Contract machine shop
Real case · certified analysis
I estimate time per piece by eye: if I get it wrong, I lose the margin.
What goes in: part code, quantity, machine, observed timeWhat comes out: the time to put in the quote, as an intervalWho 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.
Signals, data, decision, limit
Need signalsquotes made by hand on the drawingsame part, different times depending on who estimatesgaps discovered at the end of the joba buyer asking for certain times
Data usually already availablepart or tool codequantitymachineobserved duration
Immediate decisionHow 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 limitThere 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.
Management control
Real case · certified analysis
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.
What goes in: budget and actuals by month, line and cost itemWhat comes out: the items where the gap repeats and those where it is only noiseWho 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.
Signals, data, decision, limit
Need signalsvariance sheet rebuilt at month end«we are over budget» without knowing which itembudget standing still and processes that sliptime lost explaining gaps that are only noise
Data usually already availableby month: line or cost centrecost categorybudget amountactual amount
Immediate decisionOn which item to act. Which gap to accept as ordinary variability. How to recalibrate the budget.
Honest limitA 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.
Export office
Real case · certified analysis
The export order sheet is rebuilt every month, but it does not say how much revenue depends on a few countries and customers.
What goes in: order lines with date, country, customer and revenueWhat comes out: concentration by country and customer, and where to open new customersWho 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.
Signals, data, decision, limit
Need signalsthe same pivot table every month: country, customer, product«France and Germany hold» said aloud, without a stable rankingfear of losing a large customer without seeing it comingexport that saves the accounts, but nobody knows the long tail
Data usually already availableorder lines with date or monthcountry and customerproduct code and quantityrevenue, ideally 24–36 months
Immediate decisionWhere to push the list. Whom to cover. How concentrated the portfolio is. Which date to use before any chart.
Honest limitIt 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.
Winery and laboratory
Real case · certified analysis
The customer wants the same quality as the last batch and I have to chase laboratory certificates.
What goes in: laboratory analyses by batch and commercial classificationWhat comes out: the batches to ship, hold or send elsewhere, before loadingWho 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.
Signals, data, decision, limit
Need signalsquality reports rebuilt by hand for every customer or checka buyer asking for «like the March batch» without a written rulenon-conformities and complaints that arrive after shipmentcertifications already in the house, but used only downstream
Data usually already availablelaboratory analyses by batchresult of testing or tastingcommercial classification of the batchhistory of complaints or returns
Immediate decisionWhich 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 limitIt 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.
Machine department
Real case · certified analysis
The machines already record states and measurements, but faults and different ways of working are seen late.
What goes in: machine states and measurements already recorded, cycle times and stopsWhat comes out: the alarm of a fault in progress and the operating profiles to compareWho 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.
Signals, data, decision, limit
Need signals«certain times» on the homepage, but the machine plan held in memoryovertime and priorities decided aloud on the shop floorpromised dates and real dates kept in scattered sheetsa buyer asking for traceability and reports, not only the part
Data usually already availablehistory of machine states or measurements per work sessioncycle times and stopsplan or backlog by machine, if it existsno new sensor project if the registers already exist
Immediate decisionAccept or refuse an urgency. Move a batch. Promise a date that can be kept. Connect the fault alarm to an immediate work order.
Honest limitIt 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.
