What the technology does well
Volume review. A data room can hold a large number of documents. Software can read through them, sort them by type and pull out the clauses, dates, parties and amounts a reviewer would otherwise hunt for by hand.
Entity matching. The same company appears under a trade name, a registered name and a transliteration; the same person appears with and without a middle name. Matching these records across documents is repetitive work that software does consistently, and it shows which names belong together.
Red-flag surfacing. A pattern such as a change-of-control clause, a guarantee that sits outside the main agreement or a date that contradicts another document can be flagged for a person to look at. A flag is a prompt to read, not a finding.
What it does not do
Materiality. Whether a clause matters depends on the transaction, the price, the client's appetite for risk and the other terms on the table. That is a judgment about this deal, and a tool that has not been told the deal cannot make it.
Context outside the documents. A missing document, a counterparty's reputation, a conversation with management: none of these are in the data room, and none can be flagged from it.
Responsibility. Behind a reviewed deliverable stands a person or a team. Software that produced a first draft is a tool in that process, not a party to it.
Where the sign-off sits
In well-run setups the pattern is usually the same: technology extends reach, experts lead. Each finding links back to the document and passage it came from, so a reviewer can test it instead of trusting it. A finding should be traceable to its source.
How to read a claim about AI in due diligence
Ask four things. What was read, and how much of it. How findings are tied to their sources. Who reviews them before they leave the building. And what the process does when it is unsure: a good one says so, and does not fill the gap.
A claim that does not answer these is marketing. One that does is a description of a process you can check.
Questions about AI in due diligence
- What does AI do in a due diligence review?
- It reads documents at volume, matches the same entity across documents, and flags patterns such as unusual clauses or contradicting dates for a reviewer to examine.
- Can AI decide whether a finding is material?
- No. Materiality depends on the transaction, the price and the client's risk appetite. A reviewer makes that call; the tool supplies the material to make it on.
- Why does each finding need a source link?
- So that a reviewer can open the document and passage it came from and test the finding, instead of relying on it as stated.
- Who is accountable for the conclusion?
- The experts who review and release the work. Technology is part of the process; accountability stays with people.
- Is AI-assisted review the same as an AI-native approach?
- Not necessarily. AI-assisted means a tool is added to an existing process. AI-native means the process is built around it, with human review designed in at defined points.
For a review of a defined project or body of material, contact CRUNCH. You can also return to the newsroom or meet the senior team behind the work.