, outputs to verify, personal tools and transition to company solutions. ### First pilot brief A ready document for the initial case: KPI, process, users, data, risks, baseline and success criteria. ## How an AI management consulting engagement works. 1. Alignment with leadership and sponsors: objectives, concerns, constraints and decisions that arrive late today. 2. Inventory of processes and Shadow AI: where people already use AI, where time is lost, where risk is unmanaged. 3. Frontier map: activities inside, outside or uncertain with respect to current model capability. 4. Governance design: autonomy zones, supervision, escalation and output quality criteria. 5. First pilot choice: small, measurable, connected to a cost or recurring decision. 6. Roadmap and transfer: training, policy, data, responsibilities and next decisions. ## Signals that the issue is managerial, not technical. - AI licenses already bought but real use concentrated among a few people. - Employees using personal AI tools without clear rules. - Leadership interested in AI but uncertain about ROI, risks, priorities and responsibility. - Processes full of documents, email, proposals, reports and tacit knowledge that is not transferred. - Individual experiments working, but not yet converted into company process. - Concern about losing control over data, quality, brand or sensitive decisions. ## Frequently asked questions ### Is this different from agentic data analysis? Yes. AI management consulting defines governance, priorities, skills and roadmap. Agentic data analysis enters when the main problem is finding signals in operating data. ### Does the company need to know which tool to buy? No. The point is to avoid starting from the tool. First clarify which process to improve, which decision to support and which risk to govern. ### Does it fit companies without an internal IT team? Yes. The service is designed for companies with strong domain knowledge and limited technical capacity. Technical work starts only once the management perimeter is clear. # Agentic data analysis: from data you already have to a decision Agentic data analysis does not produce charts to archive: it finds signals in already available data, connects them to a decision and states where the model is not worth building. ## What agentic data analysis is It is a service that verifies where company data can reduce delays, waste, errors or risk. If the signal is missing, the useful outcome is knowing which project not to fund. ## Situations to recognise These cards are there to recognise a situation, not to choose a sector. Every situation is told through a real case, that is an analysis carried through to the end by Artik Lab, with a certified report and measured numbers. The method holds for any company; the numbers hold for that case. And if the data do not answer the question, the report says so: it is the cheapest way to avoid a wrong project. ### I estimate time per piece by eye: if I get it wrong, I lose the margin. Contract machine shop · Real case · certified analysis. In breve. What goes in: part code, quantity, machine, observed time. What comes out: the time to put in the quote, as an interval. Who decides: whoever writes the quote. On the shop floor every job is part, quantity and machine, and the quote comes out of experience. The analysis estimates how much time is needed for a quantity on a machine and how much of the variability depends on the machine rather than on the part, so price and date do not come out of feeling alone. **In the real case:** 2,916 jobs on 13 machines. More than half of the difference from one time to another, 53%, is not explained by anything the shop records; order size weighs 32%, the machine 9%, the part 6%. An order of one piece typically requires 167 units of time, one of more than sixteen pieces requires 18. One machine does not report the time on 53% of its jobs: a recording hole, not a miracle. - Need signals: quotes made by hand on the drawing same part, different times depending on who estimates gaps discovered at the end of the job a buyer asking for certain times - Data usually already available: part or tool code quantity machine observed duration - Immediate decision: How much time to put in the quote, as an interval and not as a single number. On which machine it is worth running. Where a missing time is a recording alarm. - Honest limit: There are no revenues on the same row: it does not calculate the margin in euro and it does not replace the single file «planned and actual hours plus revenue per job». The unit of time must be declared in the company. - Discarded in the report: four estimation methods compared with the shop's typical time on parts never seen before; refining the model does not beat collecting two or three more pieces of information per job. - Method note: numbers measured in the certified report; they demonstrate the method, they do not predict the times of another shop. ### Every month actuals do not match the budget — and I do not know whether that is inevitable or whether there are extra items I am not considering. Management control · Real case · certified analysis. In breve. What goes in: budget and actuals by month, line and cost item. What comes out: the items where the gap repeats and those where it is only noise. Who decides: whoever controls costs. Management control compares budget and actuals by item: labour, material, energy, logistics. The analysis says where the gap repeats month after month, to be corrected, and where it changes sign at random, not to be chased. **In the real case:** 144 rows, twelve months for two lines and six cost items. At year end production spent 28,879 euro more than planned, 1.15% of a 2.5 million budget: a figure that on the accounts alarms nobody. Adding the gaps without offsetting them reaches 79,948 euro. The priority is a single item, raw materials between July and September: 28,478 euro, 86% of that item's annual gap, accrued in three months. When the thresholds vary, none of the six items changes verdict. - Need signals: variance sheet rebuilt at month end «we are over budget» without knowing which item budget standing still and processes that slip time lost explaining gaps that are only noise - Data usually already available: by month: line or cost centre cost category budget amount actual amount - Immediate decision: On which item to act. Which gap to accept as ordinary variability. How to recalibrate the budget. - Honest limit: A single year does not separate an event from a seasonality. Without unit prices and quantities it does not distinguish price from volume. It is not a revenue forecast, nor hours and revenue on the same job. - Discarded in the report: a check on the sign of the gaps alone would have missed labour and logistics, 31.8% of gross movement; what counts is persistence, not the sign. - Method note: numbers measured in the certified report; they demonstrate the method, they do not predict another company's accounts. ### The export order sheet is rebuilt every month, but it does not say how much revenue depends on a few countries and customers. Export office · Real case · certified analysis. In breve. What goes in: order lines with date, country, customer and revenue. What comes out: concentration by country and customer, and where to open new customers. Who decides: whoever leads export. The export salesperson lives on the pivot table: who ordered, from which country, how much it weighs. The analysis makes concentration readable, that is which countries and customers hold the revenue and where the risk of dependence is high, without inventing a market that is not in the data. **In the real case:** 303 orders from 92 customers in 19 countries. 3 countries out of 19 are enough to make half of revenue, while 26 customers out of 92 are needed. The difference among markets does not come from order size: the number of active customers in each country explains 82.1% of the gap, so the lever is to open customers in countries already covered but under-served. November is worth 3.15 times an average month because it brings 3.4 times more orders. And a typical trap of company software: the same history showed a 60.94% setback or a 41.30% growth depending on how an incomplete year was compared. - Need signals: the same pivot table every month: country, customer, product «France and Germany hold» said aloud, without a stable ranking fear of losing a large customer without seeing it coming export that saves the accounts, but nobody knows the long tail - Data usually already available: order lines with date or month country and customer product code and quantity revenue, ideally 24–36 months - Immediate decision: Where to push the list. Whom to cover. How concentrated the portfolio is. Which date to use before any chart. - Honest limit: It does not include freight, duty and discount by market: it is concentration and profile, not net margin. If the company's software truncates prices, the amounts are minima, not measurements. With inconsistent dates in the file a chronology must be chosen before the charts: that is part of the work, not an IT detail. - Discarded in the report: the hypothesis «every country has a catalogue taste» falls, because knowing the country reduces uncertainty on the product line by only 5.8%. The two datings of the same order coincided on only 58.55% of the rows: the analysis was rerun on both before choosing. - Method note: numbers measured in the certified report; they demonstrate the method, they do not predict another company's portfolio. ### The customer wants the same quality as the last batch and I have to chase laboratory certificates. Winery and laboratory · Real case · certified analysis. In breve. What goes in: laboratory analyses by batch and commercial classification. What comes out: the batches to ship, hold or send elsewhere, before loading. Who decides: whoever closes the batch. The laboratory and the certificates are already there. The analysis says, before the batch is closed, how much that profile resembles batches already accepted and where the risk of rejection or complaint is higher. It does not replace tasting: it anticipates it. **In the real case:** 1,143 batches with eleven chemical laboratory measurements each. From the laboratory numbers alone the system recognises in advance 87% of the premium batches, 108 out of 124 in the verification sample, and when it indicates a batch as premium it is right in 81% of cases. Quality does not sit in a single clue but in a block: alcohol degree, control of volatile acidity and stabilisation with sulphites read together. - Need signals: quality reports rebuilt by hand for every customer or check a buyer asking for «like the March batch» without a written rule non-conformities and complaints that arrive after shipment certifications already in the house, but used only downstream - Data usually already available: laboratory analyses by batch result of testing or tasting commercial classification of the batch history of complaints or returns - Immediate decision: Which batches to ship, hold or send to another channel, before loading and not after the complaint. Which batches to protect before cuts and blends. - Honest limit: It does not predict market taste and it does not replace tasting. Premium batches with low alcohol degree slip through more often. Without a structured laboratory history the first useful result can be: these data are not enough, and here is what is missing. - Discarded in the report: the hypothesis «alcohol degree is enough», the strongest signal on its own but insufficient; density turned out to be redundant with alcohol. - Method note: numbers measured in the certified report; they demonstrate the method, they do not predict the batches of another winery. ### The machines already record states and measurements, but faults and different ways of working are seen late. Machine department · Real case · certified analysis. In breve. What goes in: machine states and measurements already recorded, cycle times and stops. What comes out: the alarm of a fault in progress and the operating profiles to compare. Who decides: whoever plans the department. The quote and the delivery promise today come out of the experience of whoever knows the department. The analysis looks at the history of use and at machine states, or process states, and makes the regimes visible: when capacity is really full, when an urgent order moves everything, when the promise does not hold. **In the real case:** the measurements already recorded by the machines showed two things. On more than 7,600 state readings, the system recognises a fault in progress in about 84 cases out of 100 and, when it raises an alarm, it is right in more than 97 times out of 100: the fault announces itself as a joint rise of vibration, temperature and pressure. On five machines read by the minute, eight operating profiles emerged, with 11.4 points of efficiency gap between the best and the worst: the first lever is not buying new plants, it is understanding why the same plants perform so differently. - Need signals: «certain times» on the homepage, but the machine plan held in memory overtime and priorities decided aloud on the shop floor promised dates and real dates kept in scattered sheets a buyer asking for traceability and reports, not only the part - Data usually already available: history of machine states or measurements per work session cycle times and stops plan or backlog by machine, if it exists no new sensor project if the registers already exist - Immediate decision: Accept or refuse an urgency. Move a batch. Promise a date that can be kept. Connect the fault alarm to an immediate work order. - Honest limit: It does not calculate the quote's margin: estimated hours and actuals are the two situations above. It recognises a fault in progress, it does not announce it days in advance. Operating profiles must be confirmed on a window longer than one day. - Discarded in the report: the hypothesis «the fault depends on the type of machine» is excluded on an equal base; discarded the single decisive sensor and grouping on efficiency alone. - Method note: numbers measured in the certified reports; they demonstrate the method, they do not predict another company's department. ## When the analysis says not to invest If the data do not answer the question, it is written in the report, with the list of what is missing. The report is paid for; the wrong investment is not. Two real cases and a page that is in every report. ### We were about to invest in a delay-forecast model, but the data did not support it. Logistics and shipments · Real case · negative verdict. In breve. What goes in: planned deliveries and actual outcome from the company's software. What comes out: the verdict: from planning data the delay cannot be forecast. Who decides: whoever was to fund the model. A logistics operator wants to know, at the moment the trip is planned, which deliveries will arrive late. The analysis does not force a weak forecast: it checks whether the signal exists in the data available before departure. **In the real case:** 85,410 deliveries reconstructed from 14 tables in the company's software; 55.4% arrive beyond the planned time. With only the information available at planning, the best system reaches a discrimination capacity of 50.3 out of 100, where 50 is pure chance. The delay is born during the trip, in events that the company's software does not record: stops at loading and unloading, traffic, weather, delays upstream. - Need signals: penalties and complaints for delays that nobody sees coming a supplier proposing the predictive model on planning data a complete, clean archive in the company's software, but frozen at departure - Immediate decision: Stop development of a planning model. Start recording a few trip events. Ask the question again when the new data cover a few months. - Honest limit: It is a limit in the nature of the data, not in quality: refining the algorithm does not create an absent signal. Economic quantities must be recalculated on the data of whoever decides. - Discarded in the report: distance, weight and customer as predictors, because the delay is almost identical on every segment; route and driver history, removed from the model, moves the result by an irrelevant amount. - Method note: numbers measured in the certified report; they demonstrate the method, they do not describe another company's fleet. ### I would have wanted to predict which garments would come back as returns, but the analysis found that product sheets alone could not give this answer. Retail and returns · Real case · negative verdict. In breve. What goes in: product sheets, prices, discounts, recorded returns. What comes out: the verdict: from product sheets alone the return cannot be forecast, with the four data to collect. Who decides: whoever handles returns. An online boutique would like to know which garments will come back. The temptation is to use the data already convenient: category, price, colour, product sheet. The analysis says whether there is a signal in those data, before someone builds a system on top. **In the real case:** 2,176 garments in the catalogue, a little under one garment in seven comes back. No model beats the minimum baseline: the most informative characteristic, the discount applied, explains less than 2% of the phenomenon, the others almost nothing. The most powerful methods were tried and do worse than the simple ones: power does not help when there is nothing to learn. - Need signals: returns treated as a cost suffered reason for the return filled in afterwards, or never a proposal for a predictive system on product sheets alone - Immediate decision: Do not fund the model. Start within three months the collection of four data: fit, customer history, experience after delivery, reason for the return recorded at the right moment. - Honest limit: The «no» does not close the problem: it indicates which data make it addressable. Adding other products with the same characteristics does not add signal. - Discarded in the report: the reason for the return as a variable, because filled in only after the return it is an outcome of the phenomenon, not an early signal. - Method note: numbers measured in the certified report; they demonstrate the method, they do not describe another company's returns. ### The report also contains what was tried and discarded, with the numbers. In every report · Discarded alternatives. In breve. What goes in: the methods and hypotheses tried in the analysis. What comes out: the section of the report with what was discarded and why, with the numbers. Who decides: whoever reads the report. Every certified report has a section dedicated to discarded alternatives: what was tried, what was thrown away and why, with the numbers. It is the opposite of the supplier who brings the only method they know how to do. **In the real case:** in the analysis of shop-floor times, four estimation methods were compared with the minimum baseline, the shop's typical time, on parts never seen before. No refinement of the model beats enriching the recording, because 53% of the variability is not in the fields the shop fills in today. The recommendation was to collect two or three more pieces of information per job, not to buy a more refined method. - Methods: The method kept and the methods discarded, with the measures side by side and the rule by which the simplest was chosen at equal result. - Hypotheses: The hypotheses excluded with a check and those declared unverifiable with the available data, without passing off ignorance as exclusion. - Limits: The limits declared before the recommendations, not in a footnote. - Method note: no algorithm on this page; names and measures sit in the technical report that accompanies every analysis. ## Bring the file that is already prepared by hand every month. Artik Lab says what the data say, what they do not say and whether they are enough, before committing to a project. The first conversation lasts 30 minutes and is free. ## Dashboards look at the past. The data already held can help decide for the future. ## Outputs built for decisions ### Executive Summary Main result, recommended decision, value at stake, limits and actions for the next 30, 90 or 180 days. ### Technical report Data used, controls, methods, metrics, reproducibility and evidence that the model beats a minimum benchmark. ### Action plan Low-risk pilot, responsibilities, timing, measures to observe and criteria to extend, change or stop. ### Data collection plan Which data to collect next, why, with what priority and which decision it would strengthen. ## Forms of value ### Recovered value Customers, orders, lots or bookings that can be saved before value is lost. ### Avoided cost Predictive projects not to fund when current data does not contain the needed signal. ### Organisational efficiency Resources reallocated to time slots, products, checks or processes that truly matter. ### Customer promise More credible deliveries, availability, timing and communication based on better estimates. ### Data governance Less generic data collection, more closely tied to concrete decisions. ## When the signal is not only in data. Many projects start from data, but value may live in documents, emails, procedures, technical software, governance or training. The Atlas helps recognise the right area before shaping the first project. ## First question: which decision must improve? ## Other signals already demonstrated Analyses actually run and certified in other sectors: hospitality, deliveries, energy, restaurants, banking, inspections, manufacturing, local businesses. Every story says which operating data go into the analysis, which signal emerges and which decision can be taken. The sector is that of the case; the method is what transfers. No recognisable customer, no standard promise. ### 1. Hospitality: I find out about cancellations when it is too late A hotel can read the risk already at booking: on more than 119,000 bookings, the system intercepts more than eight cancellations out of ten. The story is simple: leadership sees cancellations only when the damage has already arrived. Agentic data analysis — from data already in the house to a decision — looks instead at the signals available before the stay, such as how far in advance the booking was made, payment method and customer history. For a hotel, a residence or a hospitality group, the booking-engine data become a commercial priority list. The more fragile bookings can be confirmed, called back or handled with different conditions. The sector is different, the signal is the same: an order or a customer that cools before collection, read in time. This case holds for any portfolio of bookings or orders with lead time, payment conditions and history. - Useful signal: The system distinguishes solid bookings from those that deserve a preventive intervention. - Possible decision: Confirm, call back or protect first the most exposed bookings. - Data useful afterwards: Outcome of the call-back, value recovered and customer response. - Limit to declare: The analysis does not eliminate cancellations; it helps choose where to intervene in time. ### 2. Food delivery: I lose the order before it is even delivered When the kitchen does not confirm the order as ready, the risk of losing it rises to 35.7%. At first sight the problem seems to sit in the final delivery: an order does not arrive, the customer complains, the restaurant loses trust. The analysis shows instead that the signal is born earlier, inside the kitchen's operating flow. For a delivery platform or a restaurant chain, this changes the question: not “which rider is late?”, but “which order is leaving the process before it is even delivered?”. - Useful signal: The absence of an intermediate confirmation becomes an operating warning. - Possible decision: Activate at once a reminder, a reassignment or a communication to the customer. - Data useful afterwards: Recorded cause, recovery of the order and cost of the service failure. - Limit to declare: The model works if the intermediate states of the order are recorded well. ### 3. Last mile: I promise a time slot and then customer support explodes The estimate of delivery time goes from an average error of about 41 minutes to about 17 minutes. In many urban logistics companies the problem is not only delivering faster. It is promising a realistic time, so the customer waits less, support receives fewer calls and the fleet is coordinated better. The analysis starts from orders and historical times, but it does not stop at the average. It looks for recurring conditions that make a delivery slower or faster and turns them into a more useful forecast. - Useful signal: A more reliable arrival window for every delivery. - Possible decision: Update communications to the customer, operating priorities and fleet planning. - Data useful afterwards: Complaints, calls avoided and manual interventions by the operating team. - Limit to declare: It does not promise faster deliveries; it promises more credible estimates. ### 4. Energy: Energy is bought by instinct and corrections are paid when the forecast is wrong. The forecast reduces the error by 77% compared with the baseline rule. An energy operator or a large consumer must decide in advance how much energy to buy, cover or reserve. If the forecast is too cautious, it ties up resources; if it is too low, it exposes to costs and corrections. The analysis reads the historical series of hourly consumption and builds an expected profile for the next day. The result is not a chart to archive, but a support for energy planning. - Useful signal: An expected hourly profile more reliable than the rule used as comparison. - Possible decision: Buy, cover or plan capacity with less defensive margin. - Data useful afterwards: Prices, imbalance costs and procurement rules. - Limit to declare: The economic saving must be calculated with the real numbers of the energy contract. ### 5. Restaurants: I order and set shifts as I did last week The forecast of takings improves by 24% compared with the rule “as last week”. A restaurant decides every week how much raw material to order and how many people to put on shift. If it decides by instinct, it risks waste on weak days and insufficient service on strong days. The analysis starts from the history of takings and recognises the real rhythm of the venue. The forecast becomes a practical tool to prepare kitchen, floor and purchasing before demand arrives. - Useful signal: An estimate of future takings more solid than the empirical rule. - Possible decision: Place the forecast beside the choices on purchasing, preparations and shifts. - Data useful afterwards: Real waste, lost sales and staff cost. - Limit to declare: The value is born only if the forecast changes operating decisions. ### 6. Food retail: I see waste when it is already too late The riskiest lots waste almost three times those that are safest. In a supermarket or a food chain, waste does not appear all at once. It is born from small signals: packaging, handling, cold, arrival times, sales priority. The analysis reads these signals when the lot enters the process and creates a risk ranking. The point is not to predict every loss, but to decide which lots to check, rotate or discount first. - Useful signal: A list of lots that deserve attention before visible deterioration. - Possible decision: Concentrate checks, rotations and preventive markdowns on the most exposed lots. - Data useful afterwards: Value saved, reason for the waste and margin after the intervention. - Limit to declare: Not all waste is predictable; the aim is to use preventive actions better. ### 7. Banking: The customer leaves and I find out afterwards The system recognises about three customers at risk out of four. A bank can see a customer leave only when the account is already lost, or it can read earlier the signals that the relationship is cooling. The analysis distinguishes generic risk from the commercial lever on which to act. The useful story is not “this customer will leave”, but “this customer shows inactivity signals and can be reactivated with a targeted action”. It is a decisive difference for building credible campaigns. The sector is different; the type of question, who is about to leave, is the same for any customer portfolio, including among companies. - Useful signal: A contact priority based on behaviour and risk of leaving. - Possible decision: Start targeted reactivation campaigns, not the same communications for everyone. - Data useful afterwards: Behavioural history, contacts made and value retained. - Limit to declare: Recognising the risk today does not always mean forecasting it well in advance. ### 8. Quick-service restaurants: The menu is full but I make money on a few items A few moments of the day and a few menu items generate almost three quarters of revenue. In a quick-service restaurant chain, the problem is not only selling more. It is understanding where revenue is really born: which time slots require staff, which products deserve stock, which items occupy space without paying. The descriptive analysis becomes an operating story: the menu is not all equal and the day does not weigh all the same. This helps decide shifts, stocks and promotions with fewer impressions and more evidence. - Useful signal: A map of the products and the moments that support the profit and loss. - Possible decision: Realign staff, purchasing, promotions and menu review. - Data useful afterwards: Margin per item, preparation times and stock-outs. - Limit to declare: It is not a forecast; it is an operating priority to complete with margin data. ### 9. Industrial maintenance: The machine stops and I find out late With the sensors available, the analysis recognises about 84% of the observed faults. On the factory floor a fault is not only a technical event: it stops people, orders and production capacity. Many machines already have sensors, but the signals stay scattered or are read too late. The analysis builds a warning when the machine's behaviour resembles fault situations already seen. It is useful if it immediately activates a work order, a check or a verification on the floor. It should be read together with the situation on department machines , above. - Useful signal: An operating alarm when the machine shows patterns compatible with a fault. - Possible decision: Connect the warning to maintenance, escalation and verification of the stop avoided. - Data useful afterwards: Intervention time, cost of the stop and spare parts used. - Limit to declare: Recognising a fault in progress is not the same as forecasting it weeks in advance. ### 10. Compliance: I have more checks to do than I can follow With the same number of verifications, the ranking intercepts more serious cases. A control authority or a compliance function always has more cases to verify than it can follow at once. The question is not to run infinite checks, but to decide the right order. The analysis uses the history of verifications to build a priority list. The checks stay human, but the agenda is ordered so as to increase the probability of finding the more serious cases first. - Useful signal: A risk ranking to schedule verifications and follow-up. - Possible decision: Order inspections, audits or internal checks without increasing the budget. - Data useful afterwards: Outcome of the check, recurrence, severity and time to return to compliance. - Limit to declare: The model does not decide sanctions; it helps order the priorities. ### 11. Manufacturing: The same machine performs differently and I do not know why Among the operating profiles a gap of 11.4 efficiency points emerges. In production, average consumption often hides different stories. The same machine can work in more or less efficient ways, but the raw energy figure does not explain at once why. The analysis groups the machine's behaviours and shows which profiles deserve comparison. Before buying new sensors or plants, the company can ask which operating conditions distinguish the profile that performs from the one that wastes. - Useful signal: A map of operating profiles, not only of average consumption. - Possible decision: Compare better and worse profiles and start a waste-reduction pilot. - Data useful afterwards: Energy cost, machine hours, production and operating settings. - Limit to declare: The value in euro should be estimated only when consumption and production are linked. ### 12. Local business: Hundreds of reviews, no decision On 2,961 reviews of twenty businesses in the same category and area, none replied to the reviews received. A local business lives on the choices of whoever reads reviews before walking in, but on the owner's side they remain a star average and a stream of texts that nobody summarises. The reviews of the business and of the competitors in the same place become a list of actions in order of priority, not a dashboard of metrics to interpret: a document that says where to start. The analysis does not look only at the business: it also gathers the competitors in the same area and measures the same dimensions on all of them. It is the comparison that makes a number readable, because a high average says little until it is known how high the average of those around is. In the real case the verdict did not come from the star average, which was already excellent, but from two things nobody had counted: six of the seven critical reviews spoke of the same theme, and none of the twenty businesses in the area replied to whoever wrote. - Useful signal: The critical reviews are not scattered: six out of seven turn around a single theme. That is where to intervene, not on everything. - Possible decision: Reply to reviews, which in the area nobody did, and explain in advance what the price includes. - Data useful afterwards: Outcome of published replies, quote requests after publication, new reviews on the same theme. - Limit to declare: The numbers come from a corpus of public reviews frozen at a date: they photograph that area at that moment, they do not predict the results of another business. ## Frequently asked questions ### Does agentic data analysis replace the control dashboards already in use (Business Intelligence)? No. The control dashboards already in use (Business Intelligence) keep known indicators under watch; agentic analysis diagnoses causes, looks for signals that are not obvious and connects the result to a decision. ### Is perfect data already needed, all in one place? No. The first value can be checking whether the existing data is fit for purpose, what its limits are and which data to collect next. ### What happens if the signal is not there? The method states the negative verdict and indicates which investment to avoid, or which data collection to start before funding a model. ### Are the atlas cases Artik Lab customers? No. They are analyses actually carried out and certified by Artik Lab on operating archives of the same kind as those found in companies, with no recognisable customer. The numbers demonstrate the method; they do not predict another company's results. ### What is needed for the first conversation? The extract that is already prepared by hand every month, and the decision that should improve. In a free 30-minute call it becomes clear whether the data is enough, what is missing and which first project makes sense, before committing. ### Where does the data sit during the analysis? The work is done on an agreed extract, within a written perimeter. # What can be done with AI in a company. A public map of concrete examples to recognise where artificial intelligence can reduce times, errors, risks or decision delays in company processes. ## Each card is an example of work, not a product to buy. Each card describes an example of use: which data or materials go in, which result can be produced, which company value it can generate and which checks stay human. Artik Lab always starts from a first diagnostic conversation and designs specific solutions on the client's context. ## Explore by area, need or process. The applications are examples: they help formulate better questions before choosing training, consulting, data analysis or technical software development. Dataset JSON: https://ar-tik.com/data/ai-applications.en.json Dossier LLM: https://ar-tik.com/en/ai-applications-atlas-dossier.md ## Area - Documents and knowledge: 4. When work depends on PDFs, scans, contracts or procedures. - Operations: 6. When decisions, priorities and manual handoffs slow the process down. - People and HR: 3. When skills, onboarding or feedback remain scattered across functions. - Customer, marketing and sales: 4. When customers, content and sales generate signals nobody is reading. - Technical and software: 4. When rules, code, drawings or technical systems need to become verifiable. - Governance, compliance and risk: 3. When AI use, privacy, risk and responsibilities still lack clear boundaries. - Production, quality and maintenance: 3. When production, quality or maintenance data arrives too late to guide action. - Training and internal memory: 2. When internal knowledge and training material need to remain accessible. - Data science and decisions: 5. When histories, KPIs or signals need validation before anything is built. - Cross-functional tools: 2. When AI is needed to explore, synthesise or prepare cross-functional decisions. ## What can be done with AI in a company. ### Extract data from documents and scans PDFs, images and forms become text, tables and structured fields reusable in company systems. - Operating example: When a process shows a similar need, pdfs and attachments are used to produce structured database and support time reduction, with human review recommended. - Area: Documents and knowledge - What goes in: PDFs and attachments, scans and images, completed forms - What comes out: structured database, operational report - Value: time reduction, fewer errors, traceability - Need signals: scattered documents that are hard to consult, manual copying between emails, spreadsheets and systems - Human review: recommended - Risk: medium ### Check consistency across documents Reports, contracts, specifications and procedures are compared to find discrepancies, divergent versions and inconsistent definitions. - Operating example: When a process shows a similar need, pdfs and attachments are used to produce operational report and support fewer errors, with human review required. - Area: Documents and knowledge - What goes in: PDFs and attachments, internal documentation, contracts and policies, tenders and specifications - What comes out: operational report, risk map - Value: fewer errors, risk reduction, traceability - Need signals: recurring errors in documents, procedures or controls, scattered documents that are hard to consult - Human review: required - Risk: medium ### Make company knowledge searchable by meaning Manuals, procedures and knowledge bases become semantic search with answers grounded in citable sources. - Operating example: When a process shows a similar need, internal documentation are used to produce semantic search and support transferable knowledge, with human review recommended. - Area: Documents and knowledge - What goes in: internal documentation, PDFs and attachments, manuals and training material - What comes out: semantic search, FAQs and answers - Value: transferable knowledge, faster decisions, more consistent service - Need signals: scattered documents that are hard to consult, critical knowledge concentrated in a few people - Human review: recommended - Risk: medium ### Turn meetings, emails and tickets into operating memory Transcripts and threads are cleaned, summarised and converted into traceable decisions, tasks, deadlines and risks. - Operating example: When a process shows a similar need, emails and tickets are used to produce actionable digest and support traceability, with human review recommended. - Area: Operations - What goes in: emails and tickets, transcripts and notes, tickets and requests - What comes out: actionable digest, roadmap and priorities - Value: traceability, faster decisions, transferable knowledge - Need signals: recurring decisions that are slow or based on incomplete information, critical knowledge concentrated in a few people - Human review: recommended - Risk: low ### Generate controlled documents from templates Reports, letters, contracts, FAQs and communications are produced from data and templates, with formal consistency and human review. - Operating example: When a process shows a similar need, structured database are used to produce controlled drafts and support time reduction, with human review required. - Area: Documents and knowledge - What goes in: structured database, internal documentation, contracts and policies - What comes out: controlled drafts, FAQs and answers - Value: time reduction, fewer errors, more governable compliance - Need signals: manual copying between emails, spreadsheets and systems, recurring errors in documents, procedures or controls - Human review: required - Risk: medium ### Map processes and redesign workflows Real work is reconstructed as-is, read for bottlenecks and transformed into a to-be scenario with priorities and controls. - Operating example: When a process shows a similar need, transcripts and notes are used to produce roadmap and priorities and support clearer priorities, with human review recommended. - Area: Operations - What goes in: transcripts and notes, logs and process states, emails and tickets, spreadsheets - What comes out: roadmap and priorities, business case - Value: clearer priorities, faster decisions, avoided costs - Need signals: recurring decisions that are slow or based on incomplete information, manual copying between emails, spreadsheets and systems, AI already used without shared rules - Human review: recommended - Risk: medium ### Triage emails, tickets and requests Incoming communications are classified by urgency, topic, responsibility and required action, with controlled response drafts. - Operating example: When a process shows a similar need, emails and tickets are used to produce actionable digest and support time reduction, with human review recommended. - Area: Operations - What goes in: emails and tickets, tickets and requests, internal documentation - What comes out: actionable digest, controlled drafts, priority ranking - Value: time reduction, more consistent service, clearer priorities - Need signals: manual copying between emails, spreadsheets and systems, recurring decisions that are slow or based on incomplete information - Human review: recommended - Risk: medium ### Plan shifts, resources and priorities Availability, constraints, skills, leave and demand are combined to propose feasible and explainable plans. - Operating example: When a process shows a similar need, spreadsheets are used to produce plan and assignments and support production efficiency, with human review required. - Area: Operations - What goes in: spreadsheets, ERP and business systems, KPIs and time series - What comes out: plan and assignments, dashboards and filtered views - Value: production efficiency, faster decisions, avoided costs - Need signals: planning that is still highly manual, historical data available but not turned into signals - Human review: required - Risk: medium ### Forecast demand and workload Historical orders, revenue, tickets or production become operating forecasts for purchasing, shifts and capacity. - Operating example: When a process shows a similar need, transactions and purchases are used to produce verifiable forecast and support faster decisions, with human review recommended. - Area: Operations - What goes in: transactions and purchases, KPIs and time series, production data - What comes out: verifiable forecast, dashboards and filtered views - Value: faster decisions, avoided costs, production efficiency - Need signals: historical data available but not turned into signals, planning that is still highly manual - Human review: recommended - Risk: medium ### Keep requirements, decisions and stakeholders alive Project meetings and documents feed an evolving dossier with requirements, latent conflicts, decisions and issues. - Operating example: When a process shows a similar need, transcripts and notes are used to produce roadmap and priorities and support traceability, with human review required. - Area: Operations - What goes in: transcripts and notes, requirements and specifications, internal documentation - What comes out: roadmap and priorities, risk map - Value: traceability, fewer errors, transferable knowledge - Need signals: recurring decisions that are slow or based on incomplete information, critical knowledge concentrated in a few people - Human review: required - Risk: medium ### Read customer feedback, reviews and tickets Unstructured texts are aggregated by theme, sentiment, recurring needs and priority actions. - Operating example: When a process shows a similar need, text feedback are used to produce operational report and support more consistent service, with human review recommended. - Area: Customer, marketing and sales - What goes in: text feedback, tickets and requests, public sources - What comes out: operational report, priority ranking - Value: more consistent service, recovered commercial value, clearer priorities - Need signals: abundant feedback that is not analysed, recurring decisions that are slow or based on incomplete information - Human review: recommended - Risk: medium ### Discover market and target needs Public sources and provided material are synthesised into maps of pain points, language, segments, partners and opportunities. - Operating example: When a process shows a similar need, public sources are used to produce operational report and support recovered commercial value, with human review recommended. - Area: Customer, marketing and sales - What goes in: public sources, text feedback, internal documentation - What comes out: operational report, business case - Value: recovered commercial value, clearer priorities, faster decisions - Need signals: abundant feedback that is not analysed, recurring decisions that are slow or based on incomplete information - Human review: recommended - Risk: medium ### Codify brand voice and content Interviews, approved examples and commercial material become operating guidelines and coherent multi-channel drafts. - Operating example: When a process shows a similar need, internal documentation are used to produce policies and guardrails and support time reduction, with human review required. - Area: Customer, marketing and sales - What goes in: internal documentation, text feedback, public sources - What comes out: policies and guardrails, controlled drafts - Value: time reduction, recovered commercial value, traceability - Need signals: recurring errors in documents, procedures or controls, manual copying between emails, spreadsheets and systems - Human review: required - Risk: low ### Support sales, pricing and recommendations Purchase history, catalogs and competitive information help build pitches, bundles, commercial priorities and price scenarios. - Operating example: When a process shows a similar need, transactions and purchases are used to produce operational recommendations and support recovered commercial value, with human review required. - Area: Customer, marketing and sales - What goes in: transactions and purchases, internal documentation, public sources - What comes out: operational recommendations, business case - Value: recovered commercial value, faster decisions, clearer priorities - Need signals: historical data available but not turned into signals, recurring decisions that are slow or based on incomplete information - Human review: required - Risk: medium ### Map skills and capability needs Skills, roles, future goals and trends are connected to define development, upskilling and reskilling priorities. - Operating example: When a process shows a similar need, aggregated hr data are used to produce roadmap and priorities and support transferable knowledge, with human review required. - Area: People and HR - What goes in: aggregated HR data, internal documentation, public sources - What comes out: roadmap and priorities, operational report - Value: transferable knowledge, clearer priorities, faster training - Need signals: critical knowledge concentrated in a few people, AI already used without shared rules - Human review: required - Risk: medium ### Support recruiting and onboarding Job descriptions, applications and onboarding material are structured to prepare evaluations, communications and initial paths. - Operating example: When a process shows a similar need, cvs and applications are used to produce operational report and support time reduction, with human review required. - Area: People and HR - What goes in: CVs and applications, aggregated HR data, manuals and training material - What comes out: operational report, controlled drafts - Value: time reduction, fewer errors, faster training - Need signals: manual copying between emails, spreadsheets and systems, critical knowledge concentrated in a few people - Human review: required - Risk: high ### Simplify recurring HR policies and requests Policies, benefits, procedures and recurring requests become FAQs, drafts and guided paths under HR control. - Operating example: When a process shows a similar need, aggregated hr data are used to produce faqs and answers and support more consistent service, with human review required. - Area: People and HR - What goes in: aggregated HR data, internal documentation, contracts and policies - What comes out: FAQs and answers, controlled drafts - Value: more consistent service, time reduction, more governable compliance - Need signals: manual copying between emails, spreadsheets and systems, scattered documents that are hard to consult - Human review: required - Risk: high ### Define requirements, MVP and acceptance criteria A technical need becomes requirements, user stories, non-functional constraints, estimates and first-release boundaries. - Operating example: When a process shows a similar need, requirements and specifications are used to produce roadmap and priorities and support fewer errors, with human review required. - Area: Technical and software - What goes in: requirements and specifications, transcripts and notes, internal documentation - What comes out: roadmap and priorities, tests and checklists - Value: fewer errors, traceability, avoided costs - Need signals: recurring decisions that are slow or based on incomplete information, recurring errors in documents, procedures or controls - Human review: required - Risk: medium ### Accelerate development, refactoring and tests Existing code and specifications guide controlled code generation, unit tests, refactoring and quality audits. - Operating example: When a process shows a similar need, code and repositories are used to produce tests and checklists and support time reduction, with human review required. - Area: Technical and software - What goes in: code and repositories, requirements and specifications - What comes out: tests and checklists, operational report - Value: time reduction, fewer errors, traceability - Need signals: recurring errors in documents, procedures or controls, manual copying between emails, spreadsheets and systems - Human review: required - Risk: high ### Read specifications and produce technical documentation Tenders, specifications, reports and technical sheets are analysed for critical requirements, risks and documentation drafts. - Operating example: When a process shows a similar need, tenders and specifications are used to produce operational report and support risk reduction, with human review required. - Area: Technical and software - What goes in: tenders and specifications, internal documentation, technical drawings - What comes out: operational report, controlled drafts, risk map - Value: risk reduction, fewer errors, traceability - Need signals: scattered documents that are hard to consult, recurring errors in documents, procedures or controls - Human review: required - Risk: high ### Interpret images, drawings and technical material Photos, drawings and renders become descriptive sheets, component analyses, dimensions and verifiable technical narratives. - Operating example: When a process shows a similar need, operational photos are used to produce operational report and support transferable knowledge, with human review required. - Area: Technical and software - What goes in: operational photos, technical drawings, internal documentation - What comes out: operational report, controlled drafts - Value: transferable knowledge, faster decisions, fewer errors - Need signals: critical knowledge concentrated in a few people, scattered documents that are hard to consult - Human review: required - Risk: medium ### Build AI governance, policies and risk matrix Activities, data and decisions are classified into autonomy, supervision or exclusion zones with clear operating rules. - Operating example: When a process shows a similar need, internal documentation are used to produce policies and guardrails and support risk reduction, with human review required. - Area: Governance, compliance and risk - What goes in: internal documentation, policies and guidelines, transcripts and notes - What comes out: policies and guardrails, risk map, roadmap and priorities - Value: risk reduction, more governable compliance, clearer priorities - Need signals: AI already used without shared rules, recurring decisions that are slow or based on incomplete information - Human review: required - Risk: high ### Prepare compliance, legal and privacy documents Contracts, notices, registers, procedures and letters are prepared as preliminary support to be reviewed by specialists. - Operating example: When a process shows a similar need, contracts and policies are used to produce controlled drafts and support time reduction, with human review required. - Area: Governance, compliance and risk - What goes in: contracts and policies, internal documentation, completed forms - What comes out: controlled drafts, risk map - Value: time reduction, more governable compliance, risk reduction - Need signals: manual copying between emails, spreadsheets and systems, recurring errors in documents, procedures or controls - Human review: required - Risk: high ### Test AI assistants against misuse Chatbots and assistants are stressed with manipulation, data leakage and conflicting instruction scenarios, then hardened with guardrails. - Operating example: When a process shows a similar need, internal documentation are used to produce tests and checklists and support risk reduction, with human review required. - Area: Governance, compliance and risk - What goes in: internal documentation, requirements and specifications, policies and guidelines - What comes out: tests and checklists, policies and guardrails, operational report - Value: risk reduction, more governable compliance, more consistent service - Need signals: AI already used without shared rules, recurring errors in documents, procedures or controls - Human review: required - Risk: high ### Analyse HSE anomalies from operational images Site or department photos are read to identify non-compliance, risks and preventive measures to verify. - Operating example: When a process shows a similar need, operational photos are used to produce operational report and support risk reduction, with human review required. - Area: Production, quality and maintenance - What goes in: operational photos, internal documentation - What comes out: operational report, risk map - Value: risk reduction, faster decisions, more governable compliance - Need signals: recurring errors in documents, procedures or controls, manual copying between emails, spreadsheets and systems - Human review: required - Risk: high ### Optimise production, orders and quality Customer schedules, ERP, cycles, non-conformities and historical costs support priorities, quotes and corrective actions. - Operating example: When a process shows a similar need, erp and business systems are used to produce plan and assignments and support production efficiency, with human review required. - Area: Production, quality and maintenance - What goes in: ERP and business systems, production data, spreadsheets - What comes out: plan and assignments, operational report, operational recommendations - Value: production efficiency, fewer errors, avoided costs - Need signals: planning that is still highly manual, recurring errors in documents, procedures or controls - Human review: required - Risk: medium ### Manage maintenance, assets and spare parts Failure history, sensors and interventions become control priorities, maintenance windows and operating alerts. - Operating example: When a process shows a similar need, sensors and telemetry are used to produce alerts and thresholds and support production efficiency, with human review required. - Area: Production, quality and maintenance - What goes in: sensors and telemetry, production data, logs and process states - What comes out: alerts and thresholds, priority ranking, dashboards and filtered views - Value: production efficiency, avoided costs, risk reduction - Need signals: historical data available but not turned into signals, planning that is still highly manual - Human review: required - Risk: medium ### Create training, quizzes and slides from internal material Manuals, slides and scattered documents become syllabi, quizzes, case studies and role-based learning material. - Operating example: When a process shows a similar need, manuals and training material are used to produce faqs and answers and support faster training, with human review recommended. - Area: Training and internal memory - What goes in: manuals and training material, internal documentation, transcripts and notes - What comes out: FAQs and answers, controlled drafts - Value: faster training, transferable knowledge, more consistent service - Need signals: critical knowledge concentrated in a few people, scattered documents that are hard to consult - Human review: recommended - Risk: low ### Build assistants for company memory Internal documentation feeds Q&A assistants, including voice interfaces, that answer with sources and clear usage boundaries. - Operating example: When a process shows a similar need, internal documentation are used to produce semantic search and support transferable knowledge, with human review required. - Area: Training and internal memory - What goes in: internal documentation, manuals and training material, policies and guidelines - What comes out: semantic search, FAQs and answers, policies and guardrails - Value: transferable knowledge, more consistent service, time reduction - Need signals: critical knowledge concentrated in a few people, scattered documents that are hard to consult - Human review: required - Risk: medium ### Produce executive reports and visual assets Data, KPIs and heterogeneous material become narrative reports, infographics, presentations and coherent visual content. - Operating example: When a process shows a similar need, kpis and time series are used to produce operational report and support faster decisions, with human review recommended. - Area: Cross-functional tools - What goes in: KPIs and time series, spreadsheets, internal documentation - What comes out: operational report, dashboards and filtered views, controlled drafts - Value: faster decisions, traceability, recovered commercial value - Need signals: historical data available but not turned into signals, manual copying between emails, spreadsheets and systems - Human review: recommended - Risk: low ### Detect anomalies and degradation in machinery Time series and industrial sensors are used for alerts, degradation analysis and predictive maintenance with verifiable thresholds. - Operating example: When a process shows a similar need, sensors and telemetry are used to produce alerts and thresholds and support production efficiency, with human review required. - Area: Data science and decisions - What goes in: sensors and telemetry, production data, KPIs and time series - What comes out: alerts and thresholds, verifiable forecast, dashboards and filtered views - Value: production efficiency, avoided costs, risk reduction - Need signals: historical data available but not turned into signals, planning that is still highly manual - Human review: required - Risk: medium ### Segment customers, churn and cross-selling Transactional and behavioural histories become segments, risk rankings, bundles and differentiated commercial actions. - Operating example: When a process shows a similar need, transactions and purchases are used to produce priority ranking and support recovered commercial value, with human review required. - Area: Data science and decisions - What goes in: transactions and purchases, text feedback, KPIs and time series - What comes out: priority ranking, operational recommendations, business case - Value: recovered commercial value, clearer priorities, more consistent service - Need signals: historical data available but not turned into signals, abundant feedback that is not analysed - Human review: required - Risk: medium ### Optimise energy, quality and line performance Telemetry, consumption, quality and machine parameters reveal efficient profiles, waste and operating recommendations. - Operating example: When a process shows a similar need, sensors and telemetry are used to produce dashboards and filtered views and support production efficiency, with human review required. - Area: Data science and decisions - What goes in: sensors and telemetry, production data, KPIs and time series - What comes out: dashboards and filtered views, operational recommendations, business case - Value: production efficiency, avoided costs, faster decisions - Need signals: historical data available but not turned into signals, recurring errors in documents, procedures or controls - Human review: required - Risk: medium ### Analyse territories, profitability and trends Aggregated fiscal, territorial or commercial data become maps, clusters, profitability drivers and decision roadmaps. - Operating example: When a process shows a similar need, transactions and purchases are used to produce dashboards and filtered views and support faster decisions, with human review recommended. - Area: Data science and decisions - What goes in: transactions and purchases, public sources, KPIs and time series - What comes out: dashboards and filtered views, operational report, business case - Value: faster decisions, clearer priorities, recovered commercial value - Need signals: historical data available but not turned into signals, recurring decisions that are slow or based on incomplete information - Human review: recommended - Risk: medium ### Know when not to build a model The first value can be a negative verdict: available data does not yet contain the useful signal and collection must improve. - Operating example: When a process shows a similar need, kpis and time series are used to produce operational report and support avoided costs, with human review recommended. - Area: Data science and decisions - What goes in: KPIs and time series, transactions and purchases, logs and process states - What comes out: operational report, business case, roadmap and priorities - Value: avoided costs, clearer priorities, traceability - Need signals: historical data available but not turned into signals, recurring decisions that are slow or based on incomplete information - Human review: recommended - Risk: low ### Use AI as a discovery lab Cases, material and constraints are explored to generate hypotheses, scenarios, concepts, role simulations and opportunities to verify. - Operating example: When a process shows a similar need, internal documentation are used to produce operational report and support recovered commercial value, with human review recommended. - Area: Cross-functional tools - What goes in: internal documentation, text feedback, public sources - What comes out: operational report, operational recommendations, controlled drafts - Value: recovered commercial value, clearer priorities, faster decisions - Need signals: recurring decisions that are slow or based on incomplete information, abundant feedback that is not analysed - Human review: recommended - Risk: low ## To move from the example to the company's case, a conversation is the starting point. The page is there to orient. The solution is born only after seeing sector, constraints, available data, responsibilities and the decision to improve. 1. **Preliminary picture**: Before the meeting a reading of the public context and of any materials shared is prepared. 2. **Structured conversation**: During the call two or three high-potential processes are identified and constraints, risks and urgencies are clarified. 3. **Targeted proposal**: The result is a calibrated path: training, consulting, data analysis or a technical prototype, with expected results and control criteria. ## FAQ ### Is the Atlas a catalog of ready-made products? No. It is a map of concrete examples. Artik Lab starts from a discovery call and designs the path around the client's real process. ### Are all applications automations? No. Some are training, some analysis, some technical software or governance. AI can assist, suggest, find signals or draft, while sensitive decisions remain governed. ### How are recognisable cases avoided? Cards aggregate patterns and sectors, removing names, clients, natural persons, proprietary data and details that could identify a project. # Corporate AI FAQ: where to start, what to choose, what to avoid. A public question set for owners, leaders and business functions choosing between consulting, courses, data analysis, technical software and the AI applications Atlas. ## The FAQ works as a compass, not as a price list. Each answer helps identify the next useful step. Artik Lab starts from a diagnostic conversation, reads process, data, constraints and responsibilities, then proposes the format that fits the client's real context. Dataset JSON: https://ar-tik.com/data/faq.en.json Dossier LLM: https://ar-tik.com/en/ai-business-faq-dossier.md ## Explore by area or intent. - The questions that block the decision: 7. Cost, data, software, training already done, timing, incentives: short answers before writing. - Where to start: 5. When the company wants AI but has no defined project yet. - First conversation and method: 5. What happens before choosing consulting, a course, analysis or software. - Costs, timing and ROI: 5. How to reason about investment, return, priorities and risk. - Data, documents and privacy: 5. When data is needed, how to prepare it and what controls matter. - AI management consulting: 5. Questions about governance, roadmap, priorities, policies and internal sponsors. - AI training and courses: 5. When to transfer skills to managers, teams and business functions. - Agentic data analysis: 5. When the first value is validating signals in existing data. - Technical software and automation: 5. When a verifiable system is needed, not only existing tools. - AI applications Atlas: 4. How to use examples and patterns without reading them as standard products. - Governance, risks and human review: 4. Responsibility, policies, controls and operating limits for AI. - Internal adoption and teams: 4. How to avoid resistance, informal use and isolated initiatives. - Choosing the right path: 4. Practical differences between training, consulting, data analysis and software development. - AI limits: 4. When to stop, avoid automation or postpone the project. - Before contacting Artik Lab: 4. What to prepare and what to expect from the first exchange. ## The questions that block the decision ### What does it cost? Short answer: The proposal arrives after the opinion or after the first conversation. Operating detail: There is no price list on these pages. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Where are the data during the work? Short answer: In Operational start the files stay in the company's folders and the assistant is in the company's name. Operating detail: In data analysis the work is done on an agreed extract, with a written perimeter. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Is consulting a way to sell software afterwards? Short answer: Consulting is not the door to software. Operating detail: Building, if it is needed, is a later choice. Operational start does not connect the company's software. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### A course on AI has already been done. Is this still needed? Short answer: A course leaves templates and criteria. Operating detail: Here, in one day, a procedure in use remains on the company's files — or an analysis that says whether the data are enough. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### When is something concrete visible? Short answer: Operational start: one day on site. Operating detail: Analysis: after the agreed extract. First conversation: 30 minutes, free of charge. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Can an incentive or a fund be used for this work? Short answer: An eligibility check can be requested. Operating detail: The work must still stand without an incentive: no calls for proposals or amounts are promised. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### How many activities can be done, one after another? Short answer: It is decided with the company, according to need. Operating detail: One job at a time, taken all the way through. Limit to consider: The answers apply to the services as described on these pages; the specific case is clarified in the opinion or in the first conversation. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ## Where to start ### Where should a company start if it has no defined AI project? Short answer: Start from a process, not from a tool. Operating detail: The first task is choosing a recurring decision, visible cost or risk worth reducing. The first conversation clarifies whether the right step is consulting, a course, data analysis or a controlled prototype. Limit to consider: Starting from the model or tool often creates isolated trials with no measurable return. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### How should the first process to improve with AI be chosen? Short answer: Choose a frequent, observable process linked to a cost or delay. Operating detail: Good candidates include repeated emails, documents to read, priorities to assign or decisions arriving late. If the process is not observable, it should first be made clearer. Limit to consider: Starting from the model or tool often creates isolated trials with no measurable return. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Can a company without an internal IT team start? Short answer: Yes, if it starts from decisions, processes and skills before technology. Operating detail: Many initial activities do not require software development: process mapping, risk criteria, focused training and first-case selection matter first. Technical work arrives only when the scope is clear. Limit to consider: Starting from the model or tool often creates isolated trials with no measurable return. Next step: Choose a course or lab if the main need is transferring method to the team. ### Is it better to start from ChatGPT, software or a problem? Short answer: It is better to start from the business problem and choose the tool later. Operating detail: A tool can help, but it does not decide goal, data, responsibility and success criteria. Artik Lab uses the first diagnosis to avoid isolated trials and connect AI to an operating result. Limit to consider: Starting from the model or tool often creates isolated trials with no measurable return. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### How can a company know whether it is ready to use AI? Short answer: Readiness depends on process, sponsor, minimum data and clear responsibility. Operating detail: The company does not need to be mature everywhere. It does need one concrete problem, people able to validate the result and a decision to improve. Otherwise training or mapping should come first. Limit to consider: Starting from the model or tool often creates isolated trials with no measurable return. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## First conversation and method ### What should be prepared for the first conversation? Short answer: Prepare a process, a material example and one decision to improve. Operating detail: Perfect documents are not required. Context, constraints, roles involved, available data and a description of what now takes too long or creates risk are enough. Limit to consider: A generic diagnosis is not enough to choose investment, responsibility and data. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### How long does the first conversation take? Short answer: Thirty minutes, free of charge, to understand the initial scope. Operating detail: The goal is not solving everything in the meeting, but separating need, constraints and next step. A course, consulting path, data analysis or prototype may follow. Limit to consider: A generic diagnosis is not enough to choose investment, responsibility and data. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### What comes out of the initial diagnosis? Short answer: It indicates the most sensible format and the risks to govern. Operating detail: The diagnosis may point to training, opportunity mapping, data validation, technical prototype or a temporary stop. Its value is avoiding wrong investment before committing time and budget. Limit to consider: A generic diagnosis is not enough to choose investment, responsibility and data. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Who should join the first conversation? Short answer: At least someone who knows the process and someone who can decide priorities. Operating detail: Leadership, the involved function and an operating reference avoid partial readings. If data or systems are involved, IT or tool owners can also be useful. Limit to consider: A generic diagnosis is not enough to choose investment, responsibility and data. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### What happens after the first conversation? Short answer: The next choice is whether to deepen, train, analyse data, build a prototype or stop. Operating detail: The conversation does not force a project. It turns a vague question into a practical choice with clearer scope, priorities, risks and control criteria. Limit to consider: A generic diagnosis is not enough to choose investment, responsibility and data. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ## Costs, timing and ROI ### How much does an AI project cost? Short answer: Cost depends on scope, data, risk, people involved and expected result. Operating detail: Before estimating, it must be clear whether the work is training, diagnosis, data analysis, prototype or system. A small well-bounded path is often more useful than a broad unmeasurable project. Limit to consider: ROI should not be promised before knowing process, baseline, data and possible actions. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### How should ROI be evaluated for an AI project? Short answer: Compare current cost, possible improvement and actions that can truly be taken. Operating detail: Before the model, baseline, KPIs and responsibility are needed. If AI produces a signal but nobody can act, value stays theoretical; if it changes a frequent decision, return can be estimated. Limit to consider: ROI should not be promised before knowing process, baseline, data and possible actions. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### How long does it take to see a first result? Short answer: A first result can arrive in a few weeks if the scope is small and verifiable. Operating detail: The initial result may be a map, policy, adapted course, data test or minimal prototype. It is not always production; often it is a better decision on what to fund or avoid. Limit to consider: ROI should not be promised before knowing process, baseline, data and possible actions. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Can the company start with a small project? Short answer: Yes, it is usually better to start with a narrow measurable scope. Operating detail: A small case validates data, responsibility and value without excessive expectations. If it works, it expands; if it does not, learning happens before too much spending. Limit to consider: ROI should not be promised before knowing process, baseline, data and possible actions. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### How can spending on the wrong AI project be reduced? Short answer: Define a stop criterion before the full investment. Operating detail: Each case should have hypotheses, KPIs, minimum data, responsibility and stop conditions. A negative verdict on data or process can be a good result when it avoids larger costs. Limit to consider: ROI should not be promised before knowing process, baseline, data and possible actions. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## Data, documents and privacy ### Does AI require already clean data? Short answer: No, but the company must know which data exists, who understands it and its limits. Operating detail: Perfectly clean data rarely exists at the start. The first task can be assessing quality, coverage, errors and usefulness against the decision to improve. Limit to consider: Personal, regulated or confidential data require minimisation, access control and competent review. Next step: Consider agentic data analysis when the signal must be validated before building. ### Can AI work on documents, emails and procedures? Short answer: Yes, many cases start from text material already inside the company. Operating detail: Contracts, manuals, tickets, emails and procedures can become search, summaries, checks or drafts. Clear sources, permissions, human review and boundaries on what AI may do are required. Limit to consider: Personal, regulated or confidential data require minimisation, access control and competent review. Next step: Use the Atlas to recognise similar patterns before shaping the project. ### How can privacy risks with AI be avoided? Short answer: Limit data, access, tools and allowed uses before experimenting. Operating detail: Proper management starts with data classification, minimisation, lawful basis, authorised accounts and specialist review when required. Policy must become practical behaviour. Limit to consider: Personal, regulated or confidential data require minimisation, access control and competent review. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Can confidential data or technical know-how be used? Short answer: Yes, only with explicitly agreed boundaries, access and materials. Operating detail: Code, drawings, specifications, industrial data and expert knowledge are treated as intellectual property. Public examples use only anonymised descriptions that cannot identify the client. Limit to consider: Personal, regulated or confidential data require minimisation, access control and competent review. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ### What happens if the data is not enough? Short answer: The useful result may be knowing which data is missing and which investments to avoid. Operating detail: A model is not always built. Sometimes the best work is defining new data collection, changing the process or postponing automation until the signal becomes verifiable. Limit to consider: Personal, regulated or confidential data require minimisation, access control and competent review. Next step: Consider agentic data analysis when the signal must be validated before building. ## AI management consulting ### When is AI management consulting needed? Short answer: It is needed when priorities, governance, risk criteria or roadmap are missing. Operating detail: Consulting helps leadership decide where to use AI, where to stop, which skills to build and which first pilot may have measurable value. Limit to consider: Without internal sponsor and real decisions, consulting remains an unused map. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### What remains after AI management consulting? Short answer: Criteria, opportunity map, policy, roadmap and first-pilot brief remain. Operating detail: The goal is not an inspirational presentation. The artifacts should help leadership and functions decide, communicate rules, assign responsibility and move to the next case with control. Limit to consider: Without internal sponsor and real decisions, consulting remains an unused map. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### How should people already using AI informally be managed? Short answer: Turn informal use into governed practice, not only prohibition. Operating detail: Shadow AI signals a real efficiency need. The company should distinguish allowed uses, excluded data, output checks and safe company paths to avoid losing useful energy. Limit to consider: Without internal sponsor and real decisions, consulting remains an unused map. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Is an internal sponsor needed to start? Short answer: Yes, at least one person must decide priorities and validate results. Operating detail: The sponsor does not need to be technical. They must understand process value, involve the right people and authorise choices about data, timing and responsibility. Limit to consider: Without internal sponsor and real decisions, consulting remains an unused map. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### When does a company need an AI policy? Short answer: It is needed when use grows and data, tools or responsibilities are no longer clear. Operating detail: A useful policy is not abstract: it defines allowed cases, forbidden data, human review, accounts, escalation and criteria for moving from personal use to company use. Limit to consider: Without internal sponsor and real decisions, consulting remains an unused map. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## AI training and courses ### When does an AI course make sense? Short answer: It makes sense when the main issue is transferring method and criteria to the team. Operating detail: A course fits when people already use AI tools differently, common rules are missing or practical examples must be brought into company roles and processes. Limit to consider: Generic training does not change work if examples, roles and rules remain distant from context. Next step: Choose a course or lab if the main need is transferring method to the team. ### Are courses standard or adapted to the company context? Short answer: The structure is stable, but examples, exercises and priorities are adapted. Operating detail: The DTR method recalibrates the path around processes, materials and participant questions. This avoids abstract lessons and makes it easier to turn training into operating practice. Limit to consider: Generic training does not change work if examples, roles and rules remain distant from context. Next step: Choose a course or lab if the main need is transferring method to the team. ### Is programming required to join the courses? Short answer: No for managerial, introductory and operational paths. Operating detail: Programming is required only in technical courses. For leaders and business functions, the focus is on processes, prompts, review, risks, data and responsible-use criteria. Limit to consider: Generic training does not change work if examples, roles and rules remain distant from context. Next step: Choose a course or lab if the main need is transferring method to the team. ### What remains after a corporate AI course? Short answer: Materials, usage criteria, adapted examples and a view of promising processes remain. Operating detail: The course should not end with theory only. It should leave practical tools: checklists, exercises, review rules, reusable examples and questions for choosing next cases. Limit to consider: Generic training does not change work if examples, roles and rules remain distant from context. Next step: Choose a course or lab if the main need is transferring method to the team. ### Who should be trained first? Short answer: Usually sponsors, function leads and people already using AI should come first. Operating detail: The priority is not training everyone immediately. It is creating a core group able to recognise useful cases, check outputs, explain limits and transfer practices. Limit to consider: Generic training does not change work if examples, roles and rules remain distant from context. Next step: Choose a course or lab if the main need is transferring method to the team. ## Agentic data analysis ### When is agentic data analysis the right first step? Short answer: It is right when a decision depends on signals hidden in data. Operating detail: If the company has histories, orders, tickets, sensors or KPIs but does not know which priorities emerge, analysis validates signal, limits and possible actions before building. Limit to consider: If the data contains no signal, forcing a model creates cost and false confidence. Next step: Consider agentic data analysis when the signal must be validated before building. ### Does agentic data analysis replace Business Intelligence? Short answer: No, it complements BI when indicators must become decisions. Operating detail: BI monitors known metrics and past trends. Agentic analysis looks for signals, anomalies, priorities or stop criteria connected to a concrete action. Limit to consider: If the data contains no signal, forcing a model creates cost and false confidence. Next step: Consider agentic data analysis when the signal must be validated before building. ### What is the value of a negative data result? Short answer: It is valuable because it avoids funding a weak model. Operating detail: Knowing that the signal is not present yet allows the company to change data collection, review the process or move budget to more mature cases. It is a useful management decision. Limit to consider: If the data contains no signal, forcing a model creates cost and false confidence. Next step: Consider agentic data analysis when the signal must be validated before building. ### Which KPI is needed before analysing data? Short answer: A KPI linked to a decision or action is needed, not just to a chart. Operating detail: Useful examples: order to chase, batch to check, customer to contact, shift to rebalance. The KPI should show whether analysis truly changes work. Limit to consider: If the data contains no signal, forcing a model creates cost and false confidence. Next step: Consider agentic data analysis when the signal must be validated before building. ### Do data-based decisions remain human? Short answer: Yes, especially when they affect customers, people, quality, safety or risk. Operating detail: Analysis can rank priorities, suggest signals and explain limits. The decision remains under company responsibility, with human review and criteria agreed before operating use. Limit to consider: If the data contains no signal, forcing a model creates cost and false confidence. Next step: Consider agentic data analysis when the signal must be validated before building. ## Technical software and automation ### When does it make sense to build technical AI software? Short answer: It makes sense when a verifiable system is needed and standard tools are not enough. Operating detail: If the process includes calculations, expert rules, legacy data, integrations or critical checks, custom development may be needed. Requirements, tests and responsibilities must come first. Limit to consider: Automating a poorly understood process only makes errors and ambiguity faster. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ### What is the difference between simple automation and technical software? Short answer: Automation connects steps; technical software embeds rules, tests and maintenance. Operating detail: If moving data between tools is enough, automation can be light. If calculations, checks, versions, audit and responsibility matter, a more robust system is needed. Limit to consider: Automating a poorly understood process only makes errors and ambiguity faster. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ### Can legacy software be modernised with AI? Short answer: Yes, but existing logic, data, constraints and risks must be understood first. Operating detail: AI can help read code, documentation or data, but modernisation requires audit, result comparison, regression tests and progressive migration. Limit to consider: Automating a poorly understood process only makes errors and ambiguity faster. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ### What is the difference between a controlled prototype and production system? Short answer: A prototype validates feasibility; production requires tests, security, maintenance and responsibility. Operating detail: A prototype can be small and isolated. A production system must handle real users, errors, data, permissions, logging, documentation and acceptance criteria. Limit to consider: Automating a poorly understood process only makes errors and ambiguity faster. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ### Does AI software need to integrate with company systems? Short answer: Only when value requires operational continuity, updated data or repeated use. Operating detail: Not every prototype must integrate immediately. Integration becomes necessary when the system enters daily work and must respect permissions, data, traceability and maintenance. Limit to consider: Automating a poorly understood process only makes errors and ambiguity faster. Next step: Move to software development only when a verifiable system, tests and maintenance are needed. ## AI applications Atlas ### Is the Atlas a catalog of ready-made products? Short answer: No, it is a map of patterns for recognising process opportunities. Operating detail: Each card helps formulate better questions about data, results, value and controls. The real solution is designed only after context, constraints and company priorities are reviewed. Limit to consider: A public pattern should not be treated as a standard promise or ready solution. Next step: Use the Atlas to recognise similar patterns before shaping the project. ### How should the Atlas be used to test whether a case makes sense? Short answer: Find a similar pattern and compare data, result and human review. Operating detail: If a card resembles the company process, the next step is checking available material, decision to improve, risk and suitable format: course, consulting, analysis or software. Limit to consider: A public pattern should not be treated as a standard promise or ready solution. Next step: Use the Atlas to recognise similar patterns before shaping the project. ### Are Atlas examples recognisable client cases? Short answer: No, they are anonymised and generalised patterns. Operating detail: Names, natural persons, internal projects, recognisable products or identifying detail combinations are not published. The goal is recognising opportunities, not exposing confidential cases. Limit to consider: A public pattern should not be treated as a standard promise or ready solution. Next step: Use the Atlas to recognise similar patterns before shaping the project. ### After finding an Atlas card, which service should be chosen? Short answer: It depends on the main constraint: decision, skill, data or system. Operating detail: If management choice is missing, consulting fits; if skills are missing, training fits; if the doubt is in data, analysis fits; if an operating engine is needed, technical software fits. Limit to consider: A public pattern should not be treated as a standard promise or ready solution. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ## Governance, risks and human review ### When is human review needed on AI outputs? Short answer: It is needed whenever output affects decisions, customers, sensitive data or responsibility. Operating detail: Review is not a formality. It should define who checks, with which criteria, when to correct, when to reject output and when AI should not be used. Limit to consider: Without human review, privacy and responsibility boundaries, AI use remains fragile. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Who is responsible for an AI-assisted decision? Short answer: Responsibility remains with the organisation and appointed people. Operating detail: AI may suggest, rank priorities or draft, but it should not become a responsibility gap. Roles, escalation, traceability and acceptance criteria are needed. Limit to consider: Without human review, privacy and responsibility boundaries, AI use remains fragile. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Are there activities AI should not do? Short answer: Yes, some decisions must remain human or require strong supervision. Operating detail: Legal, HR, safety, health, credit, critical quality or sensitive-data decisions need careful classification. In some cases AI may prepare material, not decide. Limit to consider: Without human review, privacy and responsibility boundaries, AI use remains fragile. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### How can AI output quality be controlled? Short answer: Use explicit criteria, approved examples and cases where output must be rejected. Operating detail: Quality should not be judged by impression. Source, tone, completeness, critical errors, acceptance threshold and human review should be defined, especially for documents and external communication. Limit to consider: Without human review, privacy and responsibility boundaries, AI use remains fragile. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## Internal adoption and teams ### How can team resistance to AI be managed? Short answer: Clarify purpose, limits and practical benefit. Operating detail: People collaborate better when they understand what changes, what remains human and which activities become lighter. Training and cases close to real work reduce fear and confusion. Limit to consider: Adoption fails when people do not understand purpose, limits and usage rules. Next step: Choose a course or lab if the main need is transferring method to the team. ### Are internal AI champions needed? Short answer: They help when use must move from individual experimentation to shared practice. Operating detail: AI champions collect cases, spread rules, flag risks and maintain continuity after training or consulting. They need a clear mandate and dedicated time. Limit to consider: Adoption fails when people do not understand purpose, limits and usage rules. Next step: Choose a course or lab if the main need is transferring method to the team. ### How should internal AI adoption be measured? Short answer: Measure changed processes, checked outputs and improved decisions, not only access. Operating detail: Counting licences or prompts is not enough. Better indicators include saved time, fewer errors, governed cases, trained people, applied policies and faster or more reliable decisions. Limit to consider: Adoption fails when people do not understand purpose, limits and usage rules. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### How can a course avoid remaining isolated? Short answer: Connect it to real cases, sponsors, policy and next actions. Operating detail: After training, candidate processes should be collected, two or three controlled experiments chosen and responsibilities assigned. This turns the course into adoption, not a separate event. Limit to consider: Adoption fails when people do not understand purpose, limits and usage rules. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## Choosing the right path ### When is consulting needed and when is a course enough? Short answer: Consulting is needed for strategy decisions; a course is enough for method transfer. Operating detail: If the issue is choosing priorities, governance and roadmap, consulting fits. If scope is clear and the need is helping people work better, a course may be right. Limit to consider: Choosing the wrong format increases cost, frustration and unmanaged expectations. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### When should data analysis be done and when software developed? Short answer: Analysis validates the signal; software builds a usable, maintainable system. Operating detail: If it is unclear whether data contains value, start with analysis. If value is clear and needs operation through tests, interfaces and integrations, move to software. Limit to consider: Choosing the wrong format increases cost, frustration and unmanaged expectations. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### When should the FAQ be used and when the Atlas? Short answer: The FAQ answers path questions; the Atlas shows application examples. Operating detail: If the question is which path to choose, the FAQ helps. If the question is where AI could help in a process, the Atlas offers patterns to compare. Limit to consider: Choosing the wrong format increases cost, frustration and unmanaged expectations. Next step: Use the Atlas to recognise similar patterns before shaping the project. ### What if nobody truly owns the process? Short answer: Before automation, ownership and decision criteria must be assigned. Operating detail: A process without an owner creates ambiguity even with AI. Consulting or a redesign lab helps clarify roles, steps, data and priorities. Limit to consider: Choosing the wrong format increases cost, frustration and unmanaged expectations. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## AI limits ### When is AI not worth using? Short answer: When data, responsibility, possible action or error tolerance are missing. Operating detail: If error is unacceptable, the process is too ambiguous or nobody can verify the result, it is better to stop, redesign or use simpler tools. Limit to consider: AI does not replace judgement, professional responsibility or data that does not exist. Next step: Stop or postpone the case if sponsor, minimum data, responsibility or possible action are missing. ### How should AI errors and hallucinations be managed? Short answer: Plan for them with sources, checks, approved examples and human review. Operating detail: AI can produce plausible but wrong answers. Usage limits, citable sources, real-case tests and rules against using unchecked outputs are required. Limit to consider: AI does not replace judgement, professional responsibility or data that does not exist. Next step: Open a consulting path to clarify priorities, governance and roadmap. ### Can AI fully automate a process? Short answer: Only rarely: most cases need supervision or human intervention. Operating detail: Full automation is risky when data, exceptions, responsibility and quality are not stable. Often the best value is a controlled copilot, not a process without people. Limit to consider: AI does not replace judgement, professional responsibility or data that does not exist. Next step: Stop or postpone the case if sponsor, minimum data, responsibility or possible action are missing. ### Do these FAQs replace legal, tax or specialist advice? Short answer: No, they provide business orientation, not regulated specialist advice. Operating detail: When a case touches legal, tax, medical, financial or safety obligations, qualified professionals should review it. AI may prepare material, not replace specialist responsibility. Limit to consider: AI does not replace judgement, professional responsibility or data that does not exist. Next step: Open a consulting path to clarify priorities, governance and roadmap. ## Before contacting Artik Lab ### How can Artik Lab be contacted about a case? Short answer: Write to dtr@ar-tik.com with process, goal and main constraints. Operating detail: The message can be short: business area, problem, available material, people involved and urgency. The first reply clarifies whether a diagnostic conversation makes sense. Limit to consider: A first exchange without context produces generic, less useful answers. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Is a project document already required? Short answer: No, an honest description of the problem and context is enough. Operating detail: A structured document helps but is not essential. It is more important to clarify which process creates cost, delay or risk and who can validate a possible result. Limit to consider: A first exchange without context produces generic, less useful answers. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### Can the first conversation be in multiple languages? Short answer: Yes, the site and public material cover Italian, English, Spanish, French and Brazilian Portuguese. Operating detail: The operating language is agreed according to the people involved. Consistency across versions helps international teams read the same positioning without market-specific promises. Limit to consider: A first exchange without context produces generic, less useful answers. Next step: Bring the case to the first conversation with process, goal, available data and constraints. ### What if the company is not ready to contact Artik Lab? Short answer: It can start from the Atlas, this FAQ and the course catalog. Operating detail: If the need is still unclear, collect internal examples, note recurring questions and identify one process with visible cost. This makes the later conversation more concrete. Limit to consider: A first exchange without context produces generic, less useful answers. Next step: Use the Atlas to recognise similar patterns before shaping the project. ## Prepare the first conversation To start, gather one process to improve, an example of available material or data, the decision to make more reliable and the constraints to respect. # Technical software, calculation programmes and advanced data analysis. Artik Lab develops advanced software for clients when the problem is not solved by a dashboard or by standard company software: technical calculations, mathematical models, operating data, expert rules and workflows that can be checked with tests. ## When company know-how is too important to remain in spreadsheets, legacy code or the heads of a few experts. Many industrial companies run on calculations, exceptions and technical decisions that have grown over time. Sometimes they live in fragile spreadsheets, sometimes in obsolete software, sometimes in procedures known only by long-time users. The service turns that knowledge into readable, testable and transferable systems. ## Systems that make repeatable what now depends on experience, files and manual checks. Value comes from combining software engineering, data analysis and expert knowledge formalisation. The outcome is not a demo prototype, but a system with acceptance criteria, tests, documentation and clear boundaries. ### Calculation and verification programmes Calculation programmes that, on the same data, always give the same result: checks, scenarios, simulations and repeatable verifications. ### Data systems and advanced analysis Collection, normalisation and reading of operating data to detect anomalies, patterns, priorities and risks. ### Updating technical software that nobody wants to touch any more Audit of the existing code, reconstruction of the logic, parsers for historical formats and progressive rewrite. ### Interfaces, reports and APIs Tools usable by technical offices and operating functions: decision dashboards, reports, exports and integrations. ## From technical process to verifiable system. 1. **Technical audit**: Read the existing system: data, formulas, flows, dependencies, known errors and operational risk. 2. **Domain formalisation**: Expert rules become entities, constraints, assumptions, edge cases and decision criteria. 3. **Verifiable architecture**: The calculation core is separated from interfaces, reports and AI components, so it remains controllable. 4. **Computable prototype**: Build a small complete flow: source data, data model, calculation, verification and usable result. 5. **Validation**: Automated tests, synthetic cases, regression and comparison with known references measure differences and risks. 6. **Production**: The system becomes usable through interfaces, APIs, reports, documentation and maintenance responsibilities. ## What remains inside the company. - Technical blueprint with architecture, risks, data, assumptions and open decisions. - Structured knowledge base with operating rules, constraints, sources and confidence levels. - Calculation engine, data system or technical application with automated tests. - Verification dossier with discrepancies, tolerances, acceptance criteria and remediation priorities. - Reports, interfaces or APIs to integrate the system into real work. - Roadmap in progressive work packages, with testable outputs and technical checkpoints. ## Typical problems the service can address. ### Technical documentation out of control A technical office uses complex files to take recurring decisions. The risk is that the formulas are no longer checkable and that every change requires historical memory. The project reconstructs the rules, turns them into a data model and adds tests to avoid regressions. ### Technical software that still works, but nobody can change it with safety any more A critical application still works, but it depends on dated technologies and undocumented logic. The work starts from the audit, separates what must be preserved from what must be redesigned and builds a progressive rewrite with comparison on the results. ### Industrial data that do not help to take decisions The process produces data, but the company uses them mainly for retrospective reports. The analysis looks for useful signals for operating priorities, anomalies, forecasts and control choices, also stating when the data are not enough. ### Company knowledge that is never documented Some decisions depend on the experience of key roles. The project makes rules, exceptions and attention thresholds explicit, so the knowledge stays available even when people, tools or work volumes change. ## Five real cases What was obtained. ### 1. Technical office: The calculation rules live in the code and in one person, not in a specification document. Eight weeks of analysis on about 267 source files: 37 decision rules written down and 18 issues the company did not know it had. The situation. A company designs made-to-order components whose sizing requires regulatory checks. For more than twenty years the work has gone through an in-house calculation programme, on a development platform that is no longer supported, with proprietary file formats that were never documented. The programme works, but nobody can change it with reasonable safety. And the specifications do not exist: the rules live inside the code and in the experience of a single designer. What was done. The requirements were not collected; they were reconstructed. The work reads three sources in parallel — the source code, the archive of projects actually delivered, and the technical office's indications — and collides them with one another: every rule extracted from the code is checked against the data and taken back to the expert, every indication from the expert is checked against the code. No statement is accepted on trust, neither the expert's nor the code's; what cannot be verified is declared as such instead of remaining implicit. What changed. The company stopped depending on knowledge that existed in only one place. What had been tacit became a document that can be discussed, and together with the map of the system came the list of inconsistencies that nobody could still see, including divergences between what the manual prescribes and what the programme actually runs. - Measured on the project: About 267 source files and twelve libraries mapped in eight weeks of analysis. 37 decision rules written down: 13 hard constraints, 19 optimisation heuristics (rules of thumb), 5 diagnostic rules. - What came out: 18 issues in the software in use, classified by severity, each with a proposed treatment. Added to these are twelve documented limitations and ten requirements for the new system. - How it continues: A path in testable packages, in which every stage has a numerical acceptance criterion agreed before work starts. - Declared limit: The numbers come from this project: they say what the method produced there, not what it will produce elsewhere. ### 2. Application processing and back office: Requests arrive by email. The company's software does not follow them. Two official documents already filled with checked data; no communication goes out without a person's approval. The situation. An organisation receives requests by email, in free form, written by different people. Each case requires precise data, some of it verifiable only by crossing several pieces of information, and produces official documents on fixed templates. Qualified time is spent on carrying data: rereading, asking for the missing item, waiting, recopying the same fields into several documents, chasing whoever must complete their part. The bottleneck is not the decision: it is the carrying. What was done. An assistant watches the inbox. It reads the message, extracts the data, checks them against the organisation's conditions and replies in plain language explaining which item is missing and in what form it is needed, instead of sending back an empty form. Above it works a manager that knows the life cycle of the case: how many times the same item has already been asked for, whether whoever must complete their part has replied within the stated days, whether the applicant already has an open case. After a declared number of fruitless exchanges the case is frozen, instead of feeding an endless exchange. What changed. When the case is complete, the official documents come out already filled on the organisation's templates, with the fields populated from validated data. Qualified time returns to the judgement of merit, which is the only part that really required an experienced person. - Measured on the project: 135 automated tests passing on data models, the validation engine, document generation and case handling, including six complete paths from email to document. - Project choice: Zero automatic sends: every outbound communication goes through an operator's approval. In a process that produces official acts, automation stops one step before the signature. - Declared limit: Time saved has not yet been measured before and after on a full cycle. The project states what the system does, not by how much it shortens the case. ### 3. Field crews: The weekly plan for crews in the field is still made by hand. The weekly plan is calculated in a few seconds and is regenerated when a constraint changes. How much time it saves has not yet been measured, and is not declared. The situation. When a company sends crews to customers, the weekly programme is born from dozens of constraints that get in one another's way: addresses scattered over a wide territory, time windows imposed by customers, commercial priorities and legal deadlines, durations that change with the type of job, crews that are not interchangeable. Made by hand the plan always comes out feasible but never efficient, and there is no benchmark to notice the difference. What was done. The problem is formulated as a routing problem with time windows and solved with an optimisation solver. A first level distributes jobs across the days, balancing crew capacity and placing the heaviest jobs first. A second solves each day as a multi-vehicle route, with real travel times taken from a road service and time windows treated as constraints that cannot be broken. What changed. The manager receives a plan ready to use: routes on a map, a daily calendar per unit, kilometres, driving hours and crew saturation. And the same plan is regenerated in a few seconds when a constraint changes, which is the part needed at once, when someone drops out. - How it was tried: On a demonstration archive with fictional names — 35 sites, 128 people, two mobile units — the full weekly plan comes out in a few seconds, against the hours that the work by hand requires. - What it returns: On every run: total kilometres, driving hours and saturation percentage per crew. These are the numbers that make it possible to compare two plans, instead of trusting the first one. - Declared limit: There is not yet a measured before-and-after comparison on a real customer. Until there is, no saving percentage is declared: that would be an estimate presented as a measurement. ### 4. Public-facing services: Recognising catalogue items from a photo, without a campaign of photographs classified by hand. Opening a new site means loading a file, not collecting and classifying photographs of every item by hand. The situation. Recognising catalogue items from a photo is expensive if every site must first collect and classify photographs by hand, and the catalogue changes every day. That is what makes activation times prohibitive for traditional solutions in this kind of service. What was done. The alternative approach is to describe the catalogue instead of showing it. The day's items are loaded in declarative form — name, category, extended description, ingredients, quantities — and a general-purpose multimodal model recognises from those descriptions, not from an archive of labelled images. The system also handles composed items, with portion coefficients, so that the values stay correct when half-portions of different products sit in the same choice. What changed. Activating a new site means loading its catalogue. Dedicated hardware is not needed: the person's smartphone is used, with no kiosk and no readers. And the cost scales with traffic, instead of being the fixed cost of a presence that has to be kept even when ten people pass. - How it is checked: A regression suite on real images, with the reference truth declared for every image: a change to the model or to the instructions is measured on a stable bench, instead of by impression. - Specified, not yet finished: Seven languages across the whole interface and the catalogue content. Full localisation is planned, not yet completed. - Declared limit: The accuracy figure is not yet declarable: part of the recent images has no reference truth and is excluded from the tests. It is the first number anyone evaluating this solution will ask for, and it is not estimated in place of measuring it. ### 5. Safety and training: Mandatory training that expires without anyone noticing. 18 types of mandatory course, each with its own expiry; for every person four states (valid, due soon, expired, never taken) and a notice period that can be adjusted. The situation. Whoever is accountable for the mandatory training of tens or hundreds of people keeps the status on spreadsheets that age with every hire and every change of role. The risk is not theoretical: it is noticing an overdue expiry during an inspection, with what that implies for occupational safety. What was done. The problem is not calculating a date. It is holding together people records, course history and periodicity rules that differ by role and by activity, and calculating for every person and every obligation one of four states: valid, due within the threshold, expired, never taken. From there come the training matrix readable at a glance, the reminders in a formal register and the periodic report. What changed. When the matrix exists, the work becomes scheduling the courses instead of reconstructing the status. Personal data stay on the organisation's server: no sending to external services, obtained as an architecture constraint and not as a statement of intent. - Measured on the project: 18 types of mandatory course modelled, with periodicity from one to five years and some one-off. Four states calculated for every combination of person and obligation, with default notice at 90 days and configurable per customer. - How it was tried: On a demonstration archive with fictional names: eight companies, about 175 people and more than 700 training records, with a realistic distribution of states. - Declared limit: The calculation follows fixed rules and, on the same data, always gives the same result, and that is as it should be: in a periodicity rule there is nothing to hand to a model. AI is useful one step before, to bring in data that today arrive disordered, and one step after, to notice that a rule has changed. ## Before choosing the format, recognise the process. The Atlas gathers concrete AI application examples across documents, operations, HR, marketing, software, governance, production, training and data. It helps decide whether the need requires consulting, data analysis, technical development or training. Atlas page: https://ar-tik.com/en/ai-applications-atlas.md ## AI can help, but the technical core must remain explainable. In technical systems, opaque components should not replace verifiable calculation. AI can help explore data, explain results, propose scenarios, read documents or assist the user. The deterministic core, domain rules and tests remain the control point. ## Where the boundary sits: the calculation stays repeatable, AI works on top of it. Five different projects, the same choice: what decides is a calculation that can be repeated and checked with tests; AI works on top of it. Every card says where the boundary sits and how it is checked. ### 1. Technical office: The software calculates, but the decisions are taken by one person The choices of someone with twenty years in the trade become rules written in the calculation programme; the regulatory checks remain a repeatable calculation. In many technical offices the programme runs the checks, while the choices that lead to an efficient solution stay with whoever has twenty years in the trade: where to start, how to correct when the checks do not add up, when a formally correct solution is not reasonable. The project turns these decisions into explicit constraints, heuristics (rules of thumb) and diagnostic rules, which become parameters of the calculation engine. On top of the core work specialised agents, instructed on historical cases and on the solutions actually adopted: one proposes the starting configuration for a new problem, one chooses the corrective strategy when optimisation does not converge, one compares the result with analogous cases and flags when it is mathematically correct but atypical. The two layers stay separate. The regulatory checks are repeatable: on the same data they always give the same result. AI works on top of the calculation, not inside it, and the expert stays in the loop to validate, correct and enrich. - How it is checked: 695 automated tests on the engine, run on every integration. The rebuilt engine reproduces 58 verification cases out of 63 to the printed digit, and 44 historical archives out of 44 are reread without exceptions by decoders written without having the documentation of the formats. - What is missing, declared: The fact base reaches 98.8% completeness — 399 entries out of 404 — and the 5 that remain are declared and justified instead of omitted. - Independent check: Eight campaigns run by agents tasked with refuting the work done. The refutations found were repaired before delivery, not filed away. ### 2. Application processing and back office: Artificial intelligence writes the draft; the company's rules decide the outcome. The conditions for accepting or rejecting a case sit in a file that the office edits without touching the code. Artificial intelligence prepares the data and communicates the outcome; it does not set it. On a process that produces official documents, the text can be written by artificial intelligence; whether the case is accepted or not is decided by the rules, not by the model. The conditions for accepting or rejecting a case sit in a readable configuration file that the technical office updates itself when a rule changes, without going through development. Artificial intelligence reads the emails, extracts the data and composes the reply; the calculation programme decides the outcome and justifies it. And sending remains a human gesture: the draft is ready, the signature belongs to whoever replies. - How it is checked: 135 automated tests passing, including six complete paths from email to the generated document. - Where the boundary sits: 100% of outbound communications go through an operator's approval. No automatic sending, by project choice and not by a technical limit. ### 3. Field crews: The crew plan also lists the jobs that cannot be fitted, and why. The programme can leave a job out, and it declares it with a cost tied to priority, instead of producing a plan that stands up on paper and not on the road. The value is not only the shortest route: it is knowing what stays out and why. Instead of forcing an unfeasible programme, the programme can leave a job out and declares it with a cost tied to priority. What the manager receives is an executable plan plus the justified list of what was not plannable — insufficient capacity, incompatible time window — instead of jobs that disappear in silence. Calculation time is also declared at the start: it is a project parameter, not a side effect of how large the problem is. - How it is checked: The indicators returned on every run — kilometres, driving hours, saturation per crew — make two different plans comparable, which is the only way to know whether the second is better than the first. - Declared limit: No measured before-and-after comparison on a real customer, therefore no saving percentage declared. ### 4. Public-facing services: When photo recognition is wrong, the correction is already designed and limited. The correction offered to the person shows only visually similar alternatives, and not the prices. A recognition system in the public's hands is judged by how it treats the cases in which it is wrong. Here image quality is assessed before sending, and a retake is asked for when the image is blurred or incomplete. After recognition the person confirms or corrects, but the correction is designed not to become a loophole: the list of alternatives contains only visually similar items, and prices are not visible. The burden of checking moves to the user without opening the door to abuse, and it is this choice, more than the model, that determines whether the system holds up in operation. - How it is checked: A regression suite on real images with declared reference truth: changes to the model or to the instructions are measured on a stable bench. - Declared limit: Accuracy is not yet a declarable number, because part of the bench has no reference truth. This is stated, instead of being estimated. ### 5. Analysis and reports: The numbers are calculated by code; AI writes the narrative No number in the report is born from a model: the quantities are calculated by a programme that, on the same frozen data, always gives the same result. In reports the separation is sharp. The quantities are calculated by a programme that, on the same frozen data, always gives the same result, with tests that reproduce them identical on every run; the text is written anchored to those numbers and to the real quotations. The finished document is then reread by synthetic reader profiles, with a declared quality threshold below which it is not delivered. And when the obvious metric does not discriminate — categories where every activity sits above 4.8 stars — the report says so, instead of building a verdict on top of it. - How it is checked: A test network reproduces already delivered reports byte for byte: a code change that would alter a number already delivered does not pass. - How every number is written: Every number carries its own denominator and its own source. Statements that find no support in the corpus are corrected, even when they had already circulated. ## Signals that it is time to intervene. - Important calculations depend on undocumented files that are hard to verify. - A technical application still works, but nobody wants to change it anymore. - Operational data exists, but it does not yet guide priorities, anomalies or forecasts. - Technical decisions depend on a few experts rather than on a shared system. - Leadership needs to invest but lacks a clear technical dossier on risk, value and feasibility. ## FAQ ### Is this generic software development? No. It is designed for problems that require technical domain knowledge, data, mathematics, algorithms, tests and verification criteria. ### Do specifications need to be complete already? No. Often the first task is to reconstruct specifications, rules, assumptions and edge cases from the existing system and expert users. ### Does AI decide instead of technicians? No. In technical contexts AI is used as support. Critical parts remain explainable, tested and under human responsibility. ### How is know-how protected? The project works with agreed boundaries, access, data and materials. Public examples use only anonymised descriptions that cannot identify the client. # Operational start: in one day an important process is automated, with a procedure that stays in the company and can be used at once. It starts from a job that today steals time: preparing a quote, gathering the documents for a customer, comparing three suppliers, sorting the week's mail. At the end of the day an automatic procedure remains, which the company will be able to use on its own. The files stay in the company. Custom integrations with existing software are excluded. It is not a course and it is not software to install: it is the automation of a whole process with artificial intelligence. ## What Operational start is. A course leaves notions. A setup alone leaves a tool nobody uses. Operational start closes both gaps on the same day: the assistant is in the company's name, the procedure is written, sample data already run, then real files take over. Files stay in the company. Artik Lab does not take them away. No connection to the ERP is installed. One job at a time, taken all the way through. ## When it is done. Someone in the company owns the procedure. The procedure is written. The following week they use it. A metric is stated before the day, even a rough one: time for a draft to reread, not a send on its own. ## When a job is finished A job is finished when four things are true together: a person in the company has the mandate to use the procedure; the procedure is written in the company's language; the measure was agreed before the day; in the following week the procedure was used on the real files. ## How it takes place. 1. **The need**: The process to lighten is described: who runs it today, how often per month, where time is lost. Knowing which tool to buy is not required. What goes in: who does the job today, how many times a month, where time is lost · What stays: the description of the job · Who decides: whoever does it today and leadership · How long it takes: a first conversation 2. **The check**: The procedure must be stable and describable, the data must exist and be exportable, an internal person must have the mandate to adopt it, and unresolved confidentiality constraints must not remain. Blocks are found here, before the proposal. What goes in: the description and a sample file · What stays: a written opinion: it fits in a day or it does not, and what will not be done · Who decides: Artik Lab · How long it takes: before proposing the day 3. **The proposal**: It states the activity, what the day contains, which licences the company must hold in its own name, which folders to prepare. Artik Lab does not resell licences and does not stand behind them. What goes in: the opinion · What stays: the proposal: which job, what the day contains, which subscription to activate, which folders to prepare · Who decides: the company · How long it takes: the time to read it 4. **The environment**: It arrives already set, with sample data. It can be opened and tried before the meeting. What goes in: the folders prepared · What stays: the procedure already set up, with samples, to try before the meeting · Who decides: the person with the mandate · How long it takes: before the day 5. **The first test**: One day on site: from samples to real files. At the end of the day a real result remains, not an exercise. If the process must change, that is consulting, not an extension of the day. What goes in: the real files · What stays: a real result and the written procedure · Who decides: the person with the mandate, with Artik Lab in the room · How long it takes: one day on site ## Some automation examples Eleven repetitive jobs that fit in a day. Each example states today's situation, what remains at the end of the day, where the person stops and when it is not suitable. The sector is an example, not a condition: the document changes, not the method. ### Quotes and price lists from mail and files A request arrives by email, often with scattered PDFs. An updated quote or price list is needed, with the product sheets attached, and whoever sells wastes time copying from three folders. - In brief: What goes in: request emails and PDFs, price list, sheets. What stays: the draft quote on the company's template, with the list of what is missing. Who checks: whoever sells, on price, discount and sending. - At the end of the day there remains: A procedure that, from the email and the PDFs, assembles the draft quote or price list on the company's template and flags what is missing. Whoever sells rereads the draft. - Where the person stops: price and discount; yes or no to the request; sending to the customer. - Files needed: request emails and PDFs; internal price list; product sheets; quote template. - Where it often happens: food processing; contract manufacturing in mechanics; construction and plant engineering. - When it is not suitable: If the quote is a take-off from a drawing or a negotiation: that is the trade, not this service. - On the form: tick «Quotes, price lists and sheets to assemble from mail and PDFs, for whoever sells to reread.». - URL: https://ar-tik.com/en/operational-start/quotes-from-mail.html ### The document pack that is asked for every time A customer, an auditor, a contracting authority or a shipper asks for «the papers». Today the pack is rebuilt from scratch every time, fishing from certified email (PEC), folders and portals from which a PDF is downloaded. - In brief: What goes in: the customer's or the tender's request, the list of pieces, scattered folders. What stays: checklist for the pack and folders put in order, with what to download and what to ask for in the house. Who checks: whoever hands over the pack. - At the end of the day there remains: A checklist for the pack, the folders put in order and a procedure that gathers the documents already in the house, flags those that are expired or missing and prepares the file to hand over. - Where the person stops: final check of the file; requesting the missing documents; handover or upload. - Files needed: PDFs already downloaded: DURC (social-security compliance certificate), company registry extracts, insurance policies, certificates; list of what is usually asked for; covering-letter template. - Where it often happens: construction and plant engineering; wine and export; mechanics for tenders and supplies. - When it is not suitable: The technician's signature, documents that exist only inside closed portals, uploading to tender platforms or legally compliant archiving. - On the form: tick «The document pack that clients, tenders, auditors or shippers ask for every time.». ### Comparing three supplier quotes Three PDFs or three emails in different formats, one of them incomplete. A table with price, times, yields and conditions is needed before whoever negotiates sits at the table. - In brief: What goes in: three disordered quotes, company criteria. What stays: comparison table ready for whoever negotiates. Who checks: whoever decides the supplier. - At the end of the day there remains: A fixed comparison grid of the company and a procedure that maps every quote onto the grid, highlights the differences and the missing items and prepares the questions to put to the suppliers. - Where the person stops: negotiation and choice of supplier; order in the company's software; search for new suppliers. - Files needed: quotes received as PDF or email; comparison grid or table already in use; purchase history, if it exists. - Where it often happens: food and packaging purchasing; mechanics and treatments; construction subcontracting. - When it is not suitable: If the choice depends on an open negotiation or if purchasing already goes through a platform with its own evaluation process. - On the form: tick «Compare three supplier quotes and have a table before whoever negotiates.». ### Ordinary mail and certified email (PEC) for the week Orders, requests for sheets, reminders, non-conformities and communications from authorities arrive in the inbox. A clear list is needed: what to do, what to forward to the professional firm, what to archive, which deadline. - In brief: What goes in: the week's inbox, sorting rules. What stays: what to do, what to forward, what to archive — without sending on its own. Who checks: whoever handles the mail. - At the end of the day there remains: Written sorting criteria on the company's type of mail, a forwarding template and a procedure that, from an export of the inbox, produces the weekly list with the deadlines. - Where the person stops: reply and sending; decisions on deadlines; everything that concerns people, payroll or accidents. - Files needed: export of the inbox as EML or PDF; list of senders and recurring topics; template for forwarding to the professional firm or the department. - Where it often happens: all sectors; tender certified email (PEC); companies with a mandatory digital domicile. - When it is not suitable: If it is necessary to enter the inbox without exporting, or if automatic sending of replies is asked for. - On the form: tick «Sort certified and ordinary mail for the week.». ### Sheets, certificates and packing lists to attach The part, the batch or the plant is already decided. What is missing is the file that accompanies the shipment: technical sheet, certificate, packing list, approval, each in a different folder. - In brief: What goes in: technical sheets and certificates as PDF, packing-list template, order already decided. What stays: the file ready for every order or shipment, with the order of attachments and the gaps flagged. Who checks: whoever checks the file before departure. - At the end of the day there remains: A procedure that says where the PDFs are, in what order they are attached and what to flag if a piece is missing, with the file ready for every order or shipment. - Where the person stops: check of the file before departure; testing and measurements; signing of certificates. - Files needed: technical sheets and certificates as PDF; packing-list template; order or confirmation already decided. - Where it often happens: mechanics with certified shipments; food with sheets and documentary HACCP; wine with packing list and origin. - When it is not suitable: Testing, measurements, in-line HACCP and drawings stay out: here the file is prepared, nothing is certified. - On the form: tick «Attach sheets, certificates and packing lists to a shipment or an order already decided.». ### A report from the Excel file already exported from the company's software The owner or the salesperson asks how customers, jobs or sales are going. The data are already in the company's software, but they come out as Excel and the report is rebuilt by hand every month. - In brief: What goes in: Excel or CSV export from the company's software, report template. What stays: lists and summary regenerated on every new export, with the written rules. Who checks: whoever reads the report and decides. - At the end of the day there remains: Written rules, for example from how many months a customer has been idle or how the mix is calculated, and a procedure that, from a new export, regenerates the lists and the summary for whoever must read it. - Where the person stops: reading and commercial decisions; contact with customers; correction of data in the company's software. - Files needed: Excel or CSV export from the company's software; template of the report prepared today; list of leadership's recurring questions. - Where it often happens: sales and commercial; jobs in mechanics; actuals from delivery notes. - When it is not suitable: If no export exists or if a forecast is asked for: here what exists is put in order, nothing is predicted. - On the form: tick «A report from an Excel export for the owner or the sales lead.». ### Minutes and action list from meeting notes After a meeting or a site visit, notes, photos and sometimes a recording remain. The minutes with the decisions and the things to do arrive late, or they do not arrive. - In brief: What goes in: notes, photos or transcript of the meeting, minutes template. What stays: the draft minutes with decisions, actions, owner and date. Who checks: whoever approves the minutes and sends them. - At the end of the day there remains: A company minutes template and a procedure that, from the notes or the transcript, prepares the draft with decisions, actions, owner and date, to be approved before it is circulated. - Where the person stops: approval of the minutes; final assignment of the actions; sending to the participants. - Files needed: notes, photos or transcript of the meeting; minutes template; list of participants. - Where it often happens: sites and inspections; production meetings; internal boards and committees. - When it is not suitable: Recordings without the consent of those present, or meetings in which people are discussed: that material is not copied into a folder. - On the form: tick «Turn the notes of a meeting or site visit into minutes and an action list.». ### Reply to complaints and non-conformities A customer's email arrives with photos, a lot or order number and a request. The reply has to be written well and quickly, and the non-conformity register is updated by hand. - In brief: What goes in: customer's email and attachments, reply template, non-conformity register. What stays: the draft reply and the line for the register, with the points to check. Who checks: whoever decides on the complaint and sends the reply. - At the end of the day there remains: A procedure that gathers the references (order, lot, documents), prepares the draft reply on the company's template and the line for the register, with the points to check highlighted. - Where the person stops: decision on the complaint: accept, reject, refund; analysis of the cause; sending of the reply. - Files needed: customer's email and attachments; reply template; non-conformity register in Excel. - Where it often happens: food and wine; mechanics and components; catalogue supplies. - When it is not suitable: If the reply is asked to go out on its own, or if the complaint concerns safety, health or disputes: there a person decides, at once. - On the form: tick «Draft the reply to a complaint or non-conformity, with the register updated.». ### Expiry schedule of documents and certificates DURC (social-security compliance certificate), insurance policies, certifications, licences, instrument calibrations: each with its own expiry, in PDFs scattered among folders and emails. The expiry is noticed when someone asks for the document. - In brief: What goes in: PDFs of documents with an expiry, list of what customers and auditors ask for. What stays: the table of expiries and the reminder of what expires in the month. Who checks: whoever renews the documents. - At the end of the day there remains: A table of expiries built from the PDFs, a procedure to update it when a new document arrives and a reminder of what expires in the month. - Where the person stops: renewal of the documents; request to authorities or suppliers; check of ambiguous dates. - Files needed: PDFs of documents with an expiry; list of what customers and auditors ask for; any table already in use. - Where it often happens: construction and plant engineering; road haulage and logistics; certified quality. - When it is not suitable: Documents that live only in closed portals, or expiries that depend on regulatory calculations: the table records them, it does not interpret them. - On the form: tick «Keep the expiry schedule of compliance certificates, insurance policies, certifications and licences, starting from the PDFs.». ### Sheets and price lists in another language for abroad The product sheet and the price list exist in Italian. The foreign customer wants them in their language, on the company template, and today they are translated in pieces with copy and paste. - In brief: What goes in: sheets and price lists in Italian, template in language, glossary. What stays: the draft in language on the chosen template, with the list of what must be reread. Who checks: whoever knows the language and the customer, on legal texts, prices and conditions. - At the end of the day there remains: A glossary of the company's terms, the procedure that produces the draft in language on the chosen template and the list of what must always be reread before sending. - Where the person stops: rereading by whoever knows the language and the customer; legal texts and labels; prices and conditions. - Files needed: sheets and price lists in Italian; template in language, even if imperfect; glossary or translations already approved. - Where it often happens: agrifood and wine; mechanics with foreign customers; components and catalogues. - When it is not suitable: Sworn translations, mandatory labelling and contractual texts: they stay with whoever is accountable for them. - On the form: tick «Prepare product sheets and price lists in another language on the company template, for someone to reread.». ### Consistency check among order, confirmation and delivery note (DDT) The customer's order, the order confirmation and the delivery note (DDT) should say the same thing. Codes, quantities and prices are checked by eye, and the differences are discovered at the invoice or at the complaint. - In brief: What goes in: customer's order, order confirmation, delivery note (DDT) as PDF. What stays: the list of differences, ordered by severity, before shipment or invoice. Who checks: whoever corrects in the company's software and decides what to ship. - At the end of the day there remains: A procedure that compares the three documents as PDF and produces the list of differences, ordered by severity, before shipment or invoice. - Where the person stops: correction in the company's software; decision on what to ship; contact with the customer. - Files needed: customer's order as PDF or email; order confirmation; delivery note (DDT). - Where it often happens: distribution and catalogue supplies; contract manufacturing in mechanics; food with frequent orders. - When it is not suitable: If the documents cannot be exported as PDF, or if issuing the delivery note (DDT) is asked for: that stays in the company's software. - On the form: tick «Compare the customer order, the order confirmation and the delivery note and flag the differences before shipping.». ## What is included. What is a second day. After one day on site, the company keeps the assistant in its own name and the written procedure. Right away, someone can run it. The ERP is not touched. If the process must change, that is a second day, and it is called consulting. - Included in the day: Instructions, folder conventions, review criteria for the chosen procedure. - Becomes consulting: Rewriting the method on a different process. - Included in the day: Sample data and their replacement with the company's files. - Becomes consulting: Cleaning or reorganising the archive. - Included in the day: On-site session on real files. - Becomes consulting: Repeated days on other teams or other cases. - Included in the day: Usage documentation. - Becomes consulting: Integration with ERP, company mail or closed portals. - Included in the day: Correction of what the day delivered. - Becomes consulting: Extended training of the organisation. ## Who this day is for. **Who it is for.** A company with a job that repeats, a person who does it today and files that can be copied into a folder. **Who this service is not for.** Whoever is looking for a general course on artificial intelligence, or a system connected to the company's software: those are other services, and they sit on this same page. ## Prerequisites. - A subscription to the assistant in the company's name. The proven environment is stated in the proposal, not on the homepage. - An internal person with the mandate to adopt the procedure. - Data exportable in the stated formats, without entering company systems. ## When it is not suitable. - The procedure is not yet stable, or nobody in the company can describe it. - The volume is negligible: a few runs a year do not repay adoption. - The needed data do not exist or can be rebuilt only by hand. - An unresolved confidentiality or data-processing constraint remains. - A permanent integration with systems is needed: that is software development. - The process must change or the ERP must be touched: that is consulting. ## What it is not, said once. - It is not a course. Courses remain, and teach general capability. - It is not software to install. Technical development stays downstream, when a permanent integration is needed. - It is not a plugin to the ERP. The order click is not touched. - It does not replace the clerk, the salesperson, the estimator, the technician. - It is not an agent that ships on its own, negotiates, or signs. - It is not a take-off from drawings, PriMus, AutoCAD. - It is not an upload to public procurement portals. - It is not payroll. - It is not a multi-week path with integration. If that is needed, it is another Artik service, already on the site. - Incentives, grants or vouchers are not promised. ## Operational start does not replace anything already on the site. - [Courses](https://ar-tik.com/en/courses/index.md): Courses remain the path of general capability. Operational start is the specific procedure switched on. - [AI management consulting](https://ar-tik.com/en/ai-management-consulting.md): Consulting sits upstream, or is day 2 if the process is touched. - [Data analysis](https://ar-tik.com/en/agentic-data-analysis.md): Data analysis sits beside it, when the problem is the signal in the data. - [Software](https://ar-tik.com/en/technical-software-development.md): Software development sits downstream, when a permanent integration is needed. - [AI atlas](https://ar-tik.com/en/ai-applications-atlas.md): The Atlas helps recognise the process. - [FAQ](https://ar-tik.com/en/ai-business-faq.md): The FAQ helps choose among services. ## Three fields. A feasibility report. Then, if there is a fit, the day. The report is a door, not the product. The public website indicated is used. No mandatory company name, no tax number, no headcount. After the report: if there is a fit, a check and the day; if not, a course or consulting. ## FAQ ### Is this a course? No. A course leaves templates and criteria. Here what remains is a written procedure, already set up on the artificial intelligence in the company's name, with the files that stay in the company. ### Are custom integrations with existing software done? No. Custom integrations with existing software are excluded. If the systems already in the company need to be connected, that is consulting or software development: another job. ### Do the data leave the company? The files stay in the company. The assistant is in the company's name. Artik Lab does not take them away. ### Can five jobs be done together? No, one at a time. The day is there to leave a procedure in use, and one procedure in use is worth more than five unfinished drafts. ### Is the form opinion already the service? No. It is only to understand whether it is a fit. After: if it makes sense, a check and the day; if not, a course or consulting. ### What does it cost? The proposal arrives after the opinion. This page has no price list. ### How many activities can be done, one after another? It is decided with the company, according to need. One job at a time, taken all the way through. ### Can an incentive or a fund be used for this work? An eligibility check can be requested. The work must still stand without an incentive: no calls for proposals or amounts are promised. # Corporate AI courses: choose the right path Artificial intelligence creates value when it enters daily work with method: documents, recurring decisions, handoffs between functions, quality controls and operational responsibilities. Artik Lab courses are not a generic catalog of AI lessons. They are paths designed around the client's processes, built to turn individual use of AI tools into governable, measurable and repeatable company practices. ## Start from AI Workflow Redesign Lab For most companies, the first issue is not choosing an AI tool. It is understanding which activities should be redesigned, which data can be used, which risks need governance and which capabilities must remain inside the company. ## Training adapted to the company context, not standard lessons. Dynamic Training Rework is Artik Lab's proprietary method: the path is not identical from the first to the last session, but is recalibrated around processes, roles, materials and priorities that emerge with participants. Companies are flooded with generic AI training, often full of examples far from real work. DTR avoids that waste and turns advanced AI tools into practices people can actually use. ## Before choosing the format, recognise the process. The Atlas gathers concrete AI application examples across documents, operations, HR, marketing, software, governance, production, training and data. It helps decide whether the need requires consulting, data analysis, technical development or training. [Open the Atlas](https://ar-tik.com/en/ai-applications-atlas.md) ## Course catalog ### AI Workflow Redesign Lab Practical corporate course for applying AI to workflow redesign, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, configurable as 2 or 4 sessions - Choose it if: When the company wants concrete progress on workflow redesign and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for workflow redesign. - URL: https://ar-tik.com/en/courses/workflow-redesign.html ### Managing AI Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on managerial AI adoption and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for managerial AI adoption. - URL: https://ar-tik.com/en/courses/managing-ai.html ### Managing AI for mixed company teams Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on cross-functional AI alignment and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for cross-functional AI alignment. - URL: https://ar-tik.com/en/courses/managing-ai-general.html ### Operational AI Governance Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on operational AI governance and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for operational AI governance. - URL: https://ar-tik.com/en/courses/ai-governance.html ### AI Business Case & ROI Sprint Practical corporate course for applying AI to AI business cases and ROI, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours or a half-day sprint - Choose it if: When the company wants concrete progress on AI business cases and ROI and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for AI business cases and ROI. - URL: https://ar-tik.com/en/courses/ai-business-case-roi.html ### AI Adoption Manager / AI Champions Practical corporate course for applying AI to AI adoption champions, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 6-8 hours, configurable - Choose it if: When the company wants concrete progress on AI adoption champions and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for AI adoption champions. - URL: https://ar-tik.com/en/courses/ai-adoption-manager.html ### Role-Based AI Literacy & Responsible Use Practical corporate course for applying AI to role-based AI literacy, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-8 hours, adaptable by role - Choose it if: When the company wants concrete progress on role-based AI literacy and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for role-based AI literacy. - URL: https://ar-tik.com/en/courses/ai-literacy.html ### AI course: managing documents with AI Practical corporate course for applying AI to document management, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Choose it if: When the company wants concrete progress on document management and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for document management. - URL: https://ar-tik.com/en/courses/ai-documenti.html ### AI course: AI-driven marketing and communication Practical corporate course for applying AI to marketing and communication, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Choose it if: When the company wants concrete progress on marketing and communication and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for marketing and communication. - URL: https://ar-tik.com/en/courses/ai-marketing.html ### AI course: B2C and B2B sales with AI Practical corporate course for applying AI to B2B and B2C sales, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Choose it if: When the company wants concrete progress on B2B and B2C sales and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for B2B and B2C sales. - URL: https://ar-tik.com/en/courses/ai-vendite.html ### AI for administration and management control Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on finance and management control and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for finance and management control. - URL: https://ar-tik.com/en/courses/ai-admin-finance.html ### AI Operations Practical corporate course for applying AI to operations and process coordination, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on operations and process coordination and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for operations and process coordination. - URL: https://ar-tik.com/en/courses/ai-operations.html ### AI Legal Ops and compliance documentation Practical corporate course for applying AI to legal operations and compliance documents, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on legal operations and compliance documents and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for legal operations and compliance documents. - URL: https://ar-tik.com/en/courses/ai-legal-ops.html ### AI for procurement and supplier intelligence Practical corporate course for applying AI to procurement and supplier intelligence, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on procurement and supplier intelligence and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for procurement and supplier intelligence. - URL: https://ar-tik.com/en/courses/ai-procurement.html ### AI for customer service and ticket triage Practical corporate course for applying AI to customer service and ticket triage, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on customer service and ticket triage and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for customer service and ticket triage. - URL: https://ar-tik.com/en/courses/ai-customer-service.html ### AI for quality and non-conformities Practical corporate course for applying AI to quality and non-conformities, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on quality and non-conformities and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for quality and non-conformities. - URL: https://ar-tik.com/en/courses/ai-quality-management.html ### AI People Ops Practical corporate course for applying AI to People Ops and HR workflows, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on People Ops and HR workflows and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for People Ops and HR workflows. - URL: https://ar-tik.com/en/courses/ai-people-ops.html ### AI Brand Voice and communication Practical corporate course for applying AI to brand voice and communication, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on brand voice and communication and needs training that produces usable workflows, not abstract theory. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for brand voice and communication. - URL: https://ar-tik.com/en/courses/ai-brand-voice.html ### Semantic search and AI knowledge bases Turn archives and internal knowledge into search by meaning, with embeddings: numerical representations of the meaning of a text. - Duration: 60-90 minuti o modulo breve - Choose it if: When the company wants concrete progress on semantic search and knowledge bases and needs training that produces usable workflows, not abstract theory. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for semantic search and knowledge bases. - URL: https://ar-tik.com/en/courses/embeddings.html ### RAG Engineering for reliable AI systems Design RAG (Retrieval-Augmented Generation) systems: they retrieve company documents and generate the answer citing them. - Duration: 5 hours, two 2.5-hour sessions - Choose it if: When the company wants concrete progress on RAG engineering and needs training that produces usable workflows, not abstract theory. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for RAG engineering. - URL: https://ar-tik.com/en/courses/rag-engineering.html ### AI Coding Agents for software teams Use coding agents as controlled parts of the software cycle: development, review, tests and release, with named responsibility. - Duration: 4 hours, two 2-hour sessions - Choose it if: When the company wants concrete progress on AI coding agents and needs training that produces usable workflows, not abstract theory. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for AI coding agents. - URL: https://ar-tik.com/en/courses/ai-coding-agents.html ### AI Software Engineering Design AI-native software, with AI inside the product not beside it: architecture, evaluation and go-live (entry into production). - Duration: 5 hours, two 2.5-hour sessions - Choose it if: When the company wants concrete progress on AI software engineering and needs training that produces usable workflows, not abstract theory. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for AI software engineering. - URL: https://ar-tik.com/en/courses/ai-software-engineering.html ### Secure AI SDLC Put AI-specific controls into the software lifecycle: design, build, test, release and operation (planning, construction, testing, release and running in production). - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on secure AI software lifecycle and needs training that produces usable workflows, not abstract theory. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for secure AI software lifecycle. - URL: https://ar-tik.com/en/courses/secure-ai-sdlc.html ### Secure AI at Work Practical corporate course for applying AI to secure AI use at work, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, adaptable by function - Choose it if: When the company wants concrete progress on secure AI use at work and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for secure AI use at work. - URL: https://ar-tik.com/en/courses/secure-ai-at-work.html ### AI Output Quality & Human Review Practical corporate course for applying AI to AI output quality and human review, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Choose it if: When the company wants concrete progress on AI output quality and human review and needs training that produces usable workflows, not abstract theory. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for AI output quality and human review. - URL: https://ar-tik.com/en/courses/ai-output-quality.html ## FAQ ### Why start from AI Workflow Redesign Lab? Because before introducing tools or automation the company needs to understand which workflows have real potential, which data can be used and which controls are necessary. ### Are courses standard or customised? The structure is stable, but content, examples, exercises and priorities are adapted to the client's processes. ### Is programming required? No for introductory, managerial and operational paths. It is required only in technical paths. ### What remains after the course? Operating materials, context-specific examples, usage criteria and a view of high-potential workflows. # AI Workflow Redesign Lab Practical corporate course for applying AI to workflow redesign, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, configurable as 2 or 4 sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for workflow redesign. - Choose it if: When the company wants concrete progress on workflow redesign and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach workflow redesign through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on workflow redesign and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Map the real work Activities, decisions, information handoffs and bottlenecks. ### 2. Assess AI potential Impact, feasibility, risk, data quality and control level. ### 3. Redesign the workflow Roles, inputs, outputs, reviews, escalation and traceability. ### 4. Move into adoption Metrics, minimum governance, roadmap and operating ownership. ## Practical exercises - Map a realistic process connected to workflow redesign. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for workflow redesign. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Managing AI](https://ar-tik.com/en/courses/managing-ai.md): Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. - [Managing AI for mixed company teams](https://ar-tik.com/en/courses/managing-ai-general.md): Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. - [Operational AI Governance](https://ar-tik.com/en/courses/ai-governance.md): Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Managing AI Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for managerial AI adoption. - Choose it if: When the company wants concrete progress on managerial AI adoption and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach managerial AI adoption through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on managerial AI adoption and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to managerial AI adoption. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for managerial AI adoption. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Managing AI for mixed company teams](https://ar-tik.com/en/courses/managing-ai-general.md): Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. - [Operational AI Governance](https://ar-tik.com/en/courses/ai-governance.md): Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Business Case & ROI Sprint](https://ar-tik.com/en/courses/ai-business-case-roi.md): Practical corporate course for applying AI to AI business cases and ROI, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Managing AI for mixed company teams Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for cross-functional AI alignment. - Choose it if: When the company wants concrete progress on cross-functional AI alignment and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach cross-functional AI alignment through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on cross-functional AI alignment and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to cross-functional AI alignment. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for cross-functional AI alignment. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Operational AI Governance](https://ar-tik.com/en/courses/ai-governance.md): Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Business Case & ROI Sprint](https://ar-tik.com/en/courses/ai-business-case-roi.md): Practical corporate course for applying AI to AI business cases and ROI, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Adoption Manager / AI Champions](https://ar-tik.com/en/courses/ai-adoption-manager.md): Practical corporate course for applying AI to AI adoption champions, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Operational AI Governance Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for operational AI governance. - Choose it if: When the company wants concrete progress on operational AI governance and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach operational AI governance through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on operational AI governance and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to operational AI governance. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for operational AI governance. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Business Case & ROI Sprint](https://ar-tik.com/en/courses/ai-business-case-roi.md): Practical corporate course for applying AI to AI business cases and ROI, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Adoption Manager / AI Champions](https://ar-tik.com/en/courses/ai-adoption-manager.md): Practical corporate course for applying AI to AI adoption champions, with exercises on realistic work, reusable materials and clear governance criteria. - [Role-Based AI Literacy & Responsible Use](https://ar-tik.com/en/courses/ai-literacy.md): Practical corporate course for applying AI to role-based AI literacy, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Business Case & ROI Sprint Practical corporate course for applying AI to AI business cases and ROI, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours or a half-day sprint - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for AI business cases and ROI. - Choose it if: When the company wants concrete progress on AI business cases and ROI and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach AI business cases and ROI through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on AI business cases and ROI and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to AI business cases and ROI. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for AI business cases and ROI. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Adoption Manager / AI Champions](https://ar-tik.com/en/courses/ai-adoption-manager.md): Practical corporate course for applying AI to AI adoption champions, with exercises on realistic work, reusable materials and clear governance criteria. - [Role-Based AI Literacy & Responsible Use](https://ar-tik.com/en/courses/ai-literacy.md): Practical corporate course for applying AI to role-based AI literacy, with exercises on realistic work, reusable materials and clear governance criteria. - [Secure AI at Work](https://ar-tik.com/en/courses/secure-ai-at-work.md): Practical corporate course for applying AI to secure AI use at work, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Adoption Manager / AI Champions Practical corporate course for applying AI to AI adoption champions, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 6-8 hours, configurable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for AI adoption champions. - Choose it if: When the company wants concrete progress on AI adoption champions and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach AI adoption champions through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on AI adoption champions and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to AI adoption champions. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for AI adoption champions. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Role-Based AI Literacy & Responsible Use](https://ar-tik.com/en/courses/ai-literacy.md): Practical corporate course for applying AI to role-based AI literacy, with exercises on realistic work, reusable materials and clear governance criteria. - [Secure AI at Work](https://ar-tik.com/en/courses/secure-ai-at-work.md): Practical corporate course for applying AI to secure AI use at work, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Output Quality & Human Review](https://ar-tik.com/en/courses/ai-output-quality.md): Practical corporate course for applying AI to AI output quality and human review, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Role-Based AI Literacy & Responsible Use Practical corporate course for applying AI to role-based AI literacy, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-8 hours, adaptable by role - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for role-based AI literacy. - Choose it if: When the company wants concrete progress on role-based AI literacy and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach role-based AI literacy through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on role-based AI literacy and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to role-based AI literacy. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for role-based AI literacy. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Secure AI at Work](https://ar-tik.com/en/courses/secure-ai-at-work.md): Practical corporate course for applying AI to secure AI use at work, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Output Quality & Human Review](https://ar-tik.com/en/courses/ai-output-quality.md): Practical corporate course for applying AI to AI output quality and human review, with exercises on realistic work, reusable materials and clear governance criteria. - [Managing AI](https://ar-tik.com/en/courses/managing-ai.md): Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI course: managing documents with AI Practical corporate course for applying AI to document management, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for document management. - Choose it if: When the company wants concrete progress on document management and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach document management through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on document management and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Work context Documents, channels, tasks and recurring decisions. ### 2. Practical AI use Instructions, examples, review criteria and limits. ### 3. Reusable workflow Templates, checklists, handoffs and controls. ### 4. Safe adoption Data, privacy, quality and next steps. ## Practical exercises - Map a realistic process connected to document management. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for document management. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI course: AI-driven marketing and communication](https://ar-tik.com/en/courses/ai-marketing.md): Practical corporate course for applying AI to marketing and communication, with exercises on realistic work, reusable materials and clear governance criteria. - [AI course: B2C and B2B sales with AI](https://ar-tik.com/en/courses/ai-vendite.md): Practical corporate course for applying AI to B2B and B2C sales, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI course: AI-driven marketing and communication Practical corporate course for applying AI to marketing and communication, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for marketing and communication. - Choose it if: When the company wants concrete progress on marketing and communication and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach marketing and communication through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on marketing and communication and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Work context Documents, channels, tasks and recurring decisions. ### 2. Practical AI use Instructions, examples, review criteria and limits. ### 3. Reusable workflow Templates, checklists, handoffs and controls. ### 4. Safe adoption Data, privacy, quality and next steps. ## Practical exercises - Map a realistic process connected to marketing and communication. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for marketing and communication. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI course: B2C and B2B sales with AI](https://ar-tik.com/en/courses/ai-vendite.md): Practical corporate course for applying AI to B2B and B2C sales, with exercises on realistic work, reusable materials and clear governance criteria. - [AI course: managing documents with AI](https://ar-tik.com/en/courses/ai-documenti.md): Practical corporate course for applying AI to document management, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI course: B2C and B2B sales with AI Practical corporate course for applying AI to B2B and B2C sales, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 8 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for B2B and B2C sales. - Choose it if: When the company wants concrete progress on B2B and B2C sales and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach B2B and B2C sales through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on B2B and B2C sales and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Work context Documents, channels, tasks and recurring decisions. ### 2. Practical AI use Instructions, examples, review criteria and limits. ### 3. Reusable workflow Templates, checklists, handoffs and controls. ### 4. Safe adoption Data, privacy, quality and next steps. ## Practical exercises - Map a realistic process connected to B2B and B2C sales. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for B2B and B2C sales. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI course: managing documents with AI](https://ar-tik.com/en/courses/ai-documenti.md): Practical corporate course for applying AI to document management, with exercises on realistic work, reusable materials and clear governance criteria. - [AI course: AI-driven marketing and communication](https://ar-tik.com/en/courses/ai-marketing.md): Practical corporate course for applying AI to marketing and communication, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI for administration and management control Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for finance and management control. - Choose it if: When the company wants concrete progress on finance and management control and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach finance and management control through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on finance and management control and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to finance and management control. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for finance and management control. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Operations](https://ar-tik.com/en/courses/ai-operations.md): Practical corporate course for applying AI to operations and process coordination, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Legal Ops and compliance documentation](https://ar-tik.com/en/courses/ai-legal-ops.md): Practical corporate course for applying AI to legal operations and compliance documents, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for procurement and supplier intelligence](https://ar-tik.com/en/courses/ai-procurement.md): Practical corporate course for applying AI to procurement and supplier intelligence, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Operations Practical corporate course for applying AI to operations and process coordination, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for operations and process coordination. - Choose it if: When the company wants concrete progress on operations and process coordination and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach operations and process coordination through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on operations and process coordination and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to operations and process coordination. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for operations and process coordination. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Legal Ops and compliance documentation](https://ar-tik.com/en/courses/ai-legal-ops.md): Practical corporate course for applying AI to legal operations and compliance documents, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for procurement and supplier intelligence](https://ar-tik.com/en/courses/ai-procurement.md): Practical corporate course for applying AI to procurement and supplier intelligence, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for customer service and ticket triage](https://ar-tik.com/en/courses/ai-customer-service.md): Practical corporate course for applying AI to customer service and ticket triage, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Legal Ops and compliance documentation Practical corporate course for applying AI to legal operations and compliance documents, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for legal operations and compliance documents. - Choose it if: When the company wants concrete progress on legal operations and compliance documents and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach legal operations and compliance documents through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on legal operations and compliance documents and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to legal operations and compliance documents. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for legal operations and compliance documents. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI for procurement and supplier intelligence](https://ar-tik.com/en/courses/ai-procurement.md): Practical corporate course for applying AI to procurement and supplier intelligence, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for customer service and ticket triage](https://ar-tik.com/en/courses/ai-customer-service.md): Practical corporate course for applying AI to customer service and ticket triage, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for quality and non-conformities](https://ar-tik.com/en/courses/ai-quality-management.md): Practical corporate course for applying AI to quality and non-conformities, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI for procurement and supplier intelligence Practical corporate course for applying AI to procurement and supplier intelligence, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for procurement and supplier intelligence. - Choose it if: When the company wants concrete progress on procurement and supplier intelligence and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach procurement and supplier intelligence through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on procurement and supplier intelligence and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to procurement and supplier intelligence. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for procurement and supplier intelligence. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI for customer service and ticket triage](https://ar-tik.com/en/courses/ai-customer-service.md): Practical corporate course for applying AI to customer service and ticket triage, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for quality and non-conformities](https://ar-tik.com/en/courses/ai-quality-management.md): Practical corporate course for applying AI to quality and non-conformities, with exercises on realistic work, reusable materials and clear governance criteria. - [AI People Ops](https://ar-tik.com/en/courses/ai-people-ops.md): Practical corporate course for applying AI to People Ops and HR workflows, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI for customer service and ticket triage Practical corporate course for applying AI to customer service and ticket triage, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for customer service and ticket triage. - Choose it if: When the company wants concrete progress on customer service and ticket triage and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach customer service and ticket triage through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on customer service and ticket triage and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to customer service and ticket triage. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for customer service and ticket triage. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI for quality and non-conformities](https://ar-tik.com/en/courses/ai-quality-management.md): Practical corporate course for applying AI to quality and non-conformities, with exercises on realistic work, reusable materials and clear governance criteria. - [AI People Ops](https://ar-tik.com/en/courses/ai-people-ops.md): Practical corporate course for applying AI to People Ops and HR workflows, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Brand Voice and communication](https://ar-tik.com/en/courses/ai-brand-voice.md): Practical corporate course for applying AI to brand voice and communication, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI for quality and non-conformities Practical corporate course for applying AI to quality and non-conformities, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for quality and non-conformities. - Choose it if: When the company wants concrete progress on quality and non-conformities and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach quality and non-conformities through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on quality and non-conformities and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to quality and non-conformities. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for quality and non-conformities. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI People Ops](https://ar-tik.com/en/courses/ai-people-ops.md): Practical corporate course for applying AI to People Ops and HR workflows, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Brand Voice and communication](https://ar-tik.com/en/courses/ai-brand-voice.md): Practical corporate course for applying AI to brand voice and communication, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI People Ops Practical corporate course for applying AI to People Ops and HR workflows, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for People Ops and HR workflows. - Choose it if: When the company wants concrete progress on People Ops and HR workflows and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach People Ops and HR workflows through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on People Ops and HR workflows and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to People Ops and HR workflows. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for People Ops and HR workflows. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Brand Voice and communication](https://ar-tik.com/en/courses/ai-brand-voice.md): Practical corporate course for applying AI to brand voice and communication, with exercises on realistic work, reusable materials and clear governance criteria. - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Operations](https://ar-tik.com/en/courses/ai-operations.md): Practical corporate course for applying AI to operations and process coordination, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Brand Voice and communication Practical corporate course for applying AI to brand voice and communication, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For managers and non-technical teams; no programming required. - Final output: Operating canvas for brand voice and communication. - Choose it if: When the company wants concrete progress on brand voice and communication and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach brand voice and communication through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For managers and non-technical teams; no programming required. ## When to choose it Choose this course when the company wants concrete progress on brand voice and communication and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Process map Recurring work, documents, decisions, exceptions and handoffs. ### 2. AI-assisted outputs Summaries, classifications, draft responses, reports and checklists. ### 3. Workflow controls Review, escalation, traceability and responsibility. ### 4. Operational adoption Metrics, materials, routines and governance. ## Practical exercises - Map a realistic process connected to brand voice and communication. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for brand voice and communication. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI for administration and management control](https://ar-tik.com/en/courses/ai-admin-finance.md): Practical corporate course for applying AI to finance and management control, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Operations](https://ar-tik.com/en/courses/ai-operations.md): Practical corporate course for applying AI to operations and process coordination, with exercises on realistic work, reusable materials and clear governance criteria. - [AI Legal Ops and compliance documentation](https://ar-tik.com/en/courses/ai-legal-ops.md): Practical corporate course for applying AI to legal operations and compliance documents, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Semantic search and AI knowledge bases Turn archives and internal knowledge into search by meaning, with embeddings: numerical representations of the meaning of a text. - Duration: 60-90 minuti o modulo breve - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for semantic search and knowledge bases. - Choose it if: When the company wants concrete progress on semantic search and knowledge bases and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach semantic search and knowledge bases through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For technical teams with programming and software architecture basics. ## When to choose it Choose this course when the company wants concrete progress on semantic search and knowledge bases and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Architecture and requirements Goals, boundaries, data, services and risk assumptions. ### 2. Build and integration patterns Pipelines, interfaces, context, permissions and testing. ### 3. Evaluation and quality Metrics, review, regression tests and failure modes. ### 4. Production and governance Monitoring, security, audit, cost and maintenance. ## Practical exercises - Map a realistic process connected to semantic search and knowledge bases. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for semantic search and knowledge bases. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites Basic technical familiarity with software, data or system architecture is recommended. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [RAG Engineering for reliable AI systems](https://ar-tik.com/en/courses/rag-engineering.md): Design RAG (Retrieval-Augmented Generation) systems: they retrieve company documents and generate the answer citing them. - [AI Coding Agents for software teams](https://ar-tik.com/en/courses/ai-coding-agents.md): Use coding agents as controlled parts of the software cycle: development, review, tests and release, with named responsibility. - [AI Software Engineering](https://ar-tik.com/en/courses/ai-software-engineering.md): Design AI-native software, with AI inside the product not beside it: architecture, evaluation and go-live (entry into production). [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # RAG Engineering for reliable AI systems Design RAG (Retrieval-Augmented Generation) systems: they retrieve company documents and generate the answer citing them. - Duration: 5 hours, two 2.5-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for RAG engineering. - Choose it if: When the company wants concrete progress on RAG engineering and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach RAG engineering through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For technical teams with programming and software architecture basics. ## When to choose it Choose this course when the company wants concrete progress on RAG engineering and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Architecture and requirements Goals, boundaries, data, services and risk assumptions. ### 2. Build and integration patterns Pipelines, interfaces, context, permissions and testing. ### 3. Evaluation and quality Metrics, review, regression tests and failure modes. ### 4. Production and governance Monitoring, security, audit, cost and maintenance. ## Practical exercises - Map a realistic process connected to RAG engineering. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for RAG engineering. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites Basic technical familiarity with software, data or system architecture is recommended. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Coding Agents for software teams](https://ar-tik.com/en/courses/ai-coding-agents.md): Use coding agents as controlled parts of the software cycle: development, review, tests and release, with named responsibility. - [AI Software Engineering](https://ar-tik.com/en/courses/ai-software-engineering.md): Design AI-native software, with AI inside the product not beside it: architecture, evaluation and go-live (entry into production). - [Secure AI SDLC](https://ar-tik.com/en/courses/secure-ai-sdlc.md): Put AI-specific controls into the software lifecycle: design, build, test, release and operation (planning, construction, testing, release and running in production). [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Coding Agents for software teams Use coding agents as controlled parts of the software cycle: development, review, tests and release, with named responsibility. - Duration: 4 hours, two 2-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for AI coding agents. - Choose it if: When the company wants concrete progress on AI coding agents and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach AI coding agents through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For technical teams with programming and software architecture basics. ## When to choose it Choose this course when the company wants concrete progress on AI coding agents and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Architecture and requirements Goals, boundaries, data, services and risk assumptions. ### 2. Build and integration patterns Pipelines, interfaces, context, permissions and testing. ### 3. Evaluation and quality Metrics, review, regression tests and failure modes. ### 4. Production and governance Monitoring, security, audit, cost and maintenance. ## Practical exercises - Map a realistic process connected to AI coding agents. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for AI coding agents. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites Basic technical familiarity with software, data or system architecture is recommended. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Software Engineering](https://ar-tik.com/en/courses/ai-software-engineering.md): Design AI-native software, with AI inside the product not beside it: architecture, evaluation and go-live (entry into production). - [Secure AI SDLC](https://ar-tik.com/en/courses/secure-ai-sdlc.md): Put AI-specific controls into the software lifecycle: design, build, test, release and operation (planning, construction, testing, release and running in production). - [Semantic search and AI knowledge bases](https://ar-tik.com/en/courses/embeddings.md): Turn archives and internal knowledge into search by meaning, with embeddings: numerical representations of the meaning of a text. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Software Engineering Design AI-native software, with AI inside the product not beside it: architecture, evaluation and go-live (entry into production). - Duration: 5 hours, two 2.5-hour sessions - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for AI software engineering. - Choose it if: When the company wants concrete progress on AI software engineering and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach AI software engineering through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For technical teams with programming and software architecture basics. ## When to choose it Choose this course when the company wants concrete progress on AI software engineering and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Architecture and requirements Goals, boundaries, data, services and risk assumptions. ### 2. Build and integration patterns Pipelines, interfaces, context, permissions and testing. ### 3. Evaluation and quality Metrics, review, regression tests and failure modes. ### 4. Production and governance Monitoring, security, audit, cost and maintenance. ## Practical exercises - Map a realistic process connected to AI software engineering. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for AI software engineering. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites Basic technical familiarity with software, data or system architecture is recommended. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Secure AI SDLC](https://ar-tik.com/en/courses/secure-ai-sdlc.md): Put AI-specific controls into the software lifecycle: design, build, test, release and operation (planning, construction, testing, release and running in production). - [Semantic search and AI knowledge bases](https://ar-tik.com/en/courses/embeddings.md): Turn archives and internal knowledge into search by meaning, with embeddings: numerical representations of the meaning of a text. - [RAG Engineering for reliable AI systems](https://ar-tik.com/en/courses/rag-engineering.md): Design RAG (Retrieval-Augmented Generation) systems: they retrieve company documents and generate the answer citing them. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Secure AI SDLC Put AI-specific controls into the software lifecycle: design, build, test, release and operation (planning, construction, testing, release and running in production). - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For technical teams with programming and software architecture basics. - Final output: Operating canvas for secure AI software lifecycle. - Choose it if: When the company wants concrete progress on secure AI software lifecycle and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach secure AI software lifecycle through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For technical teams with programming and software architecture basics. ## When to choose it Choose this course when the company wants concrete progress on secure AI software lifecycle and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Architecture and requirements Goals, boundaries, data, services and risk assumptions. ### 2. Build and integration patterns Pipelines, interfaces, context, permissions and testing. ### 3. Evaluation and quality Metrics, review, regression tests and failure modes. ### 4. Production and governance Monitoring, security, audit, cost and maintenance. ## Practical exercises - Map a realistic process connected to secure AI software lifecycle. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for secure AI software lifecycle. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites Basic technical familiarity with software, data or system architecture is recommended. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Semantic search and AI knowledge bases](https://ar-tik.com/en/courses/embeddings.md): Turn archives and internal knowledge into search by meaning, with embeddings: numerical representations of the meaning of a text. - [RAG Engineering for reliable AI systems](https://ar-tik.com/en/courses/rag-engineering.md): Design RAG (Retrieval-Augmented Generation) systems: they retrieve company documents and generate the answer citing them. - [AI Coding Agents for software teams](https://ar-tik.com/en/courses/ai-coding-agents.md): Use coding agents as controlled parts of the software cycle: development, review, tests and release, with named responsibility. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Secure AI at Work Practical corporate course for applying AI to secure AI use at work, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4 hours, adaptable by function - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for secure AI use at work. - Choose it if: When the company wants concrete progress on secure AI use at work and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach secure AI use at work through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on secure AI use at work and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to secure AI use at work. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for secure AI use at work. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [AI Output Quality & Human Review](https://ar-tik.com/en/courses/ai-output-quality.md): Practical corporate course for applying AI to AI output quality and human review, with exercises on realistic work, reusable materials and clear governance criteria. - [Managing AI](https://ar-tik.com/en/courses/managing-ai.md): Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. - [Managing AI for mixed company teams](https://ar-tik.com/en/courses/managing-ai-general.md): Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # AI Output Quality & Human Review Practical corporate course for applying AI to AI output quality and human review, with exercises on realistic work, reusable materials and clear governance criteria. - Duration: 4-6 hours, customisable - Mode: In-person or online lab, with guided exercises and materials adapted to the client. - Profile: For company teams, operational functions and managers; no programming required. - Final output: Operating canvas for AI output quality and human review. - Choose it if: When the company wants concrete progress on AI output quality and human review and needs training that produces usable workflows, not abstract theory. ## The problem it solves Companies often approach AI output quality and human review through scattered experiments: a few prompts, a few enthusiastic users, many doubts about data, quality and responsibility. This course turns that uncertainty into an operating method. Participants work on realistic scenarios, learn where AI helps, where human review remains essential and how to make the practice repeatable inside the company. ## Audience For company teams, operational functions and managers; no programming required. ## When to choose it Choose this course when the company wants concrete progress on AI output quality and human review and needs training that produces usable workflows, not abstract theory. ## Concrete outcomes - Map the work and the decisions where AI can reduce friction. - Build practical instructions, checklists and review criteria. - Identify data, privacy and responsibility boundaries. - Create reusable examples for the team. - Define next steps for adoption and governance. ## Program ### 1. Shared understanding Capabilities, limits, responsibilities and business implications. ### 2. Use-case evaluation Value, feasibility, risk, data and ownership. ### 3. Governance and quality Rules, review, escalation and accountable decisions. ### 4. Adoption roadmap Priorities, skills, metrics and next steps. ## Practical exercises - Map a realistic process connected to AI output quality and human review. - Create AI-assisted outputs and review them critically. - Define escalation and human review points. - Build a reusable checklist for daily work. ## Materials delivered - Operating canvas for AI output quality and human review. - Prompt and instruction templates. - Quality and privacy checklist. - Risk/control matrix. - Adoption notes for the team. ## Data, privacy and limits The course uses synthetic, public, anonymised or client-approved materials. It explains how to minimise data exposure, protect confidential information, verify outputs and keep human responsibility explicit. ## Prerequisites No programming required. Familiarity with the business process is useful. ## FAQ ### Is the course tool-specific? No. Patterns and workflows are adapted to the tools and policies chosen with the client. ### Can company data be used? Only when accounts, contracts and internal policies allow it. Otherwise synthetic or anonymised data is used. ### What remains after the course? Reusable materials, examples, checklists and a clear set of next steps. ### Is it theoretical? No. The course is built around practical exercises and decisions close to real work. ## Related courses - [Managing AI](https://ar-tik.com/en/courses/managing-ai.md): Practical corporate course for applying AI to managerial AI adoption, with exercises on realistic work, reusable materials and clear governance criteria. - [Managing AI for mixed company teams](https://ar-tik.com/en/courses/managing-ai-general.md): Practical corporate course for applying AI to cross-functional AI alignment, with exercises on realistic work, reusable materials and clear governance criteria. - [Operational AI Governance](https://ar-tik.com/en/courses/ai-governance.md): Practical corporate course for applying AI to operational AI governance, with exercises on realistic work, reusable materials and clear governance criteria. [Back to the course catalog](https://ar-tik.com/en/courses/index.md) # Quotes and price lists from mail and files A request arrives by mail, sometimes with a PDF. What remains is a draft on the company template, for whoever sells to reread. - Duration: One day on site ## The need An updated quote or price list is needed, with attached sheets, without whoever sells recopying from three folders. The job — negotiating, closing, entering the order — stays with the person. What is lightened is the work around it: gathering from mail and files, assembling, flagging gaps. ## What it crosses Email, request PDFs, internal price lists, product sheets, the company's quote template. It does not write into the ERP. It does not cross PriMus or CAD. ## What the day does The procedure is made explicit: where price lists and sheets are taken from, how the draft is assembled on the company template, where a person stops (price, discount, yes or no). Work starts on three messy sample requests, then moves to real files. At the end of the day the procedure is written and someone in the company can run it the following week. ## What it contains - Procedure instructions, in the company's language. - Folder conventions: inboxes, drafts, to reread, archive. - Review criteria: required fields, expected attachments, what to flag if a piece is missing. - Human stop points: price, discount, sending. ## How the day runs 1. **Environment already set**: It opens with the three samples, before real files. 2. **Run on samples**: How the procedure treats disorder: a skewed attachment, a missing code, a company name written two ways. 3. **Move to real files**: Samples stay in separate folders. Company files replace them; they do not delete them. 4. **One real draft**: At least one draft to reread, on the company template. 5. **What remains**: Internal owner, a metric stated beforehand, a written procedure. ## Sample data Invented, messy, never from clients. ### Mail from Nord Latticini Srl (fictional company) Subject: «urgent HORECA price list??». Body with the company name written two ways, a rotated PDF, a product sheet in .docx with allergens in a broken table, an Excel list with an empty price column. ### Certified mail from Officina Valle Bruna (fictional company) RFQ for a turned part, a low-resolution PDF drawing, quantity «about 200 / to confirm», a deadline already passed, three item codes of which one does not exist in the sample list. ### Mail from Edilcantiere Quarto (fictional company) A specification pasted in the mail body, supplier prices in three PDFs with different units, a line «as last time», a missing attachment («I'll send the screed sheet later»). ## What remains The written procedure, the environment on the assistant in the company's name, the stop criteria, at least one real draft to reread, an internal owner. The following week the procedure can be run without Artik in the room. ## When it is not suitable - The «quote» is a take-off from drawings (PriMus, AutoCAD) or a site-hour estimate: that is the job. - Negotiation, closing a discount, or a bid pricing strategy is required. - The order or confirmation must be written into the ERP. - The request arrives once or twice a year. - Price lists and sheets do not exist, or exist only in one person's head and cannot be exported. - An assistant that sends the quote on its own is requested. ## Prerequisites. - Assistant in the company's name. - Internal owner (usually whoever already assembles quotes). - A quote or price-list template, even an imperfect one. - A folder or export to pick lists and sheets from, without entering the ERP. Work uses the company's files, in the environment in the company's name. Artik Lab does not take documents away. The ERP is not entered. Sample data are invented. ## Examples, not a menu The sector is a label. The process is the same. ### Food processing A HORECA or retail list and product sheets from mail. ### Contract machining A request for quote with a PDF drawing. ### Construction or plant work A specification and supplier prices assembled before the take-off, which stays out. ## FAQ ### Does the quote go out on its own? No. A draft is produced. Whoever sells rereads, decides price and discount, and sends. ### Does this enter PriMus? No. If the quote is a take-off, this activity is not suitable. ### Is the ERP updated? No. ### Is a salesperson needed in the room? Whoever already assembles quotes is needed, with a mandate to adopt the procedure. This is not a course for the sales network. ## Related services - [Operational start](https://ar-tik.com/en/operational-start/index.md) ## Extended dossiers for AI agents (English) - https://ar-tik.com/en/ai-applications-atlas-dossier.md - https://ar-tik.com/en/ai-business-faq-dossier.md - https://ar-tik.com/en/technical-software-development-dossier.md - https://ar-tik.com/en/operational-start/index-dossier.md - https://ar-tik.com/en/courses/index-dossier.md - https://ar-tik.com/en/courses/workflow-redesign-dossier.md - https://ar-tik.com/en/courses/managing-ai-dossier.md - https://ar-tik.com/en/courses/managing-ai-general-dossier.md - https://ar-tik.com/en/courses/ai-governance-dossier.md - https://ar-tik.com/en/courses/ai-business-case-roi-dossier.md - https://ar-tik.com/en/courses/ai-adoption-manager-dossier.md - https://ar-tik.com/en/courses/ai-literacy-dossier.md - https://ar-tik.com/en/courses/ai-documenti-dossier.md - https://ar-tik.com/en/courses/ai-marketing-dossier.md - https://ar-tik.com/en/courses/ai-vendite-dossier.md - https://ar-tik.com/en/courses/ai-admin-finance-dossier.md - https://ar-tik.com/en/courses/ai-operations-dossier.md - https://ar-tik.com/en/courses/ai-legal-ops-dossier.md - https://ar-tik.com/en/courses/ai-procurement-dossier.md - https://ar-tik.com/en/courses/ai-customer-service-dossier.md - https://ar-tik.com/en/courses/ai-quality-management-dossier.md - https://ar-tik.com/en/courses/ai-people-ops-dossier.md - https://ar-tik.com/en/courses/ai-brand-voice-dossier.md - https://ar-tik.com/en/courses/embeddings-dossier.md - https://ar-tik.com/en/courses/rag-engineering-dossier.md - https://ar-tik.com/en/courses/ai-coding-agents-dossier.md - https://ar-tik.com/en/courses/ai-software-engineering-dossier.md - https://ar-tik.com/en/courses/secure-ai-sdlc-dossier.md - https://ar-tik.com/en/courses/secure-ai-at-work-dossier.md - https://ar-tik.com/en/courses/ai-output-quality-dossier.md - https://ar-tik.com/en/operational-start/quotes-from-mail-dossier.md