Engagement snapshot
- Mandate
- Remove repetitive shipment intake work while preserving customs and carrier-specific accuracy checks.
- Timeline
- 11 weeks from document mapping to live lane rollout.
- Team shape
- Ops director, 2 automation engineers, 1 data engineer, and 5 shipment coordinators.
The problem
Operations staff manually extracted shipment details from bills of lading, invoices, and booking emails before entering them into the transport management system.
What we built
Built document ingestion, validation, and booking-draft automation with carrier-specific rules and human review on risky exceptions.
Operating context
Shipment volume was high enough that intake speed mattered, but the real cost came from rekey errors and inconsistent booking prep. The business needed a controlled automation layer rather than more manual headcount.
Key constraints
- Carrier and route rules varied significantly, especially for required document fields.
- Human reviewers needed clean exception queues instead of raw extraction dumps.
- The automation had to preserve source traceability for customs and customer inquiries.
What we built
Document normalization
Converted bills of lading, invoices, and email attachments into a structured shipment packet before validation.
Rule-aware validation
Applied carrier, route, and commodity-specific checks before drafting booking entries.
Exception handling queue
Surfaced only ambiguous or incomplete files to coordinators with highlighted evidence and required actions.
Delivery path
Lane analysis
Started with the most repetitive trade lanes to prove extraction quality and operational savings quickly.
Shadow-mode comparison
Compared automated drafts against human booking prep to score accuracy and exception usefulness.
Live intake rollout
Released by lane with throughput, correction, and error telemetry visible to operations leadership.
Why it mattered
The workflow became faster because coordinators stopped retyping the easy parts. They spent their effort on the documents that were actually incomplete or risky, which improved both throughput and booking quality.
Implementation notes
- Exception design is the product in document automation, not just extraction quality.
- Trade-lane variation needs to be modeled early or users will not trust the drafts.
- Operational trust improved once source snippets were attached to every extracted field.