A brief tour.

Technology changes the surface. Behaviour doesn’t.

This is one of those rules that holds across decades. Each new technology arrives, the surface looks different, and the human behaviour underneath stays remarkably constant. Which is, when you think about it for a moment, a more useful rule than it sounds. The question is always the same: how is the surface different now, and what does the unchanged behaviour look like in the new conditions?

The old economics.

The economics of due diligence have changed, recently, and quietly. On a $10M deal, no buyer was ever going to spend $80,000 on a legal team to scrutinise every contract, every customer file, every payroll record. They would sample, spot-check, lean on warranties, and move on.

Then AI happened.

The new economics.

The same review that used to need three associates and a senior partner billing for two weeks can be done by one lawyer running tools across the entire data room over two or three half-days.

If you’re selling a business in the sub-$20M bracket, this matters more to you than it does at the top of the market. Buyers who would have done light-touch diligence are now doing deep diligence on your deal. The things they’re finding aren’t always the things sellers historically had to defend.

What buyers are actually running.

The tooling varies, the pattern is becoming consistent. Contract review tools (Luminance, Kira, increasingly Harvey-style general models) ingest every PDF in your contracts folder and extract: change-of-control provisions, assignment restrictions, exclusivity, MFN clauses, termination rights, indemnity caps, governing law.

Financial review tools cross-reference your management accounts against your statutory accounts against your BAS against your tax returns, looking for inconsistencies that previously took accountants days to find.

Customer data tools run more sophisticated concentration analysis. Not just “top 10 customers,” but rolling concentration over 36 months, churn cohorts, payment-behaviour patterns.

Increasingly, buyers also run language models across employment contracts, leases, and corporate register extracts, looking for anything that pattern-matches to a restraint, a personal guarantee, or a PPSR issue.

What AI catches well.

Pattern inconsistencies. If six contracts have uncapped liability and you didn’t flag them, the AI finds them in twenty minutes. The buyer’s lawyer used to find these after careful reading. Now they’re surfaced in a pre-populated report.

Cross-document mismatches. Your marketing materials say 120 active customers. Your CRM shows 134. Your contracts folder has 98 subfolders. AI flags this immediately. The answer may be entirely innocent: inactive accounts, oral arrangements, customers transitioning. The point is you need your answers ready before the question lands.

Buried obligations. Change-of-control clauses in supplier contracts nobody has looked at in five years. MFN clauses introducing risk across your portfolio.

Financial reconciliation gaps. Revenue recognition that differs between accounts and contracts. Bad debts not provisioned consistently. Related-party transactions that don’t tie out.

What AI misses.

This is where sellers have leverage, and where good advisors earn their fees.

  • AI is poor at commercial context. It will flag a clause as problematic without understanding that the clause has never been enforced, has been verbally waived for three years, or relates to a counterparty replaced in 2023. It will treat a missing document as a gap when the document was never required to exist.
  • AI is poor at materiality. It flags everything pattern-matching “risk” without weighting commercial significance. A $2,000-a-year supplier contract with an aggressive indemnity gets the same flag as a $2M customer contract with the identical clause.
  • AI is poor at history. The reason your largest customer’s contract has a one-page typed amendment from 2019 is a story. The AI sees an anomaly. A human sees a negotiation.
  • AI is poor at the unwritten. The relationship with your top three customers, the goodwill with your landlord, the long history with your bank. None of this is in the data room. None of it is in the AI’s field of view.

What this means for sellers.

The disclosure schedule is no longer a tidying-up exercise at the end of SPA drafting. It is now the primary tool for managing what AI surfaces, and it should be considered that way from the start.

  • Disclose against the data room early and comprehensively. If you know there are six contracts with uncapped liability, list them. If your CRM and your accounts do not reconcile, explain why. The cost of AI finding these issues is almost nothing. The cost of you not having an answer ready is much higher.
  • Build a reconciliation narrative. For every material discrepancy AI will find between documents, prepare a one-paragraph explanation that sits in the data room. The AI can read this too, which means the buyer sees the discrepancy with the mitigation, not without it. This cures buyer anxiety before it has arisen and reinforces deal momentum.
  • Be deliberate about what goes into the data room. Emails, draft contracts, internal memos, board papers, side letters, all parseable in seconds. A draft contract you sent to a customer three years ago and never signed will be picked up and read as if it were live. If a document should not be in the data room, it should never get there.
  • Use AI yourself before they do. Run the same tools across your own data room as part of preparation. Anything the buyer’s AI will find, you should find first and have a position on. The cost of doing this is now trivial compared to the cost of being surprised.
  • Think about file naming and structure. AI tools work better on well-organised data rooms. That sounds like a buyer benefit, but it cuts the other way too: poorly organised data rooms create false flags. False flags create buyer anxiety. An anxious buyer armed with a few red flags might decide to pick up the phone and request a price chip.

The bottom line.

The fundamentals of disclosure have not changed. You disclose what you know. You cannot disclose what you do not know. You negotiate the warranty package around the gap.

What has changed is the assumption that minor inconsistencies do not matter. They will be found, at almost no cost, and they may be used as ammunition in sufficient quantity to matter.

The sellers doing best in this environment treat the data room as a structured history of the company. Not a document dump. The picture you want to paint is: here is the business; here is what is true about it; here is what is anomalous and why; here is what we are and are not warranting.