Back to catalog / Agent workflows
Build a receipt review and audit decision
Extract receipt fields, reconcile line items and totals, and route anomalies to a human audit queue using measurable criteria.
Goal
Extract receipt fields, reconcile line items and totals, and route anomalies to a human audit queue using measurable criteria.
Steps
- Define structured receipt fields, audit decisions, and labeled examples. [src-openai-build-receipt-review-audit-decision]
- Build extraction, then check line items, subtotal, tax, and total with rules. [src-openai-build-receipt-review-audit-decision]
- Measure error types and audit costs, refine routing, and retain uncertain cases for review. [src-openai-build-receipt-review-audit-decision]
How to verify
Compare merchant, date, items, tax, and total with held-out labels; confirm planted arithmetic errors trigger audit.
Troubleshooting
When an audit decision is wrong, separate OCR extraction errors from rule errors; do not hide amount-recognition defects by relaxing thresholds.
Permissions
Prerequisites
- OpenAI API key, receipt images, structured labels, and an isolated evaluation set.
Permissions
- Images may be uploaded to a model API and incur charges; results are audit recommendations only and do not update payment or reimbursement systems.
Limitations
The source is a small-data, eval-driven prototype; add representative cases and have finance review the rules.
Pinned upstream source
- Repository
- openai/openai-cookbook
- File path
- examples/partners/eval_driven_system_design/receipt_inspection.ipynb
- Pinned commit
5bd6cb40b6b84b69b2032d810610aa85064102df- Source checked
- 2026-10-14 · Reviewed against the pinned upstream source
Repository metrics
These values describe the upstream repository, not popularity or likes for an individual skill or playbook.
- Stars
- Not collected
- Forks
- Not collected
- Observed at
- Not collected
No metrics snapshot is available yet.
License: MIT
Main topic
Turn ecommerce campaign materials into a reviewable weekly brief