A buyer described on Searchfunder how he connected an AI assistant to his newly acquired company's phone records, email, calendar and QuickBooks, and had it audit every sales lead from the previous two months while he worked on something else. It came back with findings he then worked through one customer at a time.
Buyers now use the same approach in due diligence, before they own anything. The last step is the important one. The AI finds things to look at. A person decides what they mean.
Where AI genuinely helps in due diligence
Reading everything, not a sample. A human reviewer samples contracts and invoices because there are too many to read. An AI tool can read all of them and list the ones with unusual terms, such as change-of-control clauses, termination rights or pricing that does not match the rest.
Classifying the ledger. A general ledger export can be tens of thousands of lines. AI is good at grouping entries, finding personal-looking spend, and pulling out the items that look like add-backs, so the reviewer starts from a shortlist.
Comparing records. Customer lists against invoices, invoices against deposits, payroll against headcount. Mismatches between records are where problems hide, and comparing is tedious work AI does well.
Summarising. First drafts of customer summaries, supplier summaries and contract abstracts that a reviewer edits rather than writes.
Where it fails
Judgement. Whether an expense is genuinely one-off, whether an owner's salary should be replaced at market rate, whether a customer's falling orders are temporary. These are judgements about the business, and AI will make them confidently whether or not it has the information to.
Anything outside the data room. AI can only read what it is given. A customer about to leave, a dispute not yet in writing, an employee planning to resign: none of it is in the files.
Errors built to look normal. Fraud and serious manipulation are designed to pass a quick review. An AI that is good at finding unusual items is, by design, weaker at finding ones made to look usual.
Consistency. Ask a general-purpose AI the same question twice about pasted numbers and you can get two different answers. A diligence finding that changes when you ask again cannot go to a lender.
Handle the seller's data carefully
Data room files are confidential and usually covered by an NDA. Before uploading anything to an AI tool, check whether the NDA permits it, where the tool stores data, whether it is used for training, and who else can access it. Business plans for AI tools often handle this differently from consumer ones. If in doubt, ask the seller.
Make the findings reliable
Treat AI output as a list of leads, not conclusions. For each finding, go back to the source document or ledger entry and confirm it. Tools that connect to the accounting system directly and show the entries behind each number are easier to verify than ones working from pasted spreadsheets.
Write down what was checked by AI and what was checked by a person. A lender or investor reading the diligence report will want to know.
What AI will not do for you
It will not replace the quality of earnings, the proof of cash, or the conversations with customers and staff that tell you whether the business will hold together after closing.
And it will not take responsibility. If a finding is wrong, the buyer carries the consequence, not the software.
Frequently asked questions
Can AI be used for due diligence on a small business acquisition?
Yes, for reading contracts and ledgers at volume, flagging unusual items and comparing records. The findings still need to be checked against the source documents by someone who can judge what they mean.
What can AI not do in due diligence?
It cannot make reliable judgements on add-backs or one-off items, see anything outside the data provided, or reliably detect errors designed to look normal. It can also give different answers to the same question.
Is it safe to upload data room files to ChatGPT or Claude?
Only if the NDA allows it and the tool's data handling suits confidential material. Check where data is stored, whether it is used for training and who can access it, and ask the seller if unsure.
How do you make AI diligence findings reliable?
Confirm every finding against its source document or ledger entry, record which checks were done by AI and which by a person, and ask the same question twice. A finding that changes is not yet a finding.
Where CFOLogic fits
We use AI on the volume work in buy-side diligence and keep the judgement with experienced reviewers, with every finding tied to the entry or document behind it. Navigator, our software, connects to QuickBooks read-only and calculates rather than estimates, so the same question gets the same answer twice.