Logistics
Series B logistics & fulfillment startup
~70% reduction
in manual re-keying time across the three highest-volume document types
Review queue, not black box
low-confidence extractions are still reviewed by a person before they hit billing
3 weeks to first production document type
shipped incrementally instead of waiting for full coverage
The Challenge
Incoming shipping manifests, customs forms, and carrier invoices arrived in a dozen inconsistent formats, and a team of five was manually re-keying them into the operations database.
Data entry backlog was growing faster than headcount, and re-keying errors were causing downstream billing disputes.
My Approach
Built a structured extraction pipeline that parses incoming documents and assigns a confidence score to every extracted field, not just the document as a whole.
Anything below the confidence threshold routes to a review queue with the source document highlighted next to the extracted value, instead of silently guessing.
Shipped incrementally against the three highest-volume document types first, then expanded coverage once accuracy held up in production.