Diligence used to take the numbers on trust and test the business around them.
That has flipped. Buy-side teams now treat the data itself as a thing to be verified, not a given to be accepted. They are no longer asking only whether the revenue is real. They are asking whether the system that produced the revenue number agrees with the system that produced the customer count, and whether the KPI in the board deck means the same thing it meant four quarters ago.
This is the part of AI-powered diligence that sellers underestimate. The screening compression from two weeks to two days gets the headline. The quieter change is what buyers do with the time they save. They spend it cross-checking your numbers against each other, line by line, in ways a human team never had the bandwidth to attempt.
If you are preparing for exit, you should know exactly what those tests are. Then you should run them on yourself.
Why accuracy became a workstream
For most of the last two decades, data accuracy was assumed inside the quality-of-earnings analysis. The QoE provider normalized EBITDA, checked the adjustments, and signed off. Buyers read the report and moved on. The underlying records, the CRM, the billing system, the general ledger, the contract files, were rarely reconciled against one another.
Two things changed. Buyers got tools that make reconciliation cheap. And a run of deals taught them that the gap between a clean QoE and a messy operating reality is where the post-close surprises live. A QoE can be technically correct and still sit on top of source data that does not agree with itself. I wrote about that gap in the QoE data problems post. The QoE tells you the number was calculated correctly. It does not tell you the inputs were consistent.
So accuracy testing became its own workstream. Not a footnote inside the financial review. A deliberate set of reconciliations the buyer runs to decide how much to trust everything else you have told them.
The four tests buyers run
These are the checks I see most often on the buy side now. None of them is exotic. All of them are unforgiving.
Billing to general ledger reconciliation
The buyer takes the revenue your billing or subscription system produced and reconciles it to the revenue posted in the general ledger. They are not checking whether the total matches at the year-end level. The auditors already did that. They are checking whether it matches at the customer level, the month level, and the product level.
This is where deferred revenue, manual journal entries, and “we fixed that in a spreadsheet” come to the surface. When the billing system says a customer paid for twelve months and the GL recognized it over nine, that is a question. When a cohort of invoices lives outside the billing system entirely because they were negotiated by hand, that is a bigger one. The reconciliation does not care that you have an explanation. It only shows the buyer where the seams are.
Customer counts across systems
Ask three systems how many customers you have and you will often get three answers. The CRM has one number, the billing system has another, and the board deck has a third that someone built for a fundraise and never reconciled back.
Buyers now pull all three and compare them. The differences are rarely fraud. They are duplicate records, churned accounts that were never deactivated, parent-child entities counted one way in sales and another way in finance, and trial accounts that drifted into the active count. Each gap forces a conversation about which number is real. And every time the answer is “it depends what you mean by customer,” the buyer learns that your operating metrics are softer than your pitch.
Revenue recognition consistency
The buyer checks whether you recognized revenue the same way across the whole period they are examining. A policy that shifted mid-period, a subsidiary that booked things differently, a large contract that got bespoke treatment, all of it shows up when you line the periods next to each other.
This is the test that punishes growth-by-acquisition stories hardest. When you have absorbed three companies, you have very likely absorbed three revenue recognition habits. If you never standardized them, the buyer’s analysis will find the inconsistency before your CFO can frame it. The point is not that you did anything wrong. The point is that an inconsistent policy reads, to a buyer, as an unmeasured risk.
KPI definitions that move between decks
This is the one sellers never see coming. Buyers now pull several quarters of board decks and track whether your headline KPIs were defined the same way each time. Net revenue retention, gross margin, CAC, active users, pipeline. They check whether the definition held.
It usually has not. A definition gets quietly adjusted to tell a better story in a tough quarter, then nobody changes it back. A metric gets calculated by a different analyst with a different assumption. The number on slide four in Q1 and the number on slide four in Q3 use different denominators, and nobody flagged it because each deck was built in isolation.
When a buyer catches a KPI that shifted definition between decks, two things happen. The specific metric loses credibility. And every other number in the deck inherits a discount, because if that one moved without disclosure, what else did.
Why the tooling matters
A human diligence team could, in theory, run all four of these tests. In practice they almost never did, because each one is tedious and the clock was always against them. They sampled. They checked the top customers, the biggest contracts, the headline metrics, and trusted that the rest held.
AI tooling removes the sampling. It ingests the full billing export, the full GL, the full CRM, and the full set of board decks, and it reconciles everything against everything. It does not get bored at row 40,000. It does not assume the small customers behave like the big ones. As I covered in the AI diligence post, the tools flag the anomaly that only appears when you cross-reference twenty documents, and they do it in hours.
That changes the seller’s exposure in a specific way. The inconsistency that used to survive diligence because nobody had time to find it now gets found on day one. The buyer surfaces a question you have not prepared for, during the period when you have the least slack to answer it. The asymmetry is real. The buyer can find the gap faster than you can explain it.
Run the tests on yourself first
The defense is straightforward to describe and uncomfortable to do. Run the buyer’s tests on yourself, before the data room exists, while you still control the timeline.
Reconcile billing to the GL at the customer and month level, not the annual total. Pull customer counts from every system that holds one and force them to agree, or document precisely why they differ. Lay your revenue recognition policy across the full period and find the places it shifted. Take three years of board decks and check every headline KPI for a definition that moved.
You will find things. Everyone does. The advantage is not that your data turns out to be perfect. The advantage is that you find the gaps first, fix what can be fixed, and prepare the explanation for what cannot. A documented note that says “we standardized revenue recognition across the acquired entities in Q3 2025, here is the bridge” is worth far more delivered upfront than extracted under questioning.
This is what reverse due diligence means in practice. You commission the same scrutiny a buyer would, on your own timeline, and you walk into the room already holding the answers to the questions the buyer’s tools are about to generate.
What clean accuracy buys you
The seller who has run these tests behaves differently in the room. When the buyer’s analysis flags a discrepancy, the seller already has the bridge. When the buyer asks which customer number is real, the seller has one number and the reconciliation behind it. When the buyer checks the board decks for KPI drift, the definitions hold, because they were locked before the process started.
That is the difference between a tightening bid and a loosening one. Consistent data lets the investment committee move with confidence. Inconsistent data multiplies follow-up requests, extends the timeline, and invites the structure that protects the buyer at your expense. Earnouts and holdbacks are priced against uncertainty, and uncertainty is exactly what a failed reconciliation creates.
Buyers now test data accuracy because the tools make it cheap and the lessons made it necessary. The tests are knowable. The only real question is whether your numbers agree with each other before the buyer checks, or after.
Run the reconciliation first. It is your number to get right before it becomes their question to ask.