What is tacit knowledge in an organization?

Tacit knowledge is the part of what an organization knows that exists only in people's experience and resists being written down: it is not in a manual, a ticket or a procedure, because the person who holds it cannot fully say how they do what they do. The concept comes from the chemist and philosopher Michael Polanyi, who in The Tacit Dimension (1966) compressed it into one line: we can know more than we can tell. Inside a company it takes very concrete forms. The shift supervisor who hears which machine is about to go down. The billing clerk who knows which three customers are invoiced differently from everyone else, and why. The developer who knows the one safe order for restarting the nightly batch when it hangs. It is real, productive knowledge, recorded nowhere, and the only asset a company owns that walks out the door on the day its holder leaves, with no copy left behind.

See also: Chesterton's fence · Do not remove an unexplained piece of code until you have found, with evidence, why it was put there.

Why writing procedures is not enough

The instinctive answer is to document everything, and it fails for a structural reason: part of this knowledge was never written down because nobody has yet recognized it as knowledge. Whoever holds it experiences it as common sense, not as transferable knowledge, so they would not write it down even if they had the time. Ikujiro Nonaka modeled the transition as the SECI cycle (socialization, externalization, combination, internalization), set out in his 1994 paper in Organization Science: tacit knowledge first moves from person to person by working side by side, then someone puts it into words, and only then can it be combined with the rest of the documentation. Skipping the first two steps produces archives nobody recognizes as useful. And this is where today's promise deserves a close reading: many organizations are using AI to index and reuse their knowledge archives, but a model can only rework what it finds. If the archive is thin, AI does not create the institutional knowledge that is missing: it produces a plausible substitute, which is worse.

An enterprise example

At a two-hundred-person engineering firm the order configurator has been running for eighteen years. The person who wrote it has retired, and with him the reason a certain discount is recalculated only for orders shipped on delivered-at-place terms. Nobody touches that rule, and nobody can explain it: this is exactly the situation Chesterton's fence describes, and lost tacit knowledge is the reason that principle is needed at all. Recovery does not start with a documentation project, it starts with traces: version history, logs, and above all the two people in the back office who use that rule every week without knowing it is a rule. An agent-maintained wiki is useful downstream of that work, never instead of it.

Why it matters for decision makers

Tacit knowledge appears on no balance sheet, so its risk surfaces only once it materializes: a resignation, a retirement, the end of a staff augmentation contract. The question worth asking is not how well documented the system is, but how many people could explain a critical part of it if the usual go-to person were not there tomorrow. Wherever the answer is a single person, there is a real operational risk, and it shrinks through shared ownership and pairing, not through an instruction to write documentation. The explicit knowledge that results then needs an authoritative home, which is what a single source of truth is for: but the first problem remains surfacing what was never written down at all.