Context
This comes from a long consulting engagement with an appliance-install business. We learned the operation the way anyone new would: through its email, WhatsApp threads and the portals the installer network uses. Every agent we've built there starts the same way, by reading and reporting.
The problem
The network pays in weekly files. Some jobs earn an extra charge depending on location and what's installed. When that line is missing, the job still looks paid, because the base install line is there. Spotting the gap means checking every job against every file, and nobody had time for that.
The setup
For every completed job, the agent checks whether it qualifies for the surcharge, then looks for that line in every payment file received since the job was done.
# the rule, in plain terms (simplified) for job in completed_jobs: if qualifies_for_surcharge(job): if not seen_in_any_payment_file(job, "surcharge"): flag(job, status="not yet seen", evidence=files_checked)
What it found
Nine completed jobs qualified for the surcharge but were paid only the base install. Each flag came with its evidence: the job, the rule it met, and which files were checked.
What broke
Payment files don't carry pay dates, so the agent can only say which file a line arrived in. A very recent job might still be paid next week. So every finding is labelled "not yet seen" rather than "never paid", and a person decides when to claim.
Why read-only
The agent could technically draft and file the claims itself. It doesn't, on purpose. The team needs to understand what it checks and why before it acts on their behalf. Full autonomy is possible here, but it's a way off, and that's the right order.
Try it yourself
- Export a list of completed jobs and the rules for any extra charges.
- Collect every payment file since your oldest open job.
- Ask an AI to match them line by line and list the gaps, along with the evidence for each.
- Label findings "not yet seen" and have a person review them before anything gets claimed.
Free · Kickstart prompt
Get a rough first version today
Paste this into Claude, ChatGPT or Grok with your own data. It gets you a useful first pass, not the tested setup.
You are helping me check whether every job I completed was paid correctly. I will paste two things: (1) a list of completed jobs with date, job ID and the extra charges each job should earn, and (2) the payment file lines I received. For each completed job, tell me whether I can find a matching payment line for the base amount and for each extra charge. Make a table: job ID, expected amount, amount found, which payment file it was in, and status (paid / short / not yet seen). Do not guess. If a line is ambiguous, say so. Label anything missing as "not yet seen", not "never paid", because it may arrive in a later file. Show your matching logic so I can check it by hand. When you want the tested version (a recommended AI, a step-by-step build prompt with follow-ups, a test checklist and a worked example; no coding needed), see the Workshaped Blueprint: https://workshaped.ai/use-cases/unpaid-surcharge-reconciliation/#blueprint
✓ Prompt tested 16 Oct 2026: we ran it with gpt-5-mini on a synthetic sample and checked the output by hand.
The Blueprint
Build this yourself: payment-file reconciliation
The idea above is free. The Blueprint is a pack of tested, copy-paste prompts: your AI builds most of the workflow with you, and you finish it in your own back-and-forth. No coding needed.
- Which AI to use, and why: ChatGPT, because its free plan can analyze CSV files with code. Fallback: Claude.
- A main build prompt: it asks you a few questions, then does the job with you step by step
- 4 follow-up prompts: “Double-check your work”, “Adapt it to your situation”, “Check the next payment file”, “Make a monthly log”
- A test-it checklist, with a synthetic sample to try before your own data
- A worked example with the expected result, so you know what good looks like
- A printable PDF guide, plus an optional developer kit (code, workflow file and tests) for technical users
Digital download, delivered instantly. Sales are final once accessed; see our Refund Policy. By continuing you agree to our Terms. Not professional advice; see Disclaimers.