The Knowledge That Leaves at Five O'Clock: Mapping UK Operational Know-How Before It Goes
Photo: handover meeting colleagues discussing documents office knowledge transfer, via showroom-assets.tradingpaints.gg
Ask a UK operations team how a month-end close actually completes and you will get two answers. The first is the process map, which is tidy, documented and broadly accurate about the sequence of steps. The second is the version delivered by the person who has done it for nine years, which includes the manual file rename, the supplier portal that must be refreshed before the batch runs, and the reason the third validation step must never be run before the second.
Only one of those answers is load-bearing. The other is a diagram.
What counts as tribal knowledge
Not everything undocumented is tribal knowledge. Some of it is simply detail that nobody needs. The knowledge worth mapping has three characteristics: it is required for a critical outcome, it is not recorded anywhere a successor could find it, and it is held by a small number of people — often one.
In practice it clusters in predictable places. Around exception handling, where the documented process describes the happy path and the real one describes what to do when the interface times out. Around timing, where the order of operations matters for reasons that were true in 2019 and have never been retested. Around external relationships, where a named contact at a supplier or a regulator responds to one person and ignores a shared inbox. And around reconciliation, where the acceptance criteria for "close enough" live entirely in someone's judgement.
Starting the map
A useful map begins as a matrix rather than a document: critical operational processes down one axis, named individuals across the other, and a mark wherever a dependency exists. The marks should distinguish between people who can perform a step and people who can recover it when it fails. The second category is far smaller and far more important.
Populate the matrix from three sources. First, the runbooks and process documentation, which tell you what is supposed to happen. Second, the incident record, which tells you who was actually called and who actually fixed things. Third, structured interviews with the people named in the second source, which tell you what they know that nobody wrote down.
Interviewing for the exception, not the process
Asking someone to describe their job produces a summary. Asking them to describe the last three times it went wrong produces something usable.
A small set of prompts works well. Walk me through the last occasion this failed and what you did first. What would you check in the first ten minutes if you came in tomorrow and the overnight batch had not completed? Which step in the documented process do you routinely skip, and why? Who do you telephone when this goes wrong, and what do you say to them? What would you tell a new starter that would not be in the handbook?
The answers should be captured verbatim before they are tidied. The tidying step is where the value leaks away, because the informal detail is exactly the part that gets edited out.
Testing exposure rather than assuming it
The bus factor — the number of people who would have to be unavailable before a process stops — is usually estimated in conversation and rarely tested. A short exercise converts it into evidence.
Pick five critical processes. For each, arrange for the primary holder to be unavailable for a working day, with no informal contact permitted, and ask a nominated deputy to complete the process from documentation alone. Record where they stopped, what they guessed, and what they had to ask. The output is not a pass or fail; it is a list of gaps with a named owner against each.
The exercise is uncomfortable, which is why it is valuable. Most UK teams discover that the documented runbook covers the first two-thirds of the process and that the final third — the part involving judgement — is entirely unwritten.
Converting knowledge into an asset
The conversion has to produce artefacts that survive contact with a real week. Three forms are enough.
Runbooks written for recovery, not for description. They should open with symptoms, not with an architecture diagram, and should state the decision points explicitly: if the reconciliation variance exceeds this threshold, escalate; if it does not, proceed and note it.
Training that is delivered by the deputy rather than the expert. The act of teaching forces the gaps into the open, and the deputy's questions often reveal assumptions the expert no longer notices.
Ownership assigned in the operational register, not just in the team's collective memory. Every critical process should have an accountable owner, a named deputy and a review date. Knowledge that has no owner decays quietly and is discovered to have decayed at the worst possible moment.
The retention question
The uncomfortable economics are straightforward. Knowledge held by one person is a single point of failure with a resignation notice period attached. The cost of extraction is real — several days of expert time, some of it spent being interviewed rather than working — but it is small compared with the cost of a failed close in the week after that person leaves.
A reasonable target is one critical process documented and reverse-shadowed per person per quarter. It is unambitious enough to survive competing priorities and consistent enough to close the exposure over a financial year.
Why it belongs in the data estate
This is not only a human resources question. The workarounds contain the real business rules: the tolerance applied to a reconciliation, the override that allows a late invoice through, the manual correction that adjusts a system of record. Those rules belong in the governed definition of the data, not in somebody's memory of what they did last November.
Mapping tribal knowledge is therefore an act of data governance as much as operational resilience. The runbook is the visible output. The recoverable, defensible business logic underneath it is the point.