Shipping-document extraction for a logistics operator

The client is a UK logistics operator handling international consignments, where paperwork accuracy directly affects customs clearance and delivery times. A single keying error can hold a shipment at the border.

Their back office spent significant effort re-keying documents that arrived in wildly inconsistent formats, and that manual step was both a cost and a source of downstream errors.

Shipping documents arrived as scans and phone photos in dozens of inconsistent formats, and re-keying them by hand delayed every consignment. The variety alone defeated simple template-based tools that only work on clean, predictable inputs.

Manual entry introduced errors that surfaced downstream as customs holds and disputes, each one expensive to unwind and damaging to delivery promises. A mistake made in seconds could cost days to resolve at the border.

The operator needed reliable extraction across messy real-world inputs, not a demo that only worked on pristine templates, but something robust enough to trust on the documents they actually receive every day.

We built a layout-aware extraction pipeline that reads the key fields regardless of format, using document structure rather than fixed positions. That structural understanding is what lets it handle the format variety that breaks template-based approaches.

Every extracted value is cross-checked against the consignment record, so mismatches are caught immediately instead of surfacing at the border as a hold. Validation at the point of extraction turns errors into caught exceptions rather than downstream incidents.

Low-confidence pages route to a reviewer with the uncertain fields highlighted, keeping the automated output trustworthy without slowing the clean cases. We tuned it on the operator's own document mix, so accuracy reflected the formats they genuinely receive rather than an idealised sample.