Can AI help us run due diligence on a corporate transaction?

Yes, for the part of due diligence that is reading rather than deciding. An AI assistant can work through a data room, pull out the terms you have asked for, point at the page each one came from, and tell you where a document you expected is not there. What it cannot do is decide which findings matter to this buyer on this deal, how to price a risk, or what goes in the disclosure letter.

The part of due diligence that is really reading

Most of the hours on a mid-market due diligence exercise go on volume, not difficulty. Someone has to open every contract in the commercial folder, find the term, the termination rights, the assignment and change of control provisions, the liability cap and the governing law, and write them into a schedule. Someone has to check whether the leases in the property folder match the list in the disclosure schedule. Someone has to notice that four of the eleven key customer contracts are unsigned.

That work is mechanical, it is where tired people make mistakes, and it is work an AI assistant can do first pass. The useful framing is that AI produces the raw material for the report and the fee earner produces the report.

Tasks that transfer well:

• Building an index of what is actually in the data room, folder by folder, against what the request list asked for.
• Extracting named clauses from a set of similar contracts into a consistent schedule.
• Flagging contracts that contain change of control, exclusivity, most favoured nation or unusual indemnity wording.
• Pulling dates out of leases, licences and finance documents so they can be checked against the warranties.
• Answering targeted questions across the whole room: which agreements are governed by Scots law, which supplier contracts expire within twelve months.

Tasks that do not transfer: deciding whether a finding is material, deciding what to escalate to the client, negotiating the position, and signing the report.

Gaps matter more than findings

On a due diligence exercise, the thing that causes trouble later is usually the document nobody saw. A seller's data room is assembled under pressure and is almost never complete. Deals go wrong on the subsidiary that was never mentioned, the guarantee that sits outside the finance folder, the lease that was uploaded as an unreadable scan.

So when you use AI here, weight your questions towards absence. Ask what is missing against the request list rather than only asking what the documents say. Ask which uploaded files could not be read. Ask which contracts referred to in other contracts do not appear anywhere in the room.

A tool that answers "the papers do not contain a signed version of this agreement" is doing more for you than one that produces a confident summary of an unsigned draft without saying it is unsigned.

Keeping the report defensible

A due diligence report is relied on by a buyer making a decision with money attached. It needs to be verifiable line by line.

Four habits keep it so:

1. Insist on page level citations. Every entry in the schedule should name the document and the page. If you cannot click through from the schedule to the source, you cannot check it quickly, and if you cannot check it quickly you will not check it at all.
2. Spot check by sampling. Take a proportion of the extracted entries across every category and verify them against the document. If the sample is clean, you have a basis for trusting the rest. If it is not, you widen the sample.
3. Verify everything that goes in the executive summary by hand. The findings the client will actually read and act on get checked individually, no exceptions.
4. Record who checked what. On a deal with a team of five working to a deadline, the record of who verified which section is what protects you if a point is challenged afterwards.

The standard is not that AI did the work carefully. It is that a qualified person took responsibility for the output.

Confidentiality in a data room

Due diligence papers are commercially sensitive, frequently covered by a non-disclosure agreement with terms about who may see them and where they may be held, and often contain personal data about the target's employees.

Before any of it goes near a tool, check three things. Where is the information held and processed. Whether it is used to train anything outside your firm. Whether the non-disclosure agreement or the engagement terms restrict processing by third parties, which some seller side agreements do in explicit terms.

Inside the firm, access should follow the deal team. On transactional work there are often information barriers in place, and a tool that lets anyone in the firm read the data room because of their seniority defeats the barrier you have built.

Where Alesis fits

Alesis is an AI assistant for UK law firms, used through the browser with one conversation for a matter. It answers questions about a matter from the matter's own papers and names the page each answer came from; where the papers do not say, it says so, and any page it could not read is flagged rather than skipped. The firm's information is held in the UK and processed only in the UK and the EU, kept apart from every other firm, and inside a firm people see only the matters they are on. It prepares drafts for a qualified person to review and sign off, and it does not provide legal advice.