Almost everything that later turns out to be wrong was never actually verified - it was assumed, because it arrived alongside things that were true.
10 min read · Level 3
The usual framing of diligence is that it is how you find out whether somebody is lying to you, and that framing is both flattering and wrong. Deliberate fraud is rare enough that a process designed around it spends most of its energy in the wrong place. What is not rare - what is close to universal - is a seller who genuinely believes their own summary, because they built it from the parts of the business they think about most and have never had a reason to total the parts they do not. The customer concentration in the example was not concealed; it was simply never computed by anyone, including the owner. That has an uncomfortable implication for how you behave in the room. If the failure mode is shared blindness rather than concealment, then an adversarial process is not just unpleasant, it is inefficient - it makes the seller defensive about numbers they have not examined either, and the fastest route to the truth is often to hand them the arithmetic and watch their face. The people who are best at this are not the most suspicious ones. They are the ones who add things up.
Every fact in front of you arrived by one of three routes, and they are not equally reliable.
Somebody told you. That is a management representation: a statement by an interested party, made in a conversation or a document they produced themselves. It may be entirely true. Its reliability is a fact about the person, not about the claim.
You inferred it. Nobody said the business had two hundred customers; the deck said 'over 200 customers' beside a revenue figure and you concluded the revenue was spread across them. That step happened in your head, at speed, and by the time you write your notes it has usually stopped feeling like an inference at all.
Or you checked it against something that was not produced for your benefit - a bank statement, a signed contract, a filed return, a public register, a customer who has no stake in the transaction. That is a primary source, and it is the only one of the three that carries weight on its own.
Diligence is the work of moving items from the first two columns into the third. It is not the work of collecting more of the first two, which is what most of it turns into.
Because the three routes feel identical once they are written down. A page of notes records 'revenue 4,200,000, over 200 customers, 40% gross margin' in a single list, in the same handwriting, with nothing to indicate that the first came from an audited statement, the second from a slide and the third from a conversation on a call. A week later the list is simply what you know.
And because failures cluster almost entirely in the unverified column. A deal that disappoints rarely does so because the checked numbers were wrong. It disappoints because something everyone had been repeating since the first meeting turned out to be nobody's actual finding - a customer count, a renewal rate, a permission that was assumed to transfer. The claim survived because it was never contested, and it was never contested because it was never labelled as a claim.
Educational, not personalised financial advice. Nothing in the Investor Academy is a recommendation to buy, sell or hold anything — what is right for you depends on your own circumstances and the rules where you live.
A small manufacturer. The materials say revenue of 4,200,000 across 'over 200 customers', which reads as a comfortably spread book of business.
Request the receivables ledger - a document produced for the company's own use, not for you - and add up what each customer actually bought. The top three account for 2,520,000. That is 2,520,000 over 4,200,000, or exactly 60.0% of revenue in three relationships. The remaining two hundred or so customers share 1,680,000 between them.
Nothing in the original statement was false. Over two hundred customers is true. The inference - that revenue was spread across them - was yours, and it was the whole basis on which the business looked stable.
Now do the second check, which almost nobody does. Take the bank statements for the same twelve months and total the cash actually received from customers. It comes to 3,780,000 against 4,200,000 of booked revenue, a difference of 420,000, or 10.0% of the year. That gap has an explanation - it might be ordinary timing, a large invoice raised in December, or it might be revenue recognised on work that was never accepted and will never be paid for. The point is not that the gap is damning. The point is that until you ran the totals, you did not know there was a gap to ask about, and the question 'why is 10% of your revenue not in your bank account' is one that can only be asked by somebody who added up two documents that were never meant to be read together.
Two checks. One turned a comfortable inference into a concentration finding; the other produced the single best question anyone asked in the whole process.
The first mistake is treating volume as verification. A data room with four hundred documents in it feels thoroughly checked, and a reader who has opened all four hundred has still only read material the seller assembled. Quantity of disclosure and independence of disclosure are unrelated properties.
The second is accepting a document because it looks official. A spreadsheet exported from an accounting system is still a spreadsheet that somebody could have edited afterwards. The test is not whether a document looks authoritative but whether it originated outside the transaction - a bank, a registrar, a counterparty, a regulator.
The third is asking the seller to confirm the seller's own claim. 'Can you confirm the top customer is only 15%?' produces a confirmation, every time, and adds nothing. The question that produces information is the one addressed to a different source entirely.
The fourth, and the one that quietly does the most damage, is checking hardest where you already doubt. Attention naturally flows to the things that feel uncertain, which means the confident-sounding claims - the ones everybody has been repeating since the first meeting - are systematically the least examined items in the file.
Told, inferred, or checked. One letter beside each. This takes fifteen minutes and is the entire technique; everything else is consequence. Do it before you build any model, because a number that enters a spreadsheet loses its provenance permanently.
Take the claims where being wrong would change the decision, and verify those first regardless of how plausible they sound. The claim you would most hate to be wrong about is usually the one nobody has questioned, precisely because questioning it would be awkward.
Not the source that could confirm it - the one that could contradict it. A customer list is confirmed by the seller and contradicted by the bank. A permission is confirmed by the seller and contradicted by the register. If no independent source exists, that is itself the finding, and it belongs in your notes as one.
Whatever is still in the told or inferred column at the point of decision is what you are actually taking on faith. Write that list out as a short paragraph and read it as though somebody else had handed it to you. Most decisions to proceed survive this; the ones that do not were going to fail later and more expensively.
On every material claim in any private transaction, and on the specific claims a public investment thesis depends on - the ones where being wrong changes the answer rather than the decimal place.
There is a real limit. Verifying every statement in a four-hundred-page disclosure to primary-source standard is not diligence, it is a way of never deciding, and it costs money and goodwill that would be better spent on three checks that matter. The discipline is selective depth, not uniform depth - and choosing where to go deep is the judgement the process cannot do for you.
A shape that appears repeatedly in post-mortems of disappointing private investments is a single load-bearing claim that was stated early, repeated in every subsequent document by every party including the buyer's own advisers, and never traced back to a source. Renewal rates and pipeline conversion are the two most common. The mechanism is ordinary: the figure appears in an initial deck, is copied into the buyer's internal note because it is the only figure available, then into the model, then into the committee paper - and by the fourth appearance it is being cited from an internal document, which makes it look like the buyer's own work.
Trace any number that appears in your own committee paper back to the document it originally came from. A figure cited from your own earlier note has not been verified; it has been laundered.
Create a free account to read the rest and apply it to your own investing.