- Most firms can answer what they got wrong on recent deals from a partner's memory and cannot answer it from their files.
- Three things get lost: the reasoning behind the conclusion, the near-miss work on deals that were passed, and the pattern that only becomes visible across eight deals.
- The most valuable diligence a firm does is on deals it does not close, and it is the work most completely destroyed.
- Post-mortems, CRMs and enterprise search each fail differently: post-mortems exclude passes, CRMs store fields rather than reasoning, and search retrieves documents rather than answers.
- Capturing a pass costs a deal team about forty minutes and produces the artifact firms end up referencing most.
- The useful question to put to any institutional-memory vendor is what specifically it remembers and what it does with that on the next deal — demonstrated against your own history, not a sample dataset.
Ask a partner this question in a meeting and you will get an answer. It will be a good answer, delivered with confidence, and it will be drawn entirely from memory — the partner's own, unaided, filtered through five years and whatever the last conversation about that deal happened to emphasise.
Ask the same question of the firm's files and you will get nothing at all. The IC memo says what the firm believed at approval. The quarterly reports say what happened. Nothing in between says what the firm got wrong, because nobody wrote it down, because there was never a moment in the process where writing it down was somebody's job.
This is the actual state of institutional memory at most funds, including very good ones, and it is worth being precise about the failure rather than gesturing at it.
What gets lost, specifically
The reasoning, not the record. Firms keep excellent records. The memo, the model, the data room index, the closing binder — all of it survives, indexed by deal, in a shared drive. What does not survive is why the team weighted the customer concentration risk as acceptable, what the counter-argument was, and who made it. The record says the firm concluded. It does not say how, or against what objection.
The near-miss. The most valuable diligence work a firm does is on deals it does not close, and it is the work most completely destroyed. The team spends six weeks and $140,000 on a target, finds the thing that kills it, passes, and the entire artifact of that discovery goes into a folder nobody opens again. When a similar business comes through eighteen months later, the firm re-buys the same insight at full price.
The pattern that takes eight deals to see. Some things are only visible across a portfolio. A firm might discover, if it could look, that every business it bought where the founder stayed on as a minority holder without an operating role underperformed the base case by a similar margin. That pattern is invisible on any single deal and invisible in any single partner's memory, because no single partner did all eight.
Why the obvious fixes have not worked
Firms have tried. The post-mortem process is the usual attempt: a session after a bad outcome, notes taken, a document filed. Two problems. It happens after exits rather than after passes, which excludes the near-misses. And the output is a narrative document that lives in a folder, which means it can only be found by someone who already knows it exists and goes looking — which is to say, by the people who were in the room, who did not need it.
The second attempt is usually the CRM. Firms load deal history into the CRM and discover that a CRM stores fields. It will tell you that you looked at the company, when, and what the outcome was. It will not tell you that the reason you passed was a supplier arrangement you have now seen three times in the same sector.
The third attempt is enterprise search, which retrieves documents. Retrieving the memo is not the same as answering the question, and the gap between them is the whole problem.
What actually has to be true
For the question at the top of this page to have a real answer, four things have to hold.
The pass has to be captured as carefully as the close. This is a process change before it is a technology change. It costs a deal team perhaps forty minutes: at the point of passing, record what was found, where it was found, and what would have had to be different. Firms that do this find within two years that it is the most-referenced artifact they own.
The history has to be queryable in the language people actually use. Nobody searches for "supplier concentration, sole-source, high-margin SKU." They ask whether the firm has seen a distributor where one supplier controlled the good product. The retrieval has to work on the question, not on the keyword.
Structure has to be imposed at ingestion, not at query time. A decade of deal folders is not a dataset. It is PDFs, Excel files, scanned contracts, email chains, and three generations of naming convention. Making it answerable means extracting entities and findings as it goes in. This is unglamorous and it is most of the work.
It has to compound without anyone maintaining it. Any system that requires a person to curate the knowledge base has already failed, because that person's time gets reallocated the first quarter things get busy. The capture has to be a byproduct of the diligence rather than a task appended to it.
A note on what everyone is now selling
"Institutional memory" has become a crowded claim in the last year. Several platforms now market compounding firm knowledge in close to identical language, and the category is at risk of the term becoming a feature checkbox that means nothing.
So the useful question to put to any vendor, including us, is not whether they have institutional memory. It is: what specifically does it remember, and what does it do with that on the next deal? A system that can retrieve the memo is a search engine. A system that can tell you the target in front of you resembles a business you passed on in 2023 for a reason that also applies here, and can point to the three documents where that reason lives, is doing something else.
Ask for that demonstrated against your own history rather than a sample dataset. It is a fair thing to ask for and the answer is informative either way.
The version of this that matters
The reason to care is not tidiness. It is that a firm which cannot answer what it got wrong is a firm that gets its diligence quality reset every time the bench turns over, and turnover at the associate and VP level in this industry runs on a two-to-four year cycle.
That means the firm's diligence is roughly as good as its most experienced person's memory, permanently, without improvement, forever. Every deal is the first deal in the only sense that matters.
The firms that break this pattern are not the ones with the best tooling. They are the ones that decided the pass was worth writing down.