The Handover Document Is Not the Handover

Her last day was a Friday and the document was fourteen pages long.

I want to be fair to that document, because it was good. Logins, migrated. Vendor contacts, with the direct lines rather than the switchboard. A numbered process for the monthly close with screenshots. She had even written down which of the two shared drives was the real one, which is more institutional honesty than most companies manage in a decade.

We printed it. Somebody said "she's a legend" and meant it. There was cake.

The first thing broke on the following Tuesday, and it was not in the document, and neither was the second thing, or the thing in September. Six weeks later I was standing in front of a supplier portal at 7:40 in the morning trying to work out why we always, always submitted the order on Wednesday when the deadline was clearly Friday, and there was nobody left in the building who knew.

What was actually in her head¶

The Wednesday thing turned out to be real, and it took a former colleague and two phone calls to recover.

The supplier's Friday deadline was true on paper. But their warehouse cut the picking list on Thursday morning, and anything submitted Thursday afternoon technically met the deadline and shipped a week later — a fact that appears in no contract, no portal, no email. She had learned it in 2022 by getting burned once, and had quietly submitted on Wednesdays ever since, and it never occurred to her to write it down because to her it wasn't a rule. It was just when you do the order.

That's the whole problem in one sentence. The most valuable thing she knew had stopped feeling like knowledge to her. It had become a habit, and habits are invisible from the inside. You cannot document what you no longer notice you are doing.

“

I shall reconsider human knowledge by starting from the fact that we can know more than we can tell.

”

Polanyi's example is a face: you can recognise one out of thousands and cannot describe how. The recognition is real, verifiable, instantly demonstrable — and it does not decompose into sentences. He called it tacit knowing, and the ordinary business version is that your best operator can tell in four seconds that a job is going to go wrong, and cannot tell you why, and is right.

You cannot get that in a handover. Not because they're withholding it. Because asking for it in written form is asking for a different thing than the thing they have.

The three layers, and which one your document caught¶

Explicit: logins, contacts, file locations95% captured

Procedural: the documented steps, in order70% captured

Tacit: why the steps are in that order10% captured

Those bars are my estimate from watching this happen, not a study, and I'd rather say so than dress it up with a fake citation. But the shape is not controversial to anyone who has run an operation through a departure.

The explicit layer is easy and everyone remembers to do it. The procedural layer is tedious and most people do a decent job under deadline. The third layer is the one that was doing the work, and it's the one nobody captures, because capturing it requires the departing person to notice which of their habits are load-bearing — and by definition they've stopped noticing.

The turn: you already have this problem¶

It's easy to read all this as a story about someone else's departure. Here's the version that should actually bother you.

Sit down and ask which single person leaving would cost you the most next month. Not "who is most valued" — who leaves the largest hole. In most small operations the honest answer is either the founder or one long-tenured operator, and in a lot of them it's a person whose job title badly understates what they hold.

Now the harder question, and the one that separates a worry from a plan: what specifically breaks, and on which day?

If you can't answer that, you don't have a staffing risk, you have an unmeasured one. And unmeasured risk is the category that shows up in the reasons small businesses actually fail wearing a different name — usually "cash flow," occasionally "bad luck," rarely "the one person who knew how the ordering worked left in March."

The org chart will not show you this either. Formal structure maps reporting, and tacit knowledge accumulates wherever the work actually happened, which is frequently three levels below where the diagram suggests it should be.

What actually transfers knowledge¶

The uncomfortable finding, from having got this wrong more than once: the document is not the intervention. Time is. Every genuinely successful handover I have seen involved two people doing the job at the same time for a while, and every failed one involved a document and a farewell.

1. 1

   Overlap, even when it feels expensive

   Two weeks of both people doing the job is the single highest-return thing available, and it is the first thing cut because it looks like paying twice for one role. You are not paying twice for a role. You are paying once for a transfer, and the alternative price — six weeks of a new person rediscovering the Wednesday rule by getting burned the same way — is higher and lands later, which is the only reason it feels cheaper.

2. 2

   Have them work from their own document, before they leave

   This is the trick that finds the gaps, and it costs nothing. In the last week, the departing person stops doing the job from memory and does it only from what they wrote. Every time they reach past the document, that's a gap — and it surfaces while the person who can fill it is still in the building. You will be startled how often they reach.

3. 3

   Write down reasons, not just steps

   "Submit the order Wednesday" is a step and it will be deleted by an efficiency-minded successor within a year. "Submit the order Wednesday because their warehouse cuts the picking list Thursday morning, learned the hard way in 2022" is a fence with a sign on it. Steps decay. Reasons defend themselves.

4. 4

   Make the new person ask about every weird step, and forbid deleting one for ninety days

   A blanket moratorium beats case-by-case judgement here, because the whole difficulty is that the dangerous steps look exactly like the useless ones. Ninety days is long enough to hit most monthly and quarterly cycles, which is when the fences reveal what they were for.

5. 5

   Run a bus-factor audit once a year, in writing, and be specific

   Not "is anyone irreplaceable." Ask: if this person vanished on the 1st, name the thing that fails by the 15th. If nobody can name it, the audit isn't finished — a vague answer means the knowledge is tacit, which is precisely the finding you were looking for. Treat what you find as a depreciating asset with an unpublished schedule, because that's what it is.

Frequently asked questions¶

The portal, at 7:40 in the morning¶

We put the Wednesday rule in the document eventually. One line, with the reason attached, and a date.

What I think about is how close we came to never recovering it. It survived because one person happened to still take our calls, and because I happened to be irritated enough at 7:40 in the morning to ask why instead of just changing it to Friday and finding out in eleven days.

That's not a process. That's luck wearing a process costume, and I'd rather not run an operation on it again.

So write the document — genuinely, write it, the fourteen pages are not wasted. Then assume it's the smallest part of what you're losing, and buy the overlap, and make the new person ask about every strange step before they tidy it away. Somebody built that fence standing right where you are, after a week that went badly, and they are not here to explain it.

It's just business — and the part that mattered was never written down.

Sources

1. Michael Polanyi — The Tacit Dimension (1966), University of Chicago Press
2. Polanyi's paradox — overview of the 'we can know more than we can tell' principle and its modern applications
3. G.K. Chesterton — The Thing (1929), the essay containing the fence principle
4. U.S. Bureau of Labor Statistics — Job Openings and Labor Turnover Survey (quits and separations data)