What was the result?
self-reportedMarshall Haas says an agency requested an additional $19,000 after he thought the project was paid up. Claude audited invoices, contracts, stated hours and communications; he says the agency agreed with his findings 24 hours later. This is a disputed extra charge he says he avoided, not cash recovered or a court ruling.
Entrepreneur Marshall Haas had Claude compare the documents behind an unexpected agency bill instead of arguing from memory. The model found discrepancies and helped assemble the case for why no more money was owed. Haas says the agency agreed within 24 hours. In the source thread he confirms connecting email and says he had already paid the agency $70,000; the agency and project are unnamed. The repeatable idea is a documented comparison for a human to review, not AI deciding a contract dispute.
Who did this?
How do I try it?
- Editorial adaptation of the reported audit: collect the signed scope, change orders, invoices, payment records, timesheets and relevant communications.
- Give Claude only authorized copies or narrowly scoped read access; retain the originals.
- Ask for a line-by-line crosswalk of each new charge against the agreement, prior invoices and evidence of approved changes.
- Check every alleged discrepancy manually. Have your accountant or lawyer assess disputed contractual obligations where appropriate.
- Write a factual query to the agency citing verified document references. You approve and send it; record the actual resolution rather than assuming every flagged item is recoverable.
Editorial adaptation, not Marshall Haas's exact prompt. Compare the attached contract, scope/change orders, prior invoices/payments, new invoice, hours and communications. Create a table: new charge, contractual basis, prior billing/payment overlap, supporting file/page or message date, discrepancy and uncertainty. Do not infer missing approvals or claim fraud. Draft neutral questions for my review; do not send anything, withhold payment, give a definitive legal opinion or change records. Separate a possible issue from a verified duplicate.
Make it run every week
A play works once. Turn it into a routine in three small steps:
- Run it once and check the result yourself.
- Give your AI a schedule. Tell it when to run this play (every week, or every month) and to message you with what it found before you act.
- Keep a log. One line per run: what was checked, anything flagged, what you did.
What should I watch out for?
Haas's own account only; no invoices, agreement or agency confirmation independently inspected. He reports agreement in 24 hours, not an independently verified $19,000 refund. An early edit mentioned $20k higher than expected; the final version explicitly says an additional $19k and is the source used. Do not call the agency dishonest based on AI flags. Build it with Page: a project scope and change-approval checklist for future work.
Did it work for you?
PageMotor keeps this site free.