The person who knew why

Organisations keep almost everything except the reasoning that connects it. As machines get better at reconstructing the past, the difference between what was said and what was decided matters more, not less.

Imagine a new commercial director reviewing a five-year supplier contract. Two things look wrong. The price is high, and one obligation seems generous to the point of carelessness. She asks the obvious question: why did we agree to this?

The company has kept everything. The contract is in the system, and so is the tender. There are evaluation spreadsheets, board papers, minutes, a consultant's report and three years of email with the supplier. Someone finds a note from a meeting that was clearly important.

None of it answers her question. The man who could have answered it left two years ago. He knew that the supplier had once threatened to walk away over that obligation, that the business gave way because a plant could not survive a transition at that point in its cycle, and that the concession was paid for by a cap on price rises elsewhere in the agreement. He had told people this, but none of it was ever written down as a reason.

What leaves with people

What that man carried was not a set of facts. The facts were all in the files. He carried the connections between them: which clause had been traded for which, whose objection had shaped the final terms, and what the business had been afraid of at the time. That kind of understanding builds up without anyone noticing, one negotiation and one difficult project at a time, until a person knows an organisation in a way no system describes. The research bears out what that means when such people leave: a 2023 review of 91 empirical studies found that how much an organisation loses depends on who leaves and what kind of knowledge goes with them (Nataliya Galan, The Learning Organization, Part I and Part II).

That finding matters wherever knowledge sits with people who will not stay: contractors and consultants who arrive for a project and leave when it ends, outsourcing that moves whole areas of operational knowledge across a company boundary, and new leadership teams who inherit hundreds of decisions made by people they never met. None of this is wrong in itself, since organisations need new people and new ideas. The question is what happens to experience when the people who gathered it move on.

What it costs

The cost rarely appears as a line anyone can point to. It shows up as a series of small, expensive repetitions. A procurement team goes into a renegotiation without knowing what was conceded last time, or what was given in return, and concedes it again. A project that was piloted and abandoned five years ago is proposed afresh, and the organisation pays to learn the same lesson. A control that looks like bureaucracy is removed, and the problem it was put there to prevent comes back. An auditor asks why a risk was accepted, and the only people who could say have gone, so the answer is assembled after the fact, which is exactly the kind of answer an auditor trusts least.

None of these is a failure of storage. The documents were there each time. What was missing was the reasoning that would have told the new people which documents mattered, and why.

Keeping everything is not the same as keeping the reason

The usual answer has been to capture more. Write the procedure down, store the project report, record the meeting. These are all sensible, and none of them quite reaches the problem.

Take a board decision. Five years on, the papers survive, the recommendation survives and the minutes say it was approved. An executive can see what was decided. But the question that matters later is usually why, and the answer to that was never in the papers. One director objected and came round only when a condition was added. Another supported it on a promise that management would come back with evidence in six months. An outside adviser shaped the discussion without ever speaking for the company. The decision is a small thing compared with the circumstances that produced it, and those circumstances are the first to go.

A transcript of that meeting would not solve this either. It would preserve everything that was said, including the speculation, the objection that was later withdrawn, and the adviser's view, all with equal weight. It cannot say which of those things the organisation came to hold. That takes a person, after the discussion, saying in effect: this is what we decided, these are our reasons, this is who disagreed, and these are the conditions we accepted.

The distinctions in that sentence are the ones that matter. Advice is not a position. Disagreement is not the decision, but it belongs on the record beside it. And something said in a meeting is not organisational knowledge until the organisation adopts it.

Reconstruction is not preservation

Artificial intelligence has changed what is possible here. Give a capable system the tender, the evaluation, the contract, the email and the meeting notes, and it can produce a clear account of why a supplier was chosen in 2021. It will find connections a person would take days to trace. That is genuinely useful.

But an account inferred from surviving material is not the same thing as a record the organisation made. The machine's explanation may be right. It may also be plausible and wrong, and nothing in it will say which. More important, nobody at the company decided that this was the explanation. The system chose which voices to believe: perhaps the confident email over the hesitant minute, or the adviser's report over the objection that changed the outcome.

The easier it becomes to search and summarise the past, the more it is worth knowing which parts of that past the organisation deliberately kept as its judgement. None of this argues for less technology. It argues for being clear about what the technology is doing. Finding the evidence is one job. Saying what it meant is another, and it belongs to people.

The strongest objection to this is that people are not neutral witnesses to their own decisions. A reason written down by the people who decided can be self-serving, tidied after the fact to look wiser than it was, while a reconstruction from the raw material might surface exactly what they left out. The objection is fair, and it is why the timing and the shape of the record matter. Reasons written at the time, before the outcome is known, are harder to polish. A record that keeps the dissent and the outside advice beside the decision cannot quietly drop the objection that was overruled. And a record that is superseded rather than edited cannot be rewritten once events have shown who was right.

Changing your mind, on the record

None of this means the past should govern the future. Markets move, regulations change, and an assumption that was reasonable in 2019 can be indefensible now. A healthy organisation changes its mind often. The problem is changing it blind.

A new executive should be able to overturn an old decision without spending six months finding out why it was made. A new procurement team should be able to renegotiate knowing what the last team traded, and for what. When the old decision rested on a condition that no longer holds, that is worth knowing. When it was simply wrong, that is worth knowing too. Either way, the earlier decision is not something to erase. It is part of the explanation for the new one, and both should stay readable.

This is why "knowledge transfer" can be the wrong way to think about it. It suggests moving a quantity of knowledge from one head to another before someone leaves, like files between folders. Experience does not move that neatly, and a two-week handover cannot carry fifteen years. What an organisation can do is keep the reasoning as it goes: decision by decision, at the time, while the people who hold it are still in the room.

The commercial director will find someone to ask, or she won't. One day her successor will ask the same question, and the answer will depend on what was kept.

Written by Claude, an AI model made by Anthropic, at Pohja's request. Pohja makes software built on this argument.