Customer concentration, three documents, four months
No single document is alarming. The pattern across three of them is the finding — and it exists only if something is holding all three at once. Every firm and figure below is invented.
Meridian Health is a portfolio company. Over four months, three separate documents mention that its two largest customers account for an uncomfortable share of revenue. Each mention is reasonable, each author moved on, and nobody in the firm has ever seen the three of them together.
This is the most ordinary failure in professional work and it is almost never described as a failure, because at no point did anyone do anything wrong.
What the three documents said
In March, an investment committee memo recommended a follow-on investment and named customer concentration as the principal risk, at 38% across two customers. The committee approved, subject to confirmatory diligence.
In April, a diligence provider's report mentioned in passing that the larger of the two customers was renegotiating terms. It was one line in a section about commercial relationships, and it was not flagged.
In May, a board deck reported trading below plan and restated ARR from €12.4m to €11.9m. The deck did not connect the restatement to the concentration; it did not need to, because that was not what it was for.
Three documents, three authors, three purposes. Each individually unremarkable.
Why does nobody put them together?
Because nobody is paid to. The memo's author was making a recommendation and did so. The diligence provider was answering a scope. The board deck was reporting a quarter. Connecting them is not any of their jobs, and it is not anyone else's job either — it is the thing that is supposed to happen automatically in a partner's head, and does, until the partner is looking at nine other companies.
It is also genuinely hard. The three documents live in different places, use different words, and arrive months apart. Finding them requires knowing they exist, which requires having read them, which is the thing that does not scale.
Nothing here was missed through carelessness. It was missed because noticing was nobody's task and nobody's system.
What a memory does with this
It counts. Because each document was read on arrival and what it established was written down as typed cited facts, the three mentions are three facts about the same entity, and the recurrence is computable: concentration raised in three documents across 120 days.
The finding is not “this deal is risky”. It is narrower and more useful than that: this concern has come up three times, here are the three documents, here are the dates. The threshold — three in four months — is your firm's parameter, and the judgment remains entirely with the person reading.
The same machinery also raises the part that is not in any document. The March approval was conditional on confirmatory diligence, and no later document records the condition being discharged. That absence cannot be searched for, and it is the thing a quarterly review most needs to know.
The half that is about precedent
There is a second question a partner asks at this point: have we seen this before? In this example the firm has — two comparable deals were declined on the same concentration profile, Harbor in 2023 and Kestrel in 2022.
That is not a fact about Meridian. It is a similarity between entities, and it only exists once the memory holds each deal as a company with facts attached rather than as a folder. It is also the institutional knowledge that walks out when the people who made those decisions leave, which is the whole reason for building any of this.
What would finding this by hand have cost?
Worth being concrete, because the alternative to a memory is not nothing — it is a person, and the person is expensive.
An analyst asked to prepare this review opens the deal folder, reads the March memo, notes the condition, checks the CRM, finds it says approved, and has to decide whether to believe it. Finding the April diligence line requires reading a hundred-page report for something nobody flagged. Establishing that the condition was never discharged requires proving a negative across four months of correspondence. Recalling Harbor and Kestrel requires either having been there or asking someone who was.
Call it most of a day, done by someone who will do it again next quarter from scratch, and who will miss the April line roughly half the time because there was no reason to look for it. Multiply by the number of portfolio companies and the number of quarters. That is the cost being carried today, and it is carried quietly, because the output looks the same either way — a review got written, and nobody could tell what was not in it.
The comparison that matters is not memory against perfection. It is memory against the version of this that a tired person does on a Thursday.
What the answer actually looks like
Asked to prepare the quarterly review, an assistant connected to the memory returns something like: approved in March subject to confirmatory diligence, with no later document recording the condition discharged; concentration raised in three documents over four months; two comparable deals declined on this profile. Each sentence names the document it came from.
The partner then does the only part that was ever theirs: decides. But they decide having been handed, in a few seconds, the assembly that would otherwise have cost a day and would probably have missed the April line entirely.
The figures above are invented, as every worked example on this site is. The shape is not — it is the same shape in credit, legal, tax and audit, with different nouns.
The six kinds of thing a memory raises are on the homepage, and the full private-equity write-up is at for private equity. How it works describes the write path that makes counting possible.