The phrase covers everything a company knows that is not written anywhere, and it is used mostly as a complaint. That is unhelpful, because the material it describes splits cleanly into three kinds with three different answers, and the standard program, which asks experienced people to document what they know, fails on two of the three.
Kind one: facts that were never written down
Where the spare key is, which account manager covers this region, what the code in the third column means, who to call when the overnight job fails. This is the largest kind, the cheapest to capture and the one most often skipped because it looks too trivial to be knowledge. It is not trivial; it is the material a new starter burns their first month on.
Kind two: exceptions and the reasons behind them
This customer is invoiced differently because of something agreed in 2019. This step is skipped for that product because of a defect nobody fixed. This approval is routed around because the named approver left. Each is a fact plus a reason, and the reason is the valuable half, because without it the next person will either break the exception or preserve it forever after it stopped being needed.
Kind three: judgement
Knowing which of two plausible readings of a situation is the real one, which customer complaint is about the stated problem and which is about something else, when to escalate and when to wait. This is genuine expertise, it is built from hundreds of cases, and it does not survive being written down by somebody in a hurry.
Why the standard program fails
Asking an experienced person to document what they know invites them to attempt kind three, which is the hardest and least transferable, and to skip kind one, which feels beneath the exercise. The result is a long document about principles and no article saying where the spare key is. Reversing that is most of the work.
What to ask instead
Three questions produce articles: which decisions do people come to you about, what makes one of those go the unusual way, and who do you call when you are unsure. The first two produce kinds one and two directly. The third produces a directory entry, which is the honest partial answer for kind three.
The part of judgement that does transfer
Not the judgement itself, but the cases. A short write-up of a real case, what looked true, what turned out to be true and what the tell was, is readable and repeatable, and a dozen of them get a new person further than any statement of principles. This is the only mechanism that reliably moves kind three and it is slow.
The timing that matters
Capture happens when the question comes up twice, not when somebody resigns. A base written in a departing expert's last fortnight gets the panic version: passwords, half a process, and a list of contacts with no context. The steady version, written because somebody asked the same thing again, is the one that holds.
Questions people ask about tribal knowledge
Is tribal knowledge always a problem?
No. Some of it is efficient: a small team that shares context does not need to write everything down for itself. It becomes a problem at the boundary, when somebody new arrives or somebody experienced leaves, and those two events are predictable enough to prepare for.
How do we start without a big programme?
Write the article the second time somebody asks. That rule alone, applied for a quarter, captures most of kind one and a good deal of kind two, and it needs no budget and no launch.
What about the person who will not write anything?
Interview them and write it yourself, then have them correct it. Correcting is far easier than composing, and the people who know the most are usually the ones with the least appetite for a blank page.
Does an assistant reading our chat history solve this?
It surfaces some of kind one, which is useful. It cannot tell an exception from a mistake, because both look like a deviation in a transcript, and that distinction is exactly what makes kind two worth having.
How much of it is worth capturing at all?
The two cheap kinds, nearly all of it, because the cost per article is minutes. Judgement, only through cases, and only where the same judgement is exercised often enough that a new person will meet it.