Yes, AI can handle a large share of shipping document work for freight forwarders: extracting data from commercial invoices, packing lists and bills of lading, cross-checking them against each other and the booking, drafting house B/Ls and pre-filling customs and filing data, suggesting HS codes for a licensed broker to confirm, and chasing shippers for missing or corrected documents. What it shouldn't do is file declarations or decide classifications on its own. Used that way, AI for freight forwarders turns the documentation desk from a re-typing operation into a review-and-exceptions operation.
This post walks through a typical shipment's document flow, shows where AI reads, checks and drafts, explains the HS code question honestly, and gives a starting point for a small or mid-sized forwarder or customs broker.
What does a freight forwarder's documentation desk do all day?
For each export or import shipment, a forwarder's documentation team typically:
- Receives the commercial invoice and packing list from the shipper, usually as PDFs or spreadsheets by email
- Checks them against the booking: consignee, notify party, ports, Incoterms, cargo description, package count, weights and volume
- Prepares or checks the shipping instructions and drafts the house bill of lading or air waybill
- Reviews the carrier's draft master B/L for errors before it's finalized
- Prepares data for customs and advance filings, which vary by country (for example, shipping bills in India, ISF filings for US imports, ICS2 entry summary data for the EU)
- Collects supporting documents: certificates of origin, phytosanitary or fumigation certificates, dangerous goods declarations, licenses
- Chases the shipper for corrections, and the consignee or bank for document requirements under a letter of credit
Every step involves reading the same facts from different documents and checking that they match. When they don't, the result ranges from an amendment fee to cargo held at customs.
How does AI extract data from shipping documents?
Commercial invoices and packing lists don't follow a standard layout. Every shipper has their own format, and many are scanned or photographed. Modern document AI, combining OCR with language models, handles this variety far better than older template-based tools.
For each document, the AI extracts the fields your team needs:
| Document | Key fields extracted |
|---|---|
| Commercial invoice | Seller, buyer, consignee, invoice number and date, Incoterms, currency, line items with description, quantity, unit price, value, country of origin, HS codes if stated |
| Packing list | Package count and type, marks and numbers, net and gross weights, dimensions and volume per package and in total, item breakdown per package |
| Bill of lading or AWB draft | Shipper, consignee, notify party, ports or airports, vessel or flight, container and seal numbers, package count, weight, description, freight terms |
| Certificates and declarations | Issuer, reference numbers, dates, product descriptions, validity |
Each extracted value carries a confidence level and a link back to where it appeared on the page. Low-confidence fields go to a person for confirmation rather than being guessed. Our guide to intelligent document processing explains this approach in more depth.
How does AI cross-check documents against each other?
Extraction alone saves typing. Cross-checking is where the real value is, because it catches errors before they become amendments or holds.
The AI compares the documents with each other and with the booking:
- Package count and type match across invoice, packing list, shipping instructions and B/L
- Gross and net weights are consistent, and the total on the packing list matches the sum of the lines
- Consignee and notify party match the booking and, for L/C shipments, the letter of credit terms
- Cargo description is consistent and specific enough for customs (not just "parts" or "samples")
- Incoterms and freight terms on the B/L match the invoice
- Container and seal numbers on the draft B/L match the terminal or carrier records
- Values and currency add up, and any HS codes on the invoice match what's being filed
When something doesn't match, the agent writes a clear exception: "Packing list shows 48 cartons, gross weight 1,240 kg; commercial invoice says 46 cartons. Please confirm with shipper." Your documentation executive decides what to do, and the agent can draft the email to the shipper.
Can AI assign HS codes?
This is the question forwarders ask most, and the honest answer is: AI can suggest HS codes; a qualified person should decide them.
Classification under the Harmonized System depends on the exact nature of the product: material, function, composition, degree of processing, and the general interpretive rules, plus national tariff extensions beyond the first six digits. It affects duty, eligibility for trade agreements, licensing requirements and restrictions. Misclassification can lead to penalties, delays and back duties.
What AI does well here:
- Reads the product description, specifications and any product literature, and proposes candidate codes with its reasoning
- Checks consistency with how the same product was classified on your previous shipments
- Flags descriptions too vague to classify ("machine parts") and drafts a question to the shipper asking for more detail
- Highlights when a proposed code may carry licensing, restrictions or special documentation requirements, for the broker to verify
What it shouldn't do is file a classification without review. Tariff schedules change, national rules differ, and binding rulings matter. A licensed customs broker or trade compliance specialist confirms the code, and the AI's reasoning gives them a head start. Rules differ by country, so check current tariff schedules and regulations, or ask counsel, rather than relying on any AI's general knowledge.
How can AI handle shipper communication?
A surprising share of documentation work is email: requesting documents, chasing missing items, sending drafts for approval and confirming corrections. An AI agent connected to the documentation inbox can:
- Send a document checklist when a booking is confirmed, tailored to the trade lane and commodity
- Track what's been received and chase what's missing ahead of the documentation cut-off
- Send the draft B/L to the shipper for approval and record the approval or corrections
- Answer routine questions ("When is the cut-off?" "Has the B/L been released?") from the job file
- Warn the shipper and your team when cut-offs are at risk
All of this is visible in the job file, so the next person on the shipment sees the full history. If you run inbound inquiries too, our guide to AI for import-export businesses covers the shipper's side of the same paperwork.
What does a typical shipment look like with AI in the loop?
For illustration, here's an export sea shipment for a small forwarder:
- Booking confirmed. The agent emails the shipper the document checklist and cut-off dates.
- Invoice and packing list arrive. The AI extracts the data, cross-checks against the booking and flags a weight mismatch.
- Shipper confirms the correction in reply to the agent's draft email, and sends a revised packing list.
- HS code candidates are proposed for two products; the broker confirms one and changes the other.
- Draft house B/L is generated from the verified data and sent to the shipper for approval.
- Customs data is pre-filled in the format your filing software expects, for the broker to review and file.
- Carrier's draft master B/L arrives and is compared with the shipping instructions; a mismatched seal number is flagged and corrected before finalization.
- Everything is logged against the job in your freight forwarding system (CargoWise, Magaya, Descartes or a local platform).
Your team still made every decision that mattered. It just didn't type anything twice.
What are the risks and limits?
Filing liability stays with you. Customs declarations and advance filings carry legal responsibility. Keep a person approving every filing.
Poor scans and handwriting. Faxed, stamped or handwritten documents reduce accuracy. Low-confidence fields must go to review, not be guessed.
L/C documentary compliance. Banks examine documents strictly under letters of credit. AI can check consistency with the L/C terms, but your team or the bank's trade desk decides compliance.
System integration. Value comes from writing verified data into your forwarding system, not just producing a spreadsheet. Check what your platform allows through APIs or imports.
Confidential commercial data. Invoices show prices, customers and suppliers. Use tools that don't train on your data and process it in an approved region. Our guide to AI data privacy and security lists what to ask vendors.
How should a freight forwarder start with AI?
- Pick one trade lane or one large shipper with consistent document volume.
- Start with extraction and cross-checking of invoices and packing lists against the booking. It's low risk and saves time immediately.
- Add B/L draft checks next, since amendments are costly.
- Add HS code suggestions with mandatory broker review.
- Add shipper communication once the document data is reliable.
Measure documents processed per person, amendment rates, holds caused by documentation errors and turnaround from receipt to draft B/L. For a broader logistics view, see how logistics companies use AI agents for shipment exceptions and our complete guide to AI for business.
Frequently Asked Questions
Can AI read bills of lading and commercial invoices accurately?
Yes, for most clear digital and scanned documents. Accuracy drops with poor scans, handwriting and unusual layouts, which is why well-designed systems flag low-confidence fields for review rather than guessing.
Can AI file customs declarations automatically?
Technically it can prepare the data, but filing should remain a human decision with a licensed broker where required. Customs filings carry legal responsibility, and errors can lead to penalties and delays.
How accurate are AI HS code suggestions?
They're useful as a starting point and for consistency checks, especially for products you've classified before. For new or complex products, a qualified person must confirm the code, because classification depends on detailed product facts and current national tariff rules.
Does AI work with CargoWise or other forwarding software?
Many forwarding platforms support data exchange through APIs, EDI or file imports. The integration options depend on your platform and license. Scoping should confirm exactly how verified data will be written back into job files.
Will AI replace documentation executives?
It removes re-typing and routine checking, so each executive can handle more shipments with fewer errors. Their role shifts toward exceptions, shipper relationships, compliance questions and quality control.
Want your documentation desk reviewing instead of retyping?
If your team spends its day copying data between invoices, packing lists and B/Ls, tell us about your shipments and your systems. We'll show you how document checks and shipper follow-ups could run with your team in control. See our approach to AI automation for operations teams.