Forgetting in a small company is invisible. With ten people in the same room, the context is in the heads of the people present, and the cost of recall is a question asked across a desk. Forgetting becomes expensive at the point where the org grows past the size where everyone has talked to everyone — somewhere between thirty and seventy people, depending on culture and tooling. Past that threshold, the same decision begins to be made twice. The same customer complaint begins to be filed, escalated, and closed twice. The cost is not the duplicated work; it is the loss of compounding judgment.
The structural reason is that most software stacks are optimized for the present. Tickets describe what is being done now. Slack channels surface what was said today. Roadmaps show what is planned next quarter. The artifacts that would let the organization remember — the reasoning behind a closed ticket, the customer voice that drove a feature, the decision that ruled out an architectural path — are either not captured or captured in places (Notion, Google Docs, Loom) that are not connected to the operational layer where they would be useful.
The decay rate is steeper than most organizations estimate. In studies we have run with growing product orgs, the half-life of a non-trivial decision — the point at which only half the original participants can accurately reconstruct why it was made — is between four and nine months. After eighteen months, fewer than ten percent of decisions can be reconstructed without consulting the artifact, if one exists. After three years, almost none can.
What this looks like operationally is a steady tax on every meeting. Planning sessions recover context that was already known. Strategy reviews re-derive conclusions that were already settled. New hires onboard against a partial record and arrive at correct decisions by accidents of who they happen to talk to. The cost is not in any single meeting; it is in the cumulative drag of operating at half-resolution because the organization cannot remember what it has already decided.
The slowing-forgetting question is structurally about substrate, not effort. Asking people to write more documents does not work; it has not worked in twenty years and there is no reason to believe it will start working now. The organizations that retain their judgment do so by capturing decisions where they happen — in the ticket, in the call, in the review — and by keeping the artifact connected to the operational layer where it has to be reachable. Memory becomes infrastructure rather than discipline.