Case study · Freight forwarding

Document OCR for a Taiwan freight forwarder

Bills of lading, invoices, and packing lists were retyped by hand. We built a system that reads them and fills in the data, leaving people to check only what needs checking.

3-4hrs → 10-15min
Per-shipment review
Freight
Industry
Taiwan
Region

The challenge

Shipping documents arrive in many layouts from many parties. Staff keyed each field into the operations system by hand, then cross-checked it against the original document to catch mistakes. That review alone took 3 to 4 hours per shipment.

The harder problem was consistency. Every supplier laid out their B/Ls and packing lists differently, so the check depended on whoever was doing it that day. Two people reading the same document could flag different things.

Approach

What we did

01

Collect real documents

Across senders and layouts, including the messy scans.

02

Build field extraction

Focused on the fields operations actually uses.

03

Add confidence checks

Uncertain fields are flagged for review instead of guessed.

04

Write into the existing system

The team keeps its current workflow.

Technical

Architecture

01

OCR and field extraction

Layout-aware reading across document formats.

02

Per-supplier correction rules

Each sender's format quirks are encoded once and applied automatically, so the fix doesn't depend on who catches it.

03

Confidence scoring

Each field scored; low-confidence fields go to a review queue.

04

System integration

Validated data written into the client's operations system.

Related

A system to build,
or one that needs taking over?

Book a consultation →