When someone leaves, what actually walks out the door
Every company runs on a few heads that were never written down. Here is how to find out which ones, before the resignation letter tells you.
Ask your office a hard question today. Any hard question.
Why do the odd accounts renew. Which vendors actually bend, and how far. What plan B was, the last time something went sideways at the worst possible hour.
Then count how many answers come back as a name.
"I don't know, John knows that."
That sentence is the whole problem, said out loud. It is also the cheapest diagnostic any company has, and almost nobody runs it on purpose.
The names are the map
Most organizations have a mental model of themselves that looks like an org chart. The real map looks nothing like it. It is a much smaller diagram, maybe eleven people in a company of two hundred, and every hard question in the building routes to one of them.
Those eleven are not always senior. They are frequently not managers. The most load-bearing person in a manufacturing business is often a scheduler nobody has promoted in nine years, because she is the only one who knows which of the three suppliers will actually take a Friday call.
You can find them in an afternoon. Sit with the questions your team escalates and mark who they get escalated to. The concentration will be worse than you expect.
What the handoff actually captures
When one of those eleven resigns, a process starts. It has a name, usually "knowledge transfer", and it usually consists of two weeks, a shared document, and a series of meetings with whoever is inheriting the work.
Here is what that process reliably captures: the systems the person logs into, the recurring meetings they attend, the reports they produce, the files they own, and the names of the external contacts they deal with.
Here is what it reliably misses: why they stopped trusting one of those contacts in 2021, which of the recurring meetings is theatre, what they do when the report looks wrong, and the four exceptions to the process that are not written in the process because the process would be unreadable if they were.
The handoff document is not a lie. It is an inventory. Inventories are useful. They are just not the thing you are losing.
Three kinds of knowledge, and only one gets written down
It helps to separate them, because organizations spend almost all of their effort on the first one.
The procedure. The steps, in order. This is what an SOP is for, and SOPs are genuinely good at it. If your business has documented procedures, this part is probably fine.
The context. Why the procedure is shaped that way. Which step exists because of a bad experience nobody wants to repeat. This is occasionally written down, usually in a comment thread on a document from four years ago, and it is effectively unfindable.
The judgment. What to do when the procedure does not apply. Which rules bend, for whom, under what pressure, and what it costs to bend them. This is never written down, because it is not a rule. It is a disposition, built out of a few hundred small cases, and the person holding it usually cannot articulate it unprompted. Ask them to "document what you know" and they will hand you the procedure, because the procedure is the part they can see.
Judgment is the part that walks out the door. It is also the part that took twenty years to build.
The part where it is worth something
There is a version of this argument that leans on scary numbers. We are not going to make it, because the honest data on knowledge loss is thin and most of the figures in circulation do not survive contact with their own footnotes.
What we can say is what one firm told us about their own experience.
A 74-year-old operator sat for a structured voice session. It ran about three and a half hours. Afterwards the firm's own estimate, on their own work and their own criteria, was that over 80 percent of what he knew was now captured.
That is their number, not ours, and it describes one person at one company. We report it because it is the only public proof we have, and we would rather have one real account than a page of borrowed statistics.
What it suggests is narrower than a headline and more useful. The knowledge is not unreachable. It is not so vast that a person cannot get it out. It is just that nobody ever sits down and asks properly, for long enough, with the right questions.
What would have to be true
If you wanted to fix this, and not just feel better about it, three things would have to change.
It would have to happen before the resignation. Two weeks' notice is not a knowledge transfer window, it is a scramble. The capture has to be routine, done on people who are staying, or it is only ever done in an emergency.
It would have to be a conversation, not a form. People cannot fill in what they know, because they do not know they know it. They can, however, answer a question about a specific hard case. The format has to do the work of asking.
And it would have to be worth something to the person answering. A system that asks people to write down everything that makes them hard to replace, with nothing offered back, gets exactly the participation it deserves. This is the part most knowledge management projects never solve, and it is why most of them fail quietly.
Start with the question
You do not need a system to run the diagnostic. You need one afternoon and the willingness to hear an uncomfortable answer.
Ask the hard questions. Count the names. Look at how short the list is.
Every company runs on a few heads that were never written down. The only real variable is whether you know whose they are before you find out the hard way.