Guide · Published 2026-10-18
AI due diligence software: what it checks, what it cannot, and how to evaluate it.
AI due diligence software reads the documents of a deal, such as the CIM, financial statements, tax returns and contracts, and reports what does not hold together: figures that differ between documents, claims that a table contradicts, and information that is missing. It does not replace the accountant, lawyer or lender who signs off; it changes what they start from.
What it is
In an acquisition, diligence is the work of testing what the seller says against the documents that are supposed to prove it. Most of that work is reading and cross-checking. AI due diligence software automates the reading and the cross-checking, and leaves the judgment to the buyer and their advisors.
It is a different category from a virtual data room. A data room stores documents and controls who can open them. Diligence software opens them. The two are often used together; see Deal OS vs a virtual data room.
What it checks
These are the checks a buyer can reasonably expect from software that reads documents.
- Tie-out: the same figure, such as revenue or EBITDA, located everywhere it appears and compared across the CIM, the financial statements and the tax return.
- Internal contradictions: two passages that cannot both be true, shown side by side.
- Add-backs: each seller adjustment itemized, quoted to its page, and given a first judgment on whether a buyer would accept it.
- Gaps: documents the data room refers to but does not contain, and questions the documents leave open, turned into a request list for the seller.
- Management claims: statements from a call compared with what the documents support.
- Citations: each finding tied to a document and a page, so a human can open the source in seconds.
What it cannot do
For SBA 7(a) acquisitions the human step is also a requirement. Under SBA SOP 50 10 8.1, effective 1 October 2026, certain change-of-ownership deals with a purchase price of $3 million or more need an independent quality of earnings analysis prepared for the lender (summary by AAFCPAs, 14 September 2026; read the SOP itself before relying on it). Software that reads documents is a first pass before that engagement, not a substitute for it.
- It cannot see what is not in the documents. A fraud that never appears on paper stays invisible.
- It does not give a legal, tax, accounting or investment opinion. An independent quality of earnings report, a lawyer's review of contracts and a lender's credit decision remain separate, human work.
- It can misread a poor scan, a table split over pages, or a figure that depends on context. That is why every finding should open to the source passage rather than ask to be trusted.
- It cannot tell you what to pay. It can show that the number you were given does not tie, which is an input to price, not a price.
How to evaluate one: criteria that do not depend on a vendor's claims
Marketing pages for these tools tend to look alike, so test the tool instead of reading the page. The table gives the criterion, the question to ask, and a way to check it yourself.
| Criterion | Question to ask | How to test it yourself |
|---|---|---|
| Citations | Does every finding point to a document and a page? | Open ten findings. Does each source passage say what the finding says? |
| Unsupported claims | What happens when the model states something the documents do not support? | Ask whether such statements are removed or shown, and whether you can see what was removed. |
| Cross-document checks | Does it compare figures between documents, or summarize each one alone? | Run a deal where you know the CIM and the tax return disagree. Does it find it? |
| Arithmetic | Are totals and ratios computed or generated as text? | Recompute three figures by hand. |
| Gaps | Does it say what is missing, not only what is present? | Remove a document you know matters and see whether the output notices. |
| Limits | Does the output say what it is not? | Look for a plain statement that it is a first pass and not an opinion. |
| Confidentiality | Where are documents hosted, are they used for training, can they be deleted, is there an NDA? | Read the security and data processing pages. Check your NDA with the seller for rules on third-party tools. |
| Format of the result | Can you share it with a lender or partner with the citations intact? | Share a sample with someone who was not in the session. |
| Cost model | Per deal, per seat, or a quote on request? | Find the price page. A tool that hides cost until a call costs you the call. |
Where Deal OS fits
Deal OS is self-serve diligence software for buyers of small companies: search funds, independent sponsors and individual acquirers. You upload a CIM and the rest of the data room and get a brief in which each claim cites its source page. A verifier checks each claim against the source text and discards the ones it cannot tie to a passage; the brief lists what was discarded.
Its checks are the ones listed above: cross-document tie-out, contradictions, an add-back review, an SBA 7(a) coverage check run as arithmetic, a management-call cross-check and an investment committee memo drafted from the verified findings. It is a first pass and says so in the output.
You can see the real output without signing up: the sample brief and the sandbox are Deal OS runs on synthetic deals. Per-deal and yearly plans are on the pricing page.
Worked example from the sample brief
What it looks like on a real brief: five things the sample brief shows, quoted from it.
- A claim contradicted by the document's own table. The CIM says on page 8 that no customer is more than 15% of revenue. Page 9 lists the top account (Northgate Property Group) at $1,366,000 against $6.21M of revenue, which the brief computes as 22.0%. The brief shows both passages word for word, each with its page, and marks the page 8 statement as false on the CIM's own numbers.
- A headline figure that two documents state differently. Trailing twelve month revenue is $6.21M on page 6 of the CIM and $5.95M as gross receipts on the tax return. The brief shows the $260K gap side by side and says to reconcile it before applying any multiple.
- A definition that changes what a number means. Page 7 says about 78% of revenue is recurring. Page 22 defines recurring as all revenue from accounts with an active inspection agreement, including repair and call-out work. The brief flags the two passages as inconsistent and asks for a revenue cut that separates fixed inspection billings from variable work.
- Add-backs sorted by whether they are likely to survive. The seller presents adjusted EBITDA of $1.42M. The brief itemizes the add-backs and rates each one; $90K of them are rated questionable or unlikely, which gives $1.33M that holds up on the documents. The arithmetic is computed, and the ratings are a first judgment for a human to confirm.
- A statement the documents do not support, thrown out. While drafting, the model proposed that the company has no outstanding litigation. No passage says so, so the verifier discarded it and the brief lists it in a verification log, with a request for a 10-year litigation history instead.
The example is Sentinel Fire & Safety, Inc., a synthetic company built to show the product. Nothing in it describes a real business, and the same output shape applies to a real data room. See the whole thing at /sample-brief.
Questions
Frequently asked questions.
What is AI due diligence software?
It is software that uses AI to read the documents of a deal and report discrepancies, gaps and unsupported claims, usually with a citation to the source page. It supports the buyer's own judgment and the work of their advisors.
Can AI due diligence software replace an accountant or a lawyer?
No. It performs the repeatable reading and cross-checking. Accounting, tax and legal opinions, and a lender's credit decision, remain with the professionals who are accountable for them.
How can I tell if its findings are reliable?
Open the source passage behind each finding and read it. A reliable tool makes that easy, discards statements it cannot tie to a passage, and shows you what it discarded.
What documents does it need?
At minimum the CIM and the financial statements. Tax returns, an add-back schedule and management-call notes let it compare more sources, so more discrepancies can be found.
Is it safe to upload a confidential CIM?
Check four things before uploading: where documents are hosted, whether they are used to train models, whether you can delete them, and whether your NDA with the seller allows third-party tools. Read the vendor's security and data processing pages for each.
See it on a deal.
Open a real brief on a synthetic deal without an account, or check how a single deal is priced.
