The Assay · 7 October 2026

The Ten-Day Test

A system built to run without you can still be waiting for you. The way to find out is to measure what stops when you are not there.

For three Tuesdays in September, a check I had built to run on its own, an AI agent on a weekly schedule, reported that it had succeeded.

It had not. Each run lasted about seven seconds and wrote nothing. As far as I can tell, it reached a step that wanted my approval, asked for it, and waited — and nobody was there to answer, because the whole point of the thing was that I would not need to be. One of its jobs was to raise anything overdue. By the time I looked, four items were overdue, and it had raised none of them.

I found out when I went to move the job elsewhere and looked at what it had actually produced. Nothing, three times, with “succeeded” beside each.

I had automated the work, not the say-so

The check could do everything the job involved except one thing: decide that it was allowed to. That part had always been me. No earlier version of the job had run without me in the room, so the question had never come up.

So the dependence did not leave when the work was automated. It moved into the system, and it arrived with a success message attached.

In last week’s issue of The Assay, No12, I argued that anything which sends, pays or deletes should happen only on a person’s confirmation. This is the other half of the same design. The steps that do not need you have to be told so, in advance and in writing, or they will wait for you as well. An automated step has no more authority than the permission written into it.

The fix was not a better model. I moved the job, rewrote each of its steps as an exact instruction, and wrote into the system’s settings a short list of exactly which of those it may run without asking me. The first run afterwards took about a minute and produced a full report.

What absence measures

Ask an owner how much the business depends on them and you get an opinion. It may be modest or it may be grand. It is almost never tested.

Three studies I read in September did not ask anyone. They measured what happened when the person was gone.

In Denmark, researchers followed firms of twenty or more staff whose chief executive spent time in hospital. In their 2011 working paper, one to four days made no meaningful difference. Ten days or more went with operating returns about 12% lower. When other senior managers were in hospital for the same time, the researchers found no effect they could distinguish from zero.

In Norway, a study published in 2022 followed 1,459 young firms whose founder had died. Sales fell by about 60% and did not come back, and the damage came mainly from founders who worked in the business. Where the founder had also been running the firm, the loss was about half as large again as when a hired chief executive died.

In the United States, a 2019 study of firms owned by top earners found that profits fell by around four-fifths when the owner died before 65. Where the owner died at 75 or older, the effect on profit was close to zero. The authors note that owners of that age were less likely to be running the firm day to day.

Most owners ask one question: is my business too dependent on the owner? Put the three studies side by side and it gets a more useful answer than yes or no. My reading of the three together: the business depends on what you do more than on what you own. And what you do can be moved, one piece at a time. It does not move on its own.

The ten-day test

None of these studies tested a planned absence, and the idea of testing a business by leaving it is not mine. In his book Clockwork, Mike Michalowicz argues for a four-week holiday, taken, as he puts it, “completely unplugged from the business so that you can see if the business will survive without you.” What I would add is a shorter version with a control, and one more question: what your automated systems are waiting for. The ten days are my test, not the researchers’. The Danish figure is simply where I set the bar.

Pick ten working days. Before they start, write down what you expect will wait for you. Then stay out: no calls to check in, no approvals from the car park. When you come back, compare your list to what actually waited.

Two kinds of thing show up. Some waited because only you knew them: which supplier to ring when the usual one is late, why one customer is on different terms. Others waited because only you were allowed to decide them, although someone else knew perfectly well what to do.

The first kind is familiar, and it is slow work to write down. The second kind is the one to sort out before any AI goes near the business, because it is what my Tuesday check ran into. Nobody had written down that it could go ahead, so it waited. A person in that position at least rings you. An automated step may simply report that it has finished.

The second kind is also the quicker of the two to move. It usually takes one sentence, written where the people who need it, and the systems, will find it: this kind of thing, up to this amount, goes ahead.

The Norwegian study found that sudden deaths did considerably more damage than expected ones. The authors had expected that people prepare when they can, though they are careful to add that it may be the turmoil of the surprise, or that the firms were different to begin with. The ten days are the absence you choose, which makes them the one you can prepare for and learn from.

In fairness

Short absences made no measurable difference in the Danish data, and two years after a long one the effect could no longer be detected. In Norway the damage was smaller in older firms — smaller, not gone. A few days away will not break most businesses. That is not the question the ten days answer.

The question is what they show you — and whether anything that reported success while you were gone actually did the work.

A business, or a system, runs without you only once it has been seen to.

If you run the ten days and would like someone to read the list with you, you know where to find me.

The Assay is written by Rudi Prenzlin at Scalable Growth, a UK practice that gives owner-managed businesses a fractional CFO and the operational structure to run without every decision going through the owner.

← All notes Begin a conversation