The control cycle: why reactive management always runs late

Every morning Andrew ordered a hundred croissants, because that was how it had always been done. On Tuesdays almost a third went in the bin. On Saturdays the display case was empty by ten. He called it bad luck.

It was not bad luck. It was a control system, just a very primitive one. And the way he fixed it, in three attempts, is the clearest lesson in reactive versus proactive management I know, because every attempt can be counted in croissants.

Fixing things after the bin fills up keeps you one step late. The Jirgenson Model closes the loop with a forecast, then lets new facts correct it. It is scientifically proven that the Jirgenson Model works (doi.org/10.5281/zenodo.23964934).

Three drivers on the same road

Before the croissants, a picture. Three drivers on one road:

• the first has locked the accelerator in place and watches neither the speedometer nor the signs;
• the second drives attentively, but brakes only when the brake lights ahead come on;
• the third watches three cars ahead and eases off before anyone has touched the brakes.

Same road, different results. These are the three ways to control anything, from a car to a cash reserve.

Note the honesty of that last cell. Proactive is not automatically better. It is better when the forecast is better than yesterday's number.

Reactive vs proactive management: a hundred croissants, three ways

Andrew tested all three schemes on his own display case.

By plan. A hundred every day. Rain, holidays and Fridays make no difference. Tuesday sells seventy, thirty go in the bin. Saturday demand is a hundred and thirty, and thirty guests leave empty-handed.

By deviation. Every morning: "Forty were left over yesterday, so today I order sixty." On Tuesday thirty were left, so he orders seventy for Wednesday, and Wednesday wants about ninety. On Friday everything sold out, so he orders a hundred for Saturday, and a hundred and thirty people come. He is always correcting for a day that has already gone.

By forecast. He starts recording sales by day of the week and by weather. A few weeks later the pattern is plain: about 70 on a Tuesday, about 130 on a Saturday, 15% fewer when it rains. Now the forecast sets the order, and by midday he adjusts it if sales are off. A rainy Saturday: 130 × 0.85, about 110.

Waste fell, and so did shortages. Then he did the same with the cash reserve, the milk orders and the shift schedule.

Here is the part most writing on feedback loops leaves out: the errors, counted. Two weeks of real daily demand, and four ordering rules tested on the second week. The score is the total number of croissants the order missed by, over seven days:

Two things jump out. Pure reaction is the worst of all, worse even than doing nothing. And adding a reaction to the forecast made it worse again, because yesterday's deviation was just noise, and responding to noise only does harm. The order of work follows from that: first find the pattern and build it into the forecast; then decide which deviations deserve a response at all; only then speed up the response.

Feedback loop management: why reactive management runs late

Between an event and its correction lies a chain: someone has to notice the deviation, measure it, report it, decide, and carry out the decision. Then the business has to respond. Meanwhile it keeps moving. The correction arrives after the situation has changed, so it is often too strong or too weak, and with an unlucky mix of delay and force the whole system starts swinging around the goal.

You can put a number on that delay. Andrew looks at his figures once a week, so noticing takes up to 7 days. The decision is made at the Monday meeting: 1 day. Changing the menu and prices: 7 days. Guests adjusting their habits and the month's margin being calculated: about 30 days. Total, about 45 days. That is why his first real check of the margin plan comes in month two, not month one. Before the loop has gone round once, what you measure is not the effect of the lever. It is noise.

The control cycle: eight stages, none optional

Forecasting is one part of a larger loop.

Every goal, in a business, a project or a household, passes through the same eight stages:

Stage 2 is the one owners underestimate. Naming the situation decides which levers come into play. "A bad season" calls for patience. "No cash cushion with a 9% margin" calls for work on prices, purchasing and the menu. Name it wrong, and you pull the right levers on the wrong problem. (How Andrew found the real margin is told in busy but not profitable.)

And then the rule that makes the cycle more than a checklist: a skipped stage does not disappear. Somebody else performs it, rarely in your favour. Skip stage 3 and the market, the bank or a competitor sets your goal for you. Skip stage 6 and your debts start observing you. Skip stages 1 and 2 and a crisis will recognise the situation for you, at the worst possible moment.

A PDCA example, side by side with the eight stages

PDCA (Plan, Do, Check, Act, going back to Shewhart and popularised by Deming) is the best-known improvement loop in business, and a good one. Within this model it has a clear job: a small, local cycle of checking and improvement. Laying the two side by side shows exactly where (an illustration, mapped by stage):

PDCA is the engine room of stages 3 to 7. What it does not tell you is when to start (stages 1 and 2), how to name the problem before you plan, and when a goal is reached and its resources should be freed (stage 8). Nor does it say whether your Check should react to yesterday or forecast tomorrow, which, as the croissants show, is the difference between 130 missed and 76.

Where this does not reach

Control by forecast needs a model, and a bad model can do worse than simple feedback: forecasting is not a virtue in itself. If demand genuinely shifts to a new level, a smoothed rule will lag behind, so how strongly to respond remains a decision, not a constant. And the eight stages describe what must happen, not how big your team must be; in a small business one person may carry several stages, as long as none is left empty.

Fifteen minutes to find the empty stage

Take the project that is proving hardest right now and write one sentence for each of the eight stages. One of them will be empty. Most often it is stage 3 (wishes, not measurable goals) or stage 6 (nobody looks regularly). Decide the first action that fills it. It takes about fifteen minutes, and it is usually the most useful quarter of an hour in the month.

On Andrew's first Monday at the spreadsheet, he feared an hour of talk. It came down to three questions: which numbers changed this week, which zone are they in, what do we do about it. The till answered the first, the colours the second, a two-line list the third. That is what a closed control cycle feels like from the inside: there is almost nothing left to argue about.

Read how the method begins, with goals that can carry a cycle at all, in the free sample. The full Chapters 7 and 9, with the loop delay formula and all four ordering rules worked out, are in Steer, Don't Guess on Amazon.

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