Logistics & Freight Forwarding

From 40 hours/week of manual document processing to 4 — across 14 document types.

A freight forwarder was processing bills of lading, packing lists, commercial invoices and customs declarations manually. AP, ops and customs staff spent a combined 40 hours/week on data entry. We built a document AI pipeline integrated with their TMS.

40 → 4
hours/week on data entry
14
document types automated
97.3%
extraction accuracy
26 wks
from kickoff to production
The problem

Every shipment required staff to read PDFs, type fields into the TMS, cross-check against customer instructions, and flag discrepancies. Errors slipped through and caused costly customs delays. Hiring more clerks wasn't the answer — it was the wrong work.

What we built

We deployed a document AI pipeline using a mix of fine-tuned extraction models and an LLM verification layer. Documents are ingested via email and FTP, classified, extracted, validated against business rules, and posted to the TMS. Exceptions are routed to a review queue with the AI's reasoning attached so staff make decisions, not data entry.

Results

97.3% straight-through processing in production. Combined manual processing time dropped from 40 hours/week to 4. Customs error rate fell by 60%. Staff redeployed to exception handling and customer service — work that actually needs humans.

Trend · cumulative outcome40 → 4

We didn't want to replace people. We wanted to stop wasting their time on data entry. That's exactly what we got.

— HB Goh, Head of Operations
Let's talk

Tell us what's breaking. We'll show you what to fix.

A 60-minute call with a senior engineer — a working conversation, not a pitch. You'll leave with a concrete next step, whether or not we work together.