Can AI help with due diligence document review?

Yes, for the first pass, and that is a bigger share of the job than it sounds. Max spent years at PwC on restructuring work, so this is a process we know from the inside rather than from a vendor demo.

What it does well

Extraction into a schedule. Two hundred contracts, and you need the counterparty, term, notice period, change of control clause and assignment position for each. That is a genuinely tedious task with a right answer, and it is the single strongest use case in the whole process.

Cross-document inconsistency. The share count in the accounts against the one in the cap table. A property listed in the fixed asset register that has no lease in the data room. A director named in one document and not another. Software compares hundreds of documents in a way a person under time pressure simply cannot.

Completeness against a checklist. Telling you on day two what has not been uploaded is worth more than telling you in week three, because the request list goes back sooner.

Making the room searchable. Asking a question and getting the three documents that bear on it, with the passages, saves hours of hunting.

What it does badly

Judgement, and specifically materiality. A model will report a non-standard indemnity in the same tone as a stray typo. Deciding which of forty findings changes the price, or kills the deal, is the work, and it is not the part being automated.

It is also poor at knowing what should be there but is not mentioned anywhere. Absence is invisible to a text search unless someone has already thought to look for it.

The rule that makes it usable

Every extracted fact carries a reference back to the document and the page. Without that, the output is unusable, because nobody can rely on a finding they cannot check, and verifying an unsourced summary takes longer than reading the source in the first place.

That single requirement is what separates a review tool people actually use from a demo that impresses once and then gets ignored.

Settle confidentiality first

Data room material is other people’s confidential information, usually under an NDA with specific terms about where it may go and who may see it. Before anything is processed, you need to know which provider is handling it, whether that provider retains inputs, and whether your NDA and the data room terms permit it.

This is normally solvable, and the tier matters more than the brand. It is a conversation to have before the deal starts, not while the clock is running.

What it realistically changes

The first pass gets faster and more complete, and the team spends its time on the findings rather than on locating them. That is a real gain on a mid-market deal and a substantial one on a portfolio review, where the same analysis repeats across many targets.