# The Number on the Page Isn’t the Asset
## Founder’s Legend | Touch Stone Publishers
—
I’ve been in rooms where a number on a page convinced experienced people to do something they shouldn’t have done. Not because they were careless. Because the number was real, the process that produced it was legitimate, and nobody in the room had a specific reason to doubt it.
That is the nature of the risk I want to talk about today.
Private equity has spent the last two years building AI into its due diligence workflows. The tools are genuinely impressive. They parse thousands of documents in hours. They surface patterns that human reviewers would miss or take weeks to find. They reduce the cost and time of VDR review in ways that are operationally significant. I understand why firms adopted them. The capability is real.
Here is what I have been watching, though, and what I think matters more than the capability: what we do with the number the tool produces.
When an AI-assisted due diligence platform reviews a virtual data room and produces a clean summary — no red flags on the financial documents, no anomalies in the contracts, positive pattern match on the AI capability descriptions — that summary is accurate about the documents it reviewed. It is not an assessment of the underlying reality those documents describe. The AI tool found what was in the documents. It has no method for finding what the documents did not record.
The DOJ established this year, in the Nate Inc. indictment, that a company raised $42 million by documenting a neural network capability that did not exist in their production system. The documentation was coherent. The presentation was credible. The documents were real documents describing a company that was not real. Every due diligence tool that reviewed the VDR found what was in the VDR. None of them found what was not there.
I am not making an argument against AI due diligence tools. I am making an argument about what we have allowed those tools to substitute for.
The hard work of an acquisition has always been the assessment of the people and the institutional knowledge behind the asset. What do they actually know? What have they actually built? Who is carrying the understanding that makes the thing work, and why are they still here, and what happens when they leave? These are human questions. They require human inquiry. They cannot be answered by pattern recognition in documents, however sophisticated the pattern recognition is.
The governance failure I see forming in the PE market is a version of one I have watched destroy value in organizations of every kind: leaders who accept the output of a process as a substitute for the judgment that the process was designed to support. The AI due diligence tool produces a clean summary. The investment committee approves a 3.2x premium. Nobody in the room is carrying the accountability for asking whether the AI in the acquired company actually does what the documents say it does.
That accountability belongs somewhere. In the organizations that fail, it belongs to everyone in general and nobody specifically. In the organizations that succeed, it belongs to a named person who is required to answer for it — before the deal closes, not after the impairment shows up on the portfolio report.
What this requires is not a more sophisticated tool. It is a more honest conversation about what the tool can and cannot do. It can find patterns in documents. It cannot validate the reality behind the documents. It can accelerate a review. It cannot substitute for the judgment that the review is supposed to produce.
The leaders I respect most are the ones who resist the seduction of a clean process. They do not feel better because the AI found nothing. They ask what the AI was designed to miss. They hold the tension between the number on the page and the reality the number is supposed to represent. That tension is uncomfortable. It is also the only thing that protects the people whose capital they are stewards of.
Every acquisition I have seen go wrong had a moment where a clean process was accepted as a substitute for a hard question. The question was available. The process produced comfort. The comfort was chosen.
The question in AI-enabled acquisitions today is a specific one: does this AI actually do what they say it does? The algorithmic audit that answers that question is not a technical add-on. It is the governance act that closes the gap between the document and the reality. It is what fiduciaries do when they take seriously what they owe to the people behind the capital they are managing.
The test of a governance model is not whether it works while everything is going well. The test is whether it was designed to find what no one wanted to find. That is what the people behind the capital are counting on.
That is what they are owed.
Glenn E. Daniels II
Touch Stone Publishers
—
*WP Category: Founder’s Legend (547) | TSP_2026-020 | May 2026*