Library › Agents & Automation AI › Finance and accounting
Brief an AP agent to match bills before approving them
- For Finance and accounting
- You practice Catching mistakes and made-up facts
- Time about 5 minutes
About this exercise
You run accounts payable and want an AI agent to check this week's vendor bills. Write its brief: approve a bill for payment only when the billed quantity and unit price exactly match both the purchase order and what was received, put any mismatch on hold with the numbers in the reason, and never pay anything (payments run Friday after your review). Output exactly 5 numbered plain-text lines: one per bill in order, then a DONE line with counts.
Common mistakes when you ask AI for this
- Not showing the log format. Left alone, the AI writes a friendly report with headings instead of a checkable action log. Try: Show it literally: "Format every line like 1. ACTION: tool(args), then ASK_HUMAN: and DONE: lines. Plain text."
- Not defining a match. The AI approves bills that look about right, or that a vendor says are fine. Try: Add: "Approve only when quantity and unit price exactly match the PO and receiving log."
- Vague hold reasons. The AI writes "discrepancy found", so you have to redo the check. Try: Add: "In each hold reason, show the two numbers that don't match."
- Letting it pay. Given pay_now, the AI may pay matched bills straight away. Try: Add: "Never use pay_now. End with DONE: N approved, N on hold."
What a strong prompt includes
- Defines a match
- Asks for numbers in reasons
- Forbids paying
- One line per bill
- Shows the action-log format
Related exercises
- Alert sales to a competitor's price change, facts only (Research & AI Search · Product managers)
- Approve or return expense claims against the policy (Productivity AI · Finance and accounting)
- Approve part of an overtime request with the math (Writing AI · Managers and team leads)
- Ask the right questions about a usage drop (Customer Support & Success AI · Customer success managers)
- Audit chatbot transcripts against guardrails as JSON (Customer Support & Success AI · Customer support teams)