The Assay · 26 August 2026

Everybody Knows the Process

A procedure tells you what happens on an ordinary day. It says nothing about the day that is not ordinary.

Take an ordinary fortnight in which you are not there. Not a crisis — a week in Cornwall with poor signal, then three days on a customer site four hundred miles away. The orders arrive. The team works. Invoices go out, the vans load, nothing burns down.

Then a customer asks for a price that sits outside the usual band. A supplier misses a delivery that three jobs depend on. A long-standing account complains about something that is arguably their own fault. And the business slows to a stop. Not because nobody knows the process but because everybody knows the process. And this is not it.

The procedure describes the ordinary day

A business can have a written procedure for almost everything and still run entirely through one person. The two facts sit together comfortably, which is why the problem survives undetected for so long.

A procedure is a description of the ordinary case, written by somebody who has already decided what counts as ordinary. It tells you the sequence. What it does not carry is the rule for the moment the sequence stops applying — and that moment is the only one anybody ever needs to ask about.

Look at what consumes a week in accounts payable. The written process is four steps: invoice in, coded, approved, paid. The week is not four steps. The week is a supplier quoting the wrong purchase order number, a scanned invoice the system cannot read, a total that differs from the order by four pence and somebody deciding whether four pence is worth an email. None of that is in the procedure. All of it is the job.

Three things, not one

What sits in the owner’s head is usually treated as a single substance. It is three, and they have nothing in common except their location.

The first is repeatable procedure — the sequence that does not vary. It should be written, and it should belong to the team rather than to you.

The second is the standard: what good enough looks like here, for this customer, at this price, on this lead time. Almost nobody writes this down, because from the inside it does not feel like knowledge. It feels like taste.

The third is live judgement — the calls that are not obvious, where the right answer depends on things nobody could have listed in advance. This layer is real, and in my experience much smaller than owners expect. It is the only one of the three that should stay with a person by design, and the only one no AI should be pointed at.

I wrote last week about the record that never got made — the fact that the way a business decides has no ledger, and lives in one head by default. This is the next question, and the more useful one. The record is not one thing. What I called instinct last week is the third of these, and only the third; until you know which of the three you are holding, you cannot tell whether the answer is a document, a delegation, or a name.

Where I got this wrong myself

For years I treated the decision to publish something as judgement. It felt like judgement — you read the draft, you know.

Then I went to hand part of my own content production to software, and software cannot be told “you know”. So I had to write down what I meant. Most of it turned out to be a standard I had simply never stated: opens on an observation, no exclamation marks, one invitation and no push, and eleven other things I could have written down at any point in the previous decade. What was left after that was four or five real calls a year. The embarrassing part was not the number. It was that I had been paying judgement’s price for a checklist’s work, and had never once looked.

The question that finds the work

There is a version of this exercise that consumes a year and produces a binder. Avoid it. The whole business does not need mapping; three decisions do.

Which three decisions would cause the most damage — in cost, in delay, or in what a customer thinks of you afterwards — if the person who normally makes them were unavailable for a month?

Ask it out loud in a room and, in my experience, the answers come quickly. They are specific, and rarely the ones on the organisation chart. What you get is not a strategy — it is a list of three places where only one person can answer, ranked by what it costs when that person is on a plane.

What an assay actually is

The word comes from metal. You do not assay a rock to establish what it is supposed to contain — the seam already has a claim attached, and the claim is generally optimistic. You assay it to find out what is in it, and whether it is worth extracting.

Applied to a business, the test is the same and so is the discipline. Not what the manual says happens. But what actually happens: which decisions run through people, which of those can be stated plainly enough for somebody else to make them, and which need a named owner, the information to decide with, and a clear point at which they come back to you anyway. That test is the method I run, and it is where the name Assay-OS comes from.

The order matters more than the name. You cannot hand a decision to AI if you cannot state it, and you should not delegate one whose standard has never been written down. Both attempts fail the same way — quietly, and later than anyone connects to the decision.

Why this is worth doing before anything else

The immediate return has nothing to do with technology. A business where the standard is written and the escalation route is known does not stall when somebody is away. Illness stops being a crisis. A key person leaving for a competitor becomes an inconvenience rather than an event. The work you currently turn down because only one person can price it becomes work you can take, and the person who normally prices it gets a fortnight off without a phone.

None of this is a new idea. Businesses have been documenting themselves for a century, mostly badly, and a filing cabinet nobody reads has earned the scepticism it gets. What has changed is the consequence. A person who meets an undescribed exception asks somebody. A system cannot: it will either handle it wrongly and with total confidence, or push it into a queue with nobody’s name on it. Both failures are the same failure, and neither is the model’s fault.

So if AI arrives in your business later, it arrives into something that can tell it what good looks like and where to stop. If it never arrives, you have a business that keeps its promises while you are in Cornwall. I have yet to meet an owner who treated that as a consolation prize.

Start with one

Pick a single decision that currently rests on one person. Start with what it is for — the outcome it exists to produce, stated without reference to how it is done today. Then write down four things: what good looks like, what information you need in order to decide, who is allowed to act without asking, and the point at which it has to come to you regardless.

It is a morning’s work, not a project. What comes back tells you which of the three layers you were missing — or, sometimes, that the procedure grew around the person rather than the outcome, in which case documenting it faithfully would be the worst thing you could do. That is the difference between assaying a business and documenting one: the first tells you what is there, the second assumes it belongs.

A business does not become less dependent on you when somebody writes down the procedure. It becomes less dependent on you when somebody else can make the right call — and say why.

The first three are yours; they need nobody’s permission and nothing bought. If the ones after that turn out to be the hard ones, that is usually where I get called in, and it will keep.

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