−63% manual data entry across a freight network

An EU freight operator moved goods quickly and moved paperwork slowly. Every shipment generated documents — bills of lading, customs declarations, carrier invoices — that arrived as PDFs and scans and had to be re-keyed into the transport management system before anything could ship.

The problem

The bottleneck was human transcription. Staff read documents and typed their contents into the TMS. It was the slowest step in the chain, the most common source of errors, and the first thing that broke when volume spiked. Worse, the documents arrived in dozens of layouts from dozens of carriers, so no rigid template could keep up.

What we built

We built a document pipeline that extracts structured data from any layout, validates it against business rules, and routes it into the TMS. Documents are classified, parsed, and checked — references that don't resolve, totals that don't add up, and dates that don't make sense are flagged before they reach the system of record. Low-confidence extractions go to a review queue with the source document side by side, so a person confirms rather than re-types.

Each step runs as a durable workflow, so a carrier outage or a malformed file pauses and resumes instead of losing work.

The result

Paperwork stopped being the constraint on throughput.

• Manual data entry down 63%
• Documents turn around in about 30 seconds instead of minutes
• 99.2% field-level extraction accuracy on validated documents

The network scaled its volume without scaling its back office.

Figures are anonymized at the client's request.