The Return-Flight Closeout: Ready, Missing, or Flagged
The Departure
The return flight lands, the trip is over, and the receipt pile comes home with you. Now the first selling block of the week is competing with a hotel folio, card rows, and three mystery charges from Tuesday.
That cleanup shouldn't wait for your attention, so give an agent the trip evidence before boarding and land to decisions instead of data entry.
The Co-Pilot
Tool: Claude Cowork
The Use Case: It reconciles one business trip into Ready, Missing, and Flagged queues while you're unavailable, then prepares a review-ready report without submitting it.
The Build: Anthropic's trustworthy-agents guide uses business-trip receipt handling as a concrete example: transcribing receipt photos, extracting details, categorizing expenses, and pausing when policy questions appear. That's the right shape for this job, but it isn't a promise that Claude connects to every expense system.
Before the return flight, create one approved trip folder. Add only the itinerary, receipt photos, relevant card rows or a trimmed CSV, the hotel folio, and your current expense policy. Remove unrelated transactions and unnecessary personal or full-card data. Use an account and environment your company approves.
Then give Cowork this instruction:
The Prompt:
"You are my Return-Flight Closeout Agent for [TRIP NAME AND DATES]. Use only the files in this approved trip folder. Reconcile every expected trip charge from the itinerary, receipt photos, relevant card rows or CSV, hotel folio, and expense policy. Build one review table with these columns: status, merchant, date, amount, currency, category, business purpose, meeting or trip, original receipt link, card-row reference, policy result, and notes. Assign every expected charge to exactly one status: READY when all required fields and evidence are present; MISSING when a receipt or required field is absent; FLAGGED when there may be a duplicate, currency mismatch, personal charge, unclear business purpose, over-policy item, or policy ambiguity. Preserve every original receipt link. Never invent or infer a missing merchant, amount, currency, attendee, purpose, or policy answer. Put unknown values in MISSING and explain what I need to supply. Put uncertain matches in FLAGGED and show the competing evidence. Check that expected trip charges = READY + MISSING + FLAGGED, with zero unexplained rows. If you can access the expense system with approved permissions, prepare a draft only. Otherwise, create a review-ready spreadsheet and a short missing-information checklist. Do not submit, send, reimburse, delete, or change any source file. Stop for my review before any external action."
The useful output isn't a prettier stack of receipts. It's a small exception queue you can clear from your phone:
Ready: evidence, required fields, and policy check are complete.
Missing: you need to add a receipt or answer a specific question.
Flagged: the agent found a mismatch, duplicate, or judgment call.
That gives the job a hard acceptance test:
expected trip charges = Ready + Missing + Flagged
Zero unexplained rows. And no silent guesses hiding inside a polished spreadsheet.
The same three-bucket specification can be adapted to ChatGPT Work, Gemini, or Grok, as long as your chosen environment can read the approved files and complete the required steps. File access, connected apps, background operation, and approval controls differ, so verify your setup before boarding.
The Upgrade
Topic: Run the Return-Flight Closeout
Pack the evidence. Create one trip folder, trim the card export to relevant rows, add the folio and receipts, and include the applicable policy. Less input means less private data exposed and fewer false matches.
Let the agent reconcile. Start the job before boarding or your final meeting. Require the three queues, original receipt links, and zero invented values while you're unavailable.
Clear exceptions, then submit. When you land, answer Missing items, decide Flagged items, and confirm the equation balances. Review every Ready row before you press the expense system's final submit button yourself.
The Landing
Use your last completed trip as the test. Put its itinerary, receipts, folio, policy, and trimmed card rows into one approved folder, run the closeout, and count the unexplained rows. Don't scale the workflow until that number is zero.
Safe travels,
Marcellus