Consider a Jamaican business that gets the chance to buy a smaller competitor. The price is fair, the fit is good, and the seller’s attorney sends over the paperwork: a sale agreement, the premises lease, the staff contracts, two loan facility letters, and an aged receivables listing. Together it runs past 300 pages, the exclusivity window is three weeks, and every one of those pages was drafted by someone whose job was to protect the other side. The traditional options are to pay an attorney to read all of it, which is expensive, or to skim it yourself and hope, which is how people end up owning a lease escalation clause they never saw.

Due diligence sounds like something that happens in glass towers, but every business signs its way into long documents: the bank facility letter, the commercial lease renewal, the franchise agreement, the equipment finance contract, the big supplier agreement with the small print about minimum volumes. This is the job AI has become excellent at — patient, tireless reading of documents far too long for a busy owner to absorb — and this week the column shows how to use it without fooling yourself about what it can and cannot do.

The stack behind a business purchase — sale agreement, lease, staff contracts, loan letters,
receivables — and the three-step workflow that gets it read. (Branded graphic by PGH Consulting, LLC)

Start with a map, not a summary

The paid versions of the leading assistants — ChatGPT, Claude, and Copilot, compared in last week’s column — will all accept a long PDF and work through it; documents running to a few hundred pages are now routine, though upload limits vary by tool and plan, so check the vendor’s own pages before you rely on one. The mistake most first-timers make is asking for “a summary,” which produces a page of smooth generalities you could have guessed from the cover letter.

Skip the summary. Ask for a map of the document first, then interrogate it section by section. (Branded graphic by PGH Consulting, LLC)

Ask for a map instead. Upload the document and start with: “List the sections of this agreement in order, with page numbers, and one line on what each does.” Now you know the terrain, and you can see where the money lives — the payment terms, the termination provisions, the schedules at the back where the surprising things are usually parked. Then interrogate it section by section, and insist on references: every answer should come with a clause number and page so you can check it against the document itself. An answer without a reference is a rumour.

Six questions do most of the work in any long contract — asked one at a time, never as a single
mega-prompt. (Branded graphic by PGH Consulting, LLC)

The six questions that earn their keep

Across contracts of every type, a handful of questions do most of the work. Ask them one at a time, not as a single mega-prompt:
1. “List every obligation this document places on me, with clause references.” Not the headline terms — all of them, including the insurance you must carry and the notices you must give.
2. “List every date, deadline, and notice period.” Renewal windows are where leases catch owners out: many require notice months in advance, and silence can mean automatic renewal on the landlord’s terms.
3. “What happens if I want to exit early, and what does it cost?” Termination clauses, penalties, and anything described as liquidated damages.
4. “Which obligations survive termination?” Guarantees, restraint-of-trade clauses, and confidentiality can outlive the agreement by years.
5. “What can the other side change without my consent?” Rates, fees, service levels — anything with the words “from time to time” nearby deserves your full attention.
6. “What would you expect to see in a document like this that is missing?” This is the question that separates careful reading from real due diligence — and, as discussed below, it is also where AI is weakest, because a model can only be tentative about what is absent.

For the business purchase above, the same method applies across the stack: ask which staff contracts transfer with the business, whether the lease requires the landlord’s consent to the sale, and whether the loan letters contain change-of-control clauses that make the debt repayable the day the business changes hands. Those three answers alone can reshape a purchase price.

AI compresses the reading, not the diligence — and the attorney, the NDA check, and the
business account are not optional. (Branded graphic by PGH Consulting, LLC)

Minutes to read, an hour to verify

The real arithmetic of this workflow is that AI compresses the reading, not the diligence. What used to be a weekend of reading becomes twenty minutes of questions — and then a focused hour of verification, because you must open the document and confirm every clause the AI has flagged before you act on it. These tools occasionally misattribute a clause, blend two provisions into one, or state a term with more confidence than the text supports. The habit that makes the whole method safe is mechanical: no flagged clause goes into your decision, your email to the seller, or your meeting with your attorney until you have read it yourself at the cited page. Scanned documents deserve extra suspicion — a poor scan can garble numbers, and a garbled number in a facility letter matters.

That hour is not a tax on the method; it is the method. You arrive at the real documents knowing exactly which ten pages of the 300 deserve your full attention.

What this does not replace

An AI assistant is not an attorney, and nothing above is legal advice. It does not know Jamaican law, the Registrar of Titles, or how a particular clause has been treated by the courts; it cannot tell you whether the lease breaches the Rent Restriction Act or whether a restraint clause would actually hold. What it changes is the quality of the conversation you pay for: instead of handing your attorney 300 unread pages, you hand over a map, six answered questions, and a shortlist of clauses that worry you. The billable hours go on judgement instead of reading.

Confidentiality needs the same discipline this column applies every week. Use a business-plan account, where your files stay out of model training — the free personal tiers are the wrong place for anyone’s sale agreement. And remember that the other side’s documents are often covered by a non-disclosure agreement; check what yours says about sharing materials with service providers before uploading, and when in doubt, ask the counterparty or strip the sensitive schedules out. Finally, the series rule stands unchanged: AI reads figures, it never generates them. The receivables listing gets checked against the seller’s books, not summarised into existence.

What to try this week
1. Practise on a document you have already signed — your lease or bank facility letter. Upload it to a business-plan assistant and ask for the map: sections, page numbers, one line each.
2. Run the six questions above, one at a time, and require a clause and page reference with every answer.
3. Verify at least three of the flagged clauses against the actual document, so the checking habit is built before a real deal depends on it.
4. Put every date and notice period it finds into your calendar, with reminders set well before each deadline — especially lease renewal windows.
5. Save the six questions as a reusable prompt, and take your annotated map to your attorney the next time a long document arrives — before you sign, not after.
Use a business-plan account for anything confidential, check your NDA before uploading anyone else’s documents, and treat every AI answer as a first read, not a final one — the signature is still yours.

Peta-Gaye Hardy is the founder of PGH Consulting, LLC, where she helps finance and operations teams adopt AI in practical, low-risk ways. She writes the weekly AI in Finance & Business column and is based between Jamaica and the United States. Learn more at www.pghconsultinggroup.com. Follow on Instagram and YouTube @pghconsultinggroup, and connect on LinkedIn at linkedin.com/in/peta-gaye-hardy.

Disclosures: This article is informational and does not constitute investment, tax, legal, or accounting advice; readers entering a transaction should engage a qualified attorney-at-law. AI tools can produce errors, and every clause or figure they identify should be verified against the source document. Product capabilities and upload limits are as published by the vendors at the time of writing (July 2026) and are subject to change. The author has no commercial relationship with OpenAI, Anthropic, Microsoft, or any product mentioned and was not compensated by them. The example transaction described is illustrative and does not depict any real business.