Underwriting has changed, and most management teams are still preparing for the old version of it.
For years, a value creation story lived in the CIM. You asserted a moat, a growth driver, a sticky customer base, and the buyer’s job was to test those claims well enough to get comfortable. The claim came first, the proof came later, and a confident narrative backed by a clean deck earned the benefit of the doubt. That benefit of the doubt is gone.
Across the market, AI has shifted PE underwriting toward competitive moats, data advantages, and embedded workflows that buyers can verify. The operative word is verifiable. Not plausible. Not well-presented. Verifiable. And the same AI that pushed underwriting in this direction is the tool buyers now use to do the verifying, faster and deeper than a diligence team could a few years ago.
This matters more now because of where pricing sits. Buyout multiples have been at or near records, and when a buyer pays at the top of a range, the investment committee wants proof the thesis is real before the wire goes out. High prices raise the evidentiary bar.
The four tests, and why three of them are data questions
The new underwriting frame comes down to four things a buyer wants to confirm before they pay. Is there a real moat? Is there a data advantage the business actually owns and can compound? Are the workflows embedded enough that customers cannot easily leave? And can the headline financials be traced to source?
Three of those four are data questions before they are anything else.
A moat is only as defensible as the evidence behind it. If you claim switching costs, the buyer wants retention by cohort, expansion within the base, and what happens to accounts that try to leave. A data advantage is, by definition, a data question. You either have proprietary data with the lineage and the lift to show it, or you have a marketing line. Embedded workflows show up in usage data and depth of integration, not in a slide that says “mission critical.”
The fourth test, clean financials, has always been a data question. What changed is that it no longer stands alone. The financials are now the floor, and the moat, the data advantage, and the workflows are where the multiple is won or lost.
What AI actually changed in the room
Buyers did not get more skeptical. They got more capable. A diligence team used to sample. They would pull a slice of contracts, a quarter of transactions, a handful of customer records, and extrapolate. Sampling left room for a confident narrative to survive, because nobody could check everything in the time. AI removed that constraint. A buyer can now run the full contract set instead of a sample, reconcile the CRM against the GL across every record, read the entire support history to test whether a product is genuinely embedded or merely installed, and decompose growth into pricing, volume, new logos, and expansion without waiting three weeks for the finance team to build the cut by hand.
When the analysis covers everything rather than a sample, the gap between what you assert and what your data actually says becomes visible in days. I wrote about this in AI-powered due diligence is raising the bar. The things you could once smooth over with a good story now get read in full.
Here is what that looks like in the room. A management team I watched go through a process presented a clean retention story. Logo churn under 5%, strong net revenue retention, a base they called sticky and growing. The deck was good. In an earlier era the buyer would have pulled a sample of accounts, confirmed the shape, and moved on.
Instead the diligence team loaded the full customer ledger and the billing exports and ran the retention calculation themselves. The number that came back was roughly double, because the company had been netting churned accounts against new logos inside the same segment and reporting the blend as retention. Nobody set out to mislead. The CFO believed the figure, because that was the number the internal report had always shown. But the report and the records did not agree, and the buyer found it in two days. The retention claim did not just get marked down. It became the thing the buyer pressure-tested everything else against.
This is why “verifiable” is the operative word. Verification used to be expensive and partial, so assertion did a lot of the work. It is now cheap and complete, so assertion does almost none of it.
Assertable versus verifiable
The distinction is the whole game. An assertable claim is one you can say. “Our data is a real competitive advantage.” “Churn is low.” These sound fine in a management presentation and cost nothing to make.
A verifiable claim is one the buyer can trace to source and confirm independently. “Net revenue retention is 118% over the last eight quarters, here is the cohort table, the definition every system agrees on, and the query that produces it.” “Gross churn is 4%, reconciled between the billing system and the GL, and here is the bridge.” Same underlying business. Completely different outcome in diligence.
The data advantage claim is where the gap opens widest, because it is the easiest to assert and the hardest to prove. Take a company that tells the buyer its pricing engine runs on a proprietary dataset no competitor can replicate. That is a strong line, and it supports a premium. Now trace it. The buyer asks where the data lives and finds it spread across a transactional database, a few analyst spreadsheets, and a vendor feed the company licenses on annual renewal. A third of the “proprietary” signal turns out to be purchased data any competitor can buy. Nobody ever ran the counterfactual that would prove the engine improves pricing. And if the lead analyst leaves, the logic lives in one person’s notebook.
The company might have a real advantage. But the claim as presented cannot be traced to source, so under the new standard it does not earn the premium it was meant to earn. The seller asserted a moat. The buyer could only verify a workflow held together by one person and a renewable license, and the multiple gets written for what was verified, not for what was claimed.
How one bad claim taxes the whole CIM
Most teams assume a single failed claim costs them that one claim. It costs them far more, and it is worth being precise about how the damage spreads. This is the dynamic I described in how buyers test data accuracy in diligence.
A CIM is a sequence of claims a buyer reads in good faith. Early on, the buyer assumes the numbers are sound and spends diligence confirming the shape of the business rather than auditing every figure. That assumption is what lets a deal move at speed. The first claim that fails to verify breaks it. Now the buyer knows the deck contains at least one number that did not survive contact with the data, so the rational response is to stop extending good faith and verify everything from scratch. That costs time, and the buyer assigns the cost back to the seller as a lower price or tougher terms.
The second-order effect is worse. Once trust is gone, ambiguity stops breaking in the seller’s favor. A borderline judgment that would have gotten the benefit of the doubt now gets challenged. A single number that cannot be traced to source converts a confirmatory diligence into an investigative one, and the deal that would have closed in 60 days on a trusted deck grinds for months instead.
Why this widens the gap between assets
This change does not affect all companies equally. It sharpens the divide I have written about in the bifurcation between premium and stalled exits.
For an A-asset, AI-driven verification is good news. A company whose claims are already traceable to source benefits when the buyer can check everything, because everything checks out. Full verification turns a strong narrative into a confirmed one and compresses diligence.
For a B-asset, the same tooling is exposure. The claims that used to survive a sampled review now meet a full read, and the gaps surface early. The deal does not necessarily die. It reprices, it slows, and the leverage moves to the buyer. The companies most hurt by AI in diligence are the ones whose value story was always more assertable than verifiable. They got away with it when verification was partial. They do not get away with it now, and AI is making the cliff steeper.
How to make your value story verifiable
The work is not glamorous, and it is not a twelve-month transformation. It is the discipline of making sure every claim you put in front of a buyer traces to a single source with a single definition.
Start by listing the claims your value creation story actually rests on. The moat, the data advantage, the embedded workflows, the growth drivers, the retention numbers. Usually a handful, not forty. For each one, ask the question the buyer’s tooling will ask. Where does this number come from? Do two systems agree on it? If a tool read every record instead of a sample, would the story hold? If you cannot answer that, the claim is a liability the moment it reaches diligence.
Then close the gaps. Reconcile the systems that disagree, settle on one definition per metric that finance, sales, and operations all sign, and build the lineage for any data you call proprietary, so the answer to a data request is a file, not a three-week project.
Our Buyer Scorecard is a useful way to grade yourself before a buyer does. It walks the categories a buyer uses to evaluate your data in diligence, and the honest version of that exercise tells you which claims are verifiable and which are still just stated.
Say less, prove more
The old playbook rewarded the confident assertion. The new one rewards the claim you can trace to source on demand. AI did not invent that standard, but it made it enforceable. Stop optimizing the narrative and start optimizing the evidence underneath it. Walk into the next process having already asked your own data the questions the buyer’s tooling will ask. The teams that do this win the multiple. The teams that keep asserting find out, in diligence, that the benefit of the doubt is no longer on the table.
For a weekly brief on PE exit dynamics, data readiness, and what buyers actually test in diligence, subscribe to Inside the Data Room.