Your firm's memory has a half-life of 4.1 years

Firms treat knowledge loss as an event. Someone senior resigns, there is a handover meeting, a folder gets shared, and the loss is booked as a one off cost of that departure.

It is not an event. It is a rate.

The two numbers

Two figures from the research, put next to each other, describe something most firms have never quantified.

The first number says your firm replaces itself roughly every four years. The second says that when a person goes, a little under half of what they knew about doing their job goes with them, because it was never anywhere else. The third is the running cost of the gap, paid weekly, by everyone still there.

Put the first two together and you get a decay curve rather than a series of incidents. A firm that does nothing loses a predictable fraction of its working memory every year, and the fraction is large enough that the memory it operates on today is mostly not the memory it had five years ago. The names on the client list are continuous. The knowledge behind them is not.

A firm's client list has decades of continuity. Its memory of those clients has about four years.

What it costs, at the numbers people actually publish

The macro figures are large enough to be easy to dismiss, so it is worth being specific about what each one measures.

• $1.3 trillion a year is Deloitte's estimate of the cost of voluntary knowledge-worker turnover across the US economy, with knowledge loss the largest single component. Not recruitment. Not onboarding. The knowledge.
• $31.5 billion a year is the estimated cost to Fortune 500 companies of failing to share knowledge they already have. This one is not about departures at all. It is the cost of knowledge that is still in the building and cannot be reached.
• 50% to 200% of annual salary is SHRM's range for the direct replacement cost of a departing knowledge worker, before any allowance for what they took with them.
• $47 million a year is an estimate of the productivity cost of inefficient knowledge sharing for an average US business.

You can argue with any individual figure, and you should treat all of them as indicative rather than precise. What survives the argument is the shape: the cost of knowledge that has left the building is smaller than the cost of knowledge that is still inside it and unreachable.

Why professional services firms have it worse

Every organisation faces this. Firms face a sharper version, for four structural reasons.

The knowledge is the product. In a manufacturer, a departing engineer leaves behind drawings, tooling and a line that still runs. In a firm, a departing partner leaves behind time sheets. What they knew about a client's board dynamics, the history of a failed procurement, the reason a particular approach was abandoned in 2021, was never an artefact.

Relationships are held personally, by design. The client relationship model that firms are built on deliberately concentrates trust in individuals. That is good for delivery and catastrophic for continuity. When the relationship holder leaves, the client does not experience a personnel change. They experience being asked questions they answered four years ago.

The pyramid guarantees churn at the base. Firms are structured so that most people leave. Up or out is not a retention failure, it is the operating model. Which means the decay rate is not something the firm can fix by being a nicer employer.

Billable pressure prices documentation out of existence. Every hour spent recording what you learned is an hour not billed. In a system where non chargeable time is the thing you defend at appraisal, the rational individual choice is to not write it down, and the aggregate result is a firm with no memory.

The compounding part

The straightforward loss is what the leaver knew. The compounding loss is what they knew about who else knew things.

This is the transactive memory point, and it is the reason the decay is faster than headcount arithmetic suggests. A fifteen year veteran is not one directory entry. They are a directory. They hold hundreds of pointers: which colleague handled that regulator, who ran the failed bid, which associate turned out to be unexpectedly good at data protection. When they leave, the firm does not lose one node. It loses a hub, and every path that ran through it.

That is why firms notice knowledge loss late and then all at once. The system tolerates the loss of leaf nodes for years, then a hub goes and a whole region of the firm's memory becomes unreachable in a single quarter.

What actually reduces the rate

Three responses are common. Two do not work.

Exit interviews and handover documents. These capture what the leaver can articulate in the two weeks they have already mentally left. They capture almost none of the tacit directory, because the leaver does not experience it as knowledge. It is just what they know.

Repositories and knowledge bases. These ask people to externalise knowledge continuously, in the exact economy where non billable time is punished. They decay from the day the mandate ends. Every firm has one. Ask when yours was last updated.

Recording work as it happens, as a by product. This is the only one with the right cost structure, because the marginal effort is close to zero. The material already exists: proposals, reports, and the email in which the actual reasoning happened. The question is only whether any of it is assembled into something navigable while the person who wrote it is still there.

The handover should not happen when someone resigns. It should have been happening all along, without anyone doing it.

What this means practically

You cannot lower the tenure number. Firms are pyramids and people move. What you can change is the fraction that leaves with them.

If 42% of what a person knows is held only by them, the target is not zero. The target is to get the durable, commercially valuable part of it, meaning who they worked with, what they delivered, what the firm learned, into a form that survives their notice period. That is a much smaller and more achievable ask than "capture what you know", and it is the part that turns out to matter when a colleague walks into that client's office three years later.

OrgAtlas is built for that specific slice. Not everything a person knows. The part of it that the next person needs.

Next: what a firm should record about an engagement, and what it never does