Institutional Memory: What Leaves With the Person Who Decided
The Compass Team
November 8, 2026
Two repair shops buy the same car lift and the same manual. One has a lower ceiling.
Its mechanics learn that raising a car past a certain height puts the roof through the plaster. They stop at a mark no one wrote down. New hires pick it up by watching, then teach the next hires.
Institutional memory runs on that chain: shared knowledge held by a group, passed between its members, alive for as long as the chain is. Wikipedia's entry on it is blunt about the mechanism.
Memory requires ongoing transmission between members, so it depends on continuity of membership. If everyone at the shop left at once, the replacements couldn't benefit from what the last group knew.
Startups run that failure at a size where it hurts. An 8-person company loses 12% of its memory when one person leaves. The founder is usually the one holding the reasons, and the reasons are the part no onboarding doc covers.
Dorothy Leonard and Walter Swap spent two years studying whether that knowledge moves at all, in 36 startups and their coaches across California, Boston, Hong Kong, Singapore, China and India. They interviewed founders and coaches in 2000 and again a year later.
The finding starts with a hard limit. Expertise is pattern recognition built over years. Chess experts carry more than 10,000 positions in their heads, by one early estimate, and later work put the range at 100,000 to over a million.
Entrepreneurship draws on financial, strategic, marketing and interpersonal judgement at once. The authors call that beyond anyone's ability to fully master, and therefore to hand to a novice.
Their study window still caught the handoff. Nine of the 36 founders were removed from operations or had their role changed significantly inside about two years. Six coaches left, and five companies went under.
The fix converges on one stage. A scoping review of 8 studies published between 2011 and 2022 found documentation is the most crucial part of knowledge management, because if the knowledge and experience aren't documented, the later stages of the process fail. It found no agreed procedure, listing planning, acquisition, registration, evaluation, submission, maintenance, publication, application and compensation across those papers.
The clearest version of the practice sits in the NHS knowledge retention and transfer toolkit, built for people leaving a job. It opens with a number worth sitting with. 75% of organisational knowledge moves through email, and it dies when the account is deleted.
So the toolkit asks for work in hand statements on every unfinished area, a calendar of what recurs across the year, a glossary for the specialist vocabulary. Then an elicitation interview to pull out what is still only in their head.
Read that list again. Every item is a process artifact.
You can rebuild a process in a week from a screenshot and two people who care. What you can't rebuild is why the pricing test died in March, why that hire went wrong two months in, why you turned down the customer who looked perfect. Those are decisions, and a decision without its reason is a rumor with a date on it.
One 2024 case study in the International Journal of Academic Research in Business and Social Sciences shows how badly this goes. A knowledge-intensive institution deleted departing employees' cloud folders under a cost-cutting procedure, before anyone had taken over the work. A two-year project lost its context.
The authors' answer is to recover meaning from the unstructured messages people left behind, which is an expensive way to rebuild something you could have written in the hour it happened.
The test I use is short. For any decision that costs real money, ask what the next person would have if you left tomorrow. If the answer is a Slack thread from four months ago that no one can find, the decision lives in someone's memory, and that expires on a schedule you don't control.
Six lines, in the hour the call is still open: what we decided, what we assumed would happen, what number we expected, what the number did, what changed, what we would tell ourselves next time.
It takes ten minutes, and the next person inherits the reasoning instead of a folder of outcomes.
The lift in the low-ceiling shop works because someone standing there remembers. A company's memory lives in its record, and the record only counts when it can still explain itself after the person who wrote it has gone.
Share this article