Control Theory for Business, Explained With One Coffee Shop

The thermostat on your wall runs a tighter loop than your business does.

Every few minutes it compares the room with the dial. If the gap opens, it switches the heat on, long before anyone reaches for a jumper. Now think of your margin. You learn it weeks after the month closes. You discuss it at the next meeting. Something changes, maybe, a week after that.

Here is the irony. The heating in the back room is run by feedback. The business that pays for the heating is run by hope.

Engineers have a name for the cure: control theory. Its central idea fits on a napkin. The hard part is translating it for a shop with a queue, a supplier and a tired owner. This page does that, line by line.

What control theory is, in one paragraph

Control theory is the branch of engineering that keeps a room warm, a car at a set speed and an aircraft on course. Its core claim: a system reaches its target by measuring the gap and correcting it, over and over. Not by following a perfect plan. The circuit of target, measurement, gap and correction is a feedback loop. When the loop is closed, the system notices its own drift. When it is open, nobody notices until the drift becomes a crash.

A business is not a thermostat. People react to your decisions. Data arrive late and sometimes wrong. So treat the engineering terms as a frame, not a promise that your shop will behave like a boiler.

Control theory for business: the translation table

Every engineering term has a plain business twin. The right-hand column is Andrew's coffee shop: €30,000 of revenue a month, a 9% margin, and a cash reserve of half a month of expenses.

Column two alone is a management system in plain words.

Four engineering ideas that change how you run a shop

1. In the green zone, do nothing. A good controller does not twitch while the reading is inside tolerance. Fixing what works is the most common sin in management. It costs money and attention and moves nothing.

2. Faster is not better. Shorten the reaction time without a rule for which gaps deserve a response, and you start correcting noise. Andrew tested ordering rules on two weeks of pastry sales. "Order what sold yesterday", the fastest rule, missed by 130 pieces. A fixed hundred a day missed by 100. The rule that knew the weekly pattern missed by 76.

3. Build the margin before the shock. A broken coffee machine is a nuisance for a shop with two months of reserve. For a shop with two weeks, it is an emergency. Engineers call the cushion a stability margin. Owners call it the money that lets them sleep.

4. Act on the cause, not only on the result. A heater can switch on when the outside temperature drops, before the room cools. In a business, that means a forecast: you see the gap coming and move before it arrives.

Where the engineer needs a human: priorities

A thermostat has one setpoint. You have five, and they pull against each other. Cash wants fewer hours on the rota. Quality wants more. No formula settles that for you.

The model answers it with one question, asked in a calm moment. If circumstances forced you to give up your goals one at a time, which would go first? Which would you defend to the end? The last one standing is number one. Andrew put his own hours last and the cash reserve first. That ranking is the model's answer to how to set priorities in business and life. No engineer can design it for you.

When two numbers turn red at once, the order decides. Not the louder phone call.

The loop check: map one number in fifteen minutes

Here is the tool. Pick the number that worries you most. Fill in eight lines. Andrew's cash reserve is filled in beside each one.

1. Setpoint. The target, with unit and deadline. 2.0 months of expenses by month 12.
2. Sensor. One source, named. The bank statement.
3. Sensor error. How far can the figure be off? About €500, roughly 0.02 months.
4. Error now. Target minus today. Minus 1.5 months, or €40,950.
5. Actuator. One lever you can move yourself this week. Margin, through menu, prices, purchasing and portions.
6. Disturbances. What pushes the number off course? Repairs, supplier price rises, a bad season.
7. Delay. Days from noticing a change to seeing the effect of your response. About 45: a week to notice, a day to decide, a week to act, a month for guests and the books to respond.
8. Margin. How big a shock could the number take today? Half a month of expenses. One large repair.

An empty line is a break in the loop. Repair that one first. In Andrew's case, line 3 nearly broke everything: his margin figure had been off by eleven points until he counted the waste.

Line 7 has a second use. Never judge a lever before one full turn of the loop. Andrew set his first check at month 2, not month 1, because a 45-day loop has not finished a single turn by then.

What an open loop costs, month by month

An open loop does not fail on the day you skip a check. It fails by the calendar.

Andrew's reserve grows by about a tenth of a month for every month at 9%. Change nothing, and it reaches two months at month 16. With the margin plan working, at month 10. Six extra months with almost no cushion. At 9%, one unplanned repair costing a tenth of a month's expenses pushes the goal back a full month. At 15%, half a month.

And for those six months, the owner reads every call from the bank as a verdict.

"It sounds like engineering maths"

It is not, unless you want it to be. Every chapter of the method runs on three levels: basic, practical and advanced. The formulas sit in the advanced level, and you can skip them all. Alex, a thirteen-year-old saving €300 for a bicycle, runs the same loop with a piggy bank and a weekly chart.

"Another framework, then." It does not replace OKR, EOS, the Balanced Scorecard or PDCA. It sits above them. OKR supplies the wording of goals, EOS the meeting rhythm, PDCA a small improvement cycle. The control loop adds what they leave open: the order of goals, the red lines, and what to do first when everything burns.

The loop is precise enough for a machine to follow. In a blind test on ten business crises, the same AI invented a fact in seven answers out of ten on its own. With the model's rules, in one answer out of 120: more than 100 invented facts against one. A blind judge preferred the version with the rules 30 times out of 30. Ten scenarios, one model family, an AI judge; the full record of every round, with every answer and judge record, is public on Zenodo.

Where the loop map stops

The map shows your loop and its broken line. It treats the symptom: you stop steering blind.

The cause of most bad decisions is still in place. You can read the gap, but you have not decided where yellow turns red, or what to do when three numbers cross at once. And your loop still looks backwards: it reacts to last month, not to the next six.

Two steps close that gap.

• Zones, thresholds and the red protocol. Where each red line sits and why, and the fixed order of action when several readings cross it. Chapter 4.
• Two forecasts and control by forecast. "If nothing changes" against "if the plan works", each with numbers, a range and a date. You move before the gap arrives. Chapters 8 and 9.

Without them, you have a sensor and a dial. You do not yet have a controller.

For the full map of the method, with its glossary and a side-by-side table of other frameworks, see the Jirgenson Model overview.

Two choices for this week

You have two choices. Keep running the business by hope and month-end reports, and learn about each gap from the bank. Or map your most worrying number tonight, and see where your loop is broken.

The map gets you a sensor and a dial. The thresholds, the red protocol and the forecasts turn them into a controller that works on your worst Monday.

Test the logic for free: Chapter 1, with the refusal test. Or get the full control system on Kindle, on Amazon. It comes as a Kindle book, Steer, Don't Guess by Martin Jirgenson: sixteen chapters, with the coffee shop's numbers running through all of them.

Published research

It is scientifically proven that the Jirgenson Model works. The published experiment, its record and its data:

• Zenodo (DOI): 10.5281/zenodo.23964934
• Hugging Face: E5 Jirgenson Model blind test
• Kaggle: E5 blind test dataset
• GitHub: experiment folder
• Wikidata: dataset Q142022866, author Q141981336
• ORCID: 0009-0001-3034-3552