A data room used to be read by people. A junior analyst worked through it over a fortnight, flagged what looked odd and passed a summary up the chain. By the time a partner looked, the story had already been shaped by whoever read it first.
That first reader is now a model. It reads every page, checks every number against every other number and compares every claim with the public record, in minutes. Whatever the process is called, a raise, an acquisition, a credit facility or a large customer's procurement review, if there is diligence, there is AI in it.
How far this has gone
Datasite surveyed 1,000 senior dealmakers in 27 countries in March 2026. Half said AI is now regularly embedded in their due diligence, and diligence was the stage where they reported the highest return on it.1 Bain's 2026 M&A report found that 45% of the more than 300 M&A executives it surveyed rely on AI tools, and that adoption more than doubled during 2025.2 In L.E.K.'s survey of 100 US buyout professionals, research and diligence synthesis were where investment teams reported using AI most.3
The people on the other side of your table are not waiting for permission. Their tools are already in the room.
5 flags · 41 questions sent back · 14 weeks in diligence
What the machine finds first
Models are good at exactly the things a tired person misses on page 40. In the companies we prepare, the first pass usually turns up the same handful of problems.
- Revenue in the deck that does not match the financial model, or a model that does not match the management accounts.
- Growth assumptions with no stated source, or a cited source that does not contain the figure.
- An employee share scheme left out of the pre-money cap table, so the price per share is wrong.
- Named customers with no signed contract behind them.
- Titles and histories that do not match company registries, professional profiles or press coverage.
- Intellectual property that was never formally assigned to the company.
None of these is fatal on its own. Found by the other side, each one becomes a question, a delay or a reason to reopen the price. Found together, they set the tone for the whole process.
Why it costs more than it used to
When a person found a discrepancy, they asked about it in a meeting and moved on. When a model finds it, it goes into a written list of questions that grows with every document. Teams then spend weeks answering questions that an afternoon of preparation would have removed, and the investor or buyer has time to wonder what else is in there.
It works the other way too. A file that clears the other side's tools on the first pass draws fewer questions, keeps the conversation on the business rather than the paperwork, and shortens the time to completion.
How to prepare
The method is simple to state and tedious to do, which is why it is usually skipped.
- Run the same read yourself. Put the full data room through the same class of tools an investor or acquirer will use, before anyone outside the company sees it.
- Keep a claims register. Every claim in the deck, the memo and the website, next to the document that supports it.
- Reconcile the numbers. One revenue figure, one customer count, one cap table, everywhere they appear.
- Fix what can be fixed, and disclose the rest. A known issue with an explanation and a date reads as competence. The same issue found by the other side reads as concealment.
- Read it again before every release. Data rooms change during a process, and each new version gets the same check.
Where it applies
Raises are the obvious case. The same preparation shortens acquisitions on either side, credit applications, listings and enterprise procurement, where a large customer's risk team now reads your security and financial documents with the same tools.
If our AI found it, so will theirs. The only questions are who finds it first, and whether you have an answer ready when they do.