By DiligensHub team · Last updated
What Is AI Due Diligence?
AI due diligence uses AI models to read and check deal documents. The term also covers reviewing a target's own AI systems; this guide is about the first meaning.
In practice it means three kinds of work:
- Extraction. Pulling out parties, dates, amounts and key terms from each document.
- Checking. Testing each document for errors and gaps: totals, signatures, dates and required clauses.
- Comparison. Finding where documents disagree with each other, or with what the seller has claimed.
Most current tools use large language models. These work with the meaning of text, not just exact keywords. That lets them read contract language they haven't seen before.
What Does AI Review Do Well?
AI reads every page with the same attention, and it handles the tedious checks people skip. It doesn't tire on the four-hundredth contract.
| Task | What AI does | Example |
|---|---|---|
| Extracting key terms | Pulls out parties, dates, amounts, renewals and notice periods | Every contract that renews in the next 90 days |
| Checking arithmetic | Recalculates totals and subtotals | A schedule whose rows don't add up to its total |
| Spotting what's missing | Finds unsigned pages, missing dates and missing exhibits | An agreement that cites Exhibit B but doesn't include it |
| Flagging risk terms | Finds change of control, exclusivity and MFN clauses | A customer contract a sale would let the customer end |
| Finding sensitive data | Finds personal data that shouldn't be shared | A payroll file with full employee identifiers |
| Comparing documents | Finds figures and claims that disagree across files | Revenue in the CIM that differs from the audited accounts |
The gain is coverage. A team can check every document, not just the ones it has time for. Reviewers then spend their hours on the findings, not on the search.
Where Must People Still Check?
People must judge what a finding means for the deal, and confirm anything the deal will rely on. AI can be wrong, and it can sound certain when it is.
- Materiality. AI can flag a clause. Only the deal team knows whether it matters at this price.
- Context. A document can be correct alone and misleading beside a side letter the AI never saw.
- Errors. AI models can state things that aren't in the document. Check each finding against its source.
- Poor scans. Faint, skewed or handwritten pages are harder to read. Check important scanned documents by hand.
- Judgment. What to ask for, what to price in, and when to walk away stay with people.
Keep a record of what people checked. If a finding matters later, in a dispute or a claim, the record shows it was reviewed.
A simple rule covers it: AI finds, people decide.
Why Does a Citation to the Source Matter?
A citation lets a reviewer confirm a finding by opening the exact page it came from. Without one, each finding must be found again by hand.
Citations also guard against invented answers. If a tool can't point to where a finding came from, treat it as unconfirmed.
For a spreadsheet, the citation should name the sheet and the row. For a scanned page, it should open the scan, not just the text read from it.
Ask the vendor how citations are produced. A citation taken from the retrieved source is stronger than one the model writes itself.
When you test a tool, click a few citations. Check that each opens the right place, and that the source says what the finding claims.
What Changes for Sellers and Buyers?
Sellers can find and fix problems before buyers see them. Buyers can cover the whole room, and send their experts where it matters.
| Who | Before AI review | With AI review |
|---|---|---|
| Seller preparing a sale | Problems surface when a buyer's advisor finds them | Problems surface during preparation, while there is time to fix them |
| Seller during the sale | New uploads are checked when someone has time | Each upload is checked as it arrives |
| Buyer's first review | Advisors sample the largest contracts first | Every contract is read, and the outliers are listed |
| Buyer's detailed review | Experts search for the clauses that matter | Experts start from a list of flagged clauses, with pages |
| Both sides | Figures are compared across documents by hand | Conflicting figures are listed side by side |
Neither side saves the judgment. Both save the search.
What Are the Risks of Using AI in Due Diligence?
The main risks are leaked documents, unchecked errors and too much trust. Each one has a simple control.
| Risk | Control |
|---|---|
| Documents leave your control | Use a tool whose provider terms bar training and limit retention, in writing |
| One party sees another's documents | Use AI that answers only from what the asker is allowed to see |
| A wrong finding goes unchecked | Require a citation for every finding, and check the ones you rely on |
| A missed issue is assumed absent | Treat AI as a first pass, not proof that a problem doesn't exist |
| Nobody owns the findings | Give each finding an owner and a recorded outcome |
What Should You Ask a Vendor About Data and Training?
Ask where your documents go, who can see them, and whether they train any AI model. Get the answers in writing.
- 1.Which AI models do you use, and from which providers?
- 2.Are our documents used to train any model, yours or the provider's?
- 3.Does the AI provider keep our documents after processing? For how long?
- 4.Where is the processing done, and where is our data stored?
- 5.Can AI be switched off for a deal?
- 6.Does the AI respect permissions? Can a user's question surface a document they can't open?
- 7.Does every finding cite where it came from?
- 8.Can we see which findings were fixed or dismissed, by whom, and why?
Question 6 matters most in a sale with several bidders. An assistant that answers from the whole room could reveal one bidder's documents to another.
How Do You Bring AI Into a Due Diligence Process?
Start on the sell side, where findings stay private. Use AI to check your own documents before any buyer sees them.
- Run AI review while you prepare the room, not after it opens.
- Give each finding an owner, and record why it was fixed or dismissed.
- Decide who sees AI findings. Usually that is the deal team, not the bidders.
- Tell your advisors which checks ran, so they spend their time elsewhere.
- On the buy side, use AI to cover the whole room, then send your experts to what it flags.
Questions
- Can AI replace due diligence lawyers and accountants?
- No. AI finds and sorts issues across many documents. Lawyers and accountants decide what each issue means, negotiate the protections, and sign off. AI changes where their time goes, not whether they are needed.
- Is AI due diligence accurate?
- It is useful for finding issues, but not for relying on unchecked. Treat every finding as a lead to confirm against its source. Test any tool on documents you already know well.
- Is it safe to upload deal documents to an AI tool?
- It depends on the tool's terms. Check that your documents aren't used for training, and how long the provider keeps them. Check that the tool respects your data room's permissions.
- What kinds of documents can AI review?
- Most tools read PDFs, Word files, spreadsheets and scanned pages. Ask how a tool handles scans, images and very long files, because limits vary from tool to tool.