LOGISTICS & SUPPLY CHAIN

AI for shipping documents: how AI reads bills of lading and invoices, and where people stay in the loop

How freight forwarders, 3PLs and shippers use AI to extract data from bills of lading, commercial invoices and customs documents, with accuracy checks and human review.

By Akash Soni, Founder & CEO, Zuroth · Updated 9 Nov 2026 · 8 min read

The short answer

AI can read shipping documents in any layout and turn them into structured data, but the value comes from the workflow around it: validate fields against shipments and master data, auto-post what passes, and route low-confidence or high-risk fields such as HS codes, values and dangerous goods to a person. Start with one document type and lane, measure the touchless rate, then expand.

Why documents slow logistics down

Forwarders and 3PLs receive thousands of documents a week from shippers, carriers, agents and customs brokers. Most arrive as PDFs or email attachments in different layouts. Teams re-key them into TMS, customs and billing systems, which causes:

  • Delays in booking, clearance and invoicing.
  • Keying errors that lead to customs holds, penalties and billing disputes.
  • Peaks of manual work when volumes spike.

What changed with AI

Template-based OCR needed a template for every layout. Today's document AI combines layout-aware models and large language models to understand documents it has never seen, extract fields and tables, and explain where each value came from.

Document Key fields Human review usually needed for
Bill of lading / air waybill Parties, ports, containers, weights, marks Consignee mismatches
Commercial invoice Line items, values, currency, Incoterms Values and currency
Packing list Packages, weights, dimensions Weight differences
Customs documents HS codes, origin, declared values HS classification
Dangerous goods declaration UN numbers, classes, packing groups Always

How to roll out document AI: 6 steps

1. Pick one document and one lane

Choose a high-volume document, such as inbound commercial invoices on your busiest trade lane. Collect 300 to 500 real samples.

2. Define fields and rules

List the fields you need, their formats and validation rules: totals must add up, container numbers must pass check digits, ports must exist in master data.

3. Extract and validate

Run extraction, then validate against bookings, purchase orders and master data. Cross-check documents against each other.

4. Set confidence thresholds

Auto-post fields above the threshold; route the rest to a review screen showing the document and extracted value side by side.

5. Integrate with your systems

Post results to your TMS, ERP or customs platform through APIs or EDI, with an audit trail of who approved what.

6. Measure and expand

Track touchless rate, field accuracy, processing time and exceptions per document. Use reviewer corrections to improve, then add documents and lanes.

How Zuroth helps

  • Document AI and AI agents for logistics, from email intake to TMS posting.
  • Review screens and exception queues built into your team's tools.
  • TMS modernization and integration when legacy systems block automation.

Our global forwarder case study shows 70% of shipping documents processed touch-free.

Next step: book a free consultation for a document sample assessment.

Frequently asked questions

Which shipping documents can AI process?

AI can extract data from bills of lading, air waybills, commercial invoices, packing lists, arrival notices, delivery orders, certificates of origin and customs declarations, whether they arrive as PDFs, scans, photos or emails. Accuracy is highest on common layouts and improves with feedback.

How accurate is AI document extraction?

Modern AI handles varied layouts far better than template-based OCR, but no system is perfect. Measure field-level accuracy on your own documents, set confidence thresholds per field and send low-confidence or high-risk fields, such as HS codes and values, to a person.

What does touchless processing mean?

Touchless processing means a document is read, validated and posted into your TMS or ERP without anyone keying data. The touchless rate is the share of documents that need no human correction, a key metric for measuring return on document AI.

Where should people stay in the loop?

People should review low-confidence fields, customs classifications, declared values, dangerous goods data and any mismatch between documents, such as a weight difference between the bill of lading and packing list. They should also approve anything sent to customs or carriers.

How does AI document processing connect to our TMS?

Extracted data is validated against shipments, bookings and master data, then posted to your TMS, ERP or customs software through APIs, EDI or, for older systems, file drops or robotic process automation. Exceptions go to a review queue inside the team's normal tools.

Akash Soni

Akash Soni

Founder & CEO, Zuroth

LinkedIn ↗

FREE CONSULTATION

Talk it through with someone who does this work.

30 minutes with a senior expert. You leave with your top 3 risks and clear next steps, whether or not we work together. NDA before we talk.

Book a free consultation →