The three failures of a firm's transactive memory

Most firms describe their knowledge problem as one thing: "we don't know what we know." That framing is why the fixes keep missing. The research separates a transactive memory system into three dimensions, and they break independently. A firm can be perfect on two and catastrophic on the third, and the remedy for each is different enough that treating the wrong one is not merely wasteful, it is the reason so many knowledge programmes end in a wiki nobody opens.

Here is each failure, what it feels like from inside, and how to tell which one you actually have.

Failure one: specialisation

The theory. Specialisation is the extent to which members of a group hold different, complementary knowledge, and the extent to which the group is aware of that division. Both halves matter. Differentiated expertise that nobody can see is not specialisation in the transactive sense, it is just a set of people who happen to know different things.

What failure looks like. Everyone claims everything. Capability statements list every service against every partner. The firm's own view of itself is a matrix with no zeroes in it, which is another way of saying it carries no information at all. Nobody can distinguish the partner who has led four regulatory price reviews from the partner who attended a webinar about them.

Where it comes from. Self reported expertise, almost always. Ask people to describe their own capability and you will get aspiration, because in a firm where staffing follows stated capability, understating yourself is career limiting. The result is a directory in which everybody is a generalist and the specialisation dimension has been flattened out of existence.

What fixes it. Evidence rather than assertion. A record of what someone has actually delivered differentiates automatically, without asking anyone to rank themselves against a colleague. Nobody has to claim to be the expert. The engagement history says it or it does not.

Failure two: credibility

The theory. Credibility is whether members trust one another's knowledge enough to rely on it rather than reproduce it. A group with perfect specialisation and zero credibility performs like a group of strangers, because every member independently redoes work that a colleague has already done correctly.

What failure looks like. Rework. A team spends three days on an analysis the firm has done twice before, because the person doing it either did not know the earlier work existed or, having found it, did not trust it enough to build on it. It also looks like partners who will only staff engagements with people they have personally worked with, which reads as a preference and is actually a credibility signal: they trust the directory entries they built themselves and no others.

Where it comes from. Credibility travels along the lines the work travelled. You trust the people you have delivered with. In a firm of thirty, that is everyone. In a firm of six hundred, it is a small neighbourhood, and the space between neighbourhoods is where trust does not reach.

There is a second source, and it is the more damaging one in 2026: a system that has been caught being wrong. The moment a firm's own knowledge tool returns something confidently incorrect, its credibility falls to zero and stays there. Nobody consults a source that might be inventing things, so the system is bypassed entirely, and the firm goes back to asking Sarah.

Credibility is not earned by being right. It is earned by being checkable.

What fixes it. Provenance. If every fact carries the document it came from, credibility no longer has to be extended on faith. A partner who does not trust an assertion can open the proposal behind it and decide in one click. That converts trust from a relationship property, which does not scale, into a property of the record, which does.

Failure three: coordination

The theory. Coordination is the group's ability to retrieve and combine what it holds, smoothly, at the moment of need. It is the dimension that turns a directory into a working system.

What failure looks like. This is the one every firm actually has. It looks like:

• A broadcast email to four hundred people asking whether anyone has done something, followed by a day of waiting and three replies, none of them from the person who did it.
• A partner walking into a stakeholder meeting cold while two colleagues who know that stakeholder well sit three desks away, unaware.
• A pitch that omits the firm's strongest relevant credential because the person writing it did not know it existed.
• A client buying a service elsewhere that the firm delivers, well, for two other clients in the same sector.

None of those are knowledge problems. In every one of them the knowledge existed, inside the firm, held by someone competent, and was not retrievable at the moment it was worth something.

Where it comes from. Retrieval in most firms depends on someone remembering at the right moment. That mechanism works beautifully at twenty people and degrades continuously from there, because it scales with the number of pairs of people who have worked together, which grows far more slowly than headcount.

What fixes it. An index that is not a person. Specifically, one that can be interrogated from the question the firm actually has ("who has worked with this stakeholder", "have we done this in water regulation") rather than from a filing structure that assumes you already know where to look.

Which one do you have?

A short diagnostic. For your firm, answer these three honestly.

| Question | If the answer is no | |---|---| | Could you list, with evidence, the five colleagues with the deepest experience in your largest client's sector? | Specialisation is invisible | | When a colleague you have never worked with tells you the firm has done something before, do you build on it, or check it yourself? | Credibility does not travel | | When you needed that answer last month, how long did it take, and did you get it from a system or from a person? | Coordination has failed |

In our experience of these conversations, firms answer no to the first and third and hesitate on the second. That combination points in one direction. The expertise is real, the trust is real inside pockets, and the retrieval mechanism is a human being with a good memory and a finite tenure.

Why the order matters

Fix coordination first, and the other two improve on their own.

An index built from evidence makes specialisation visible without asking anyone to rank themselves. An index that shows its sources makes credibility checkable without requiring a prior relationship. Neither of those is true in reverse: no amount of trust building or capability tagging creates retrieval.

This is the practical reason OrgAtlas is an atlas rather than a repository. The knowledge is not missing. The route to it is.

Next: the three questions no firm can answer quickly, which are coordination failure in its three commercial forms