Gains:
- Ability to draft basic foreign trade documents such as commercial invoice, proforma, packing list and bill of lading with artificial intelligence and have them checked for consistency
- Ability to have artificial intelligence check cross-data consistency between documents (amount, weight, amount, party information) and catch errors early
- Being able to protect that the draft declaration produced by artificial intelligence does not replace the official declaration and that the final declaration is the responsibility of the customs consultant.
The unseen hero of foreign trade is documents. Before a container can leave the port, a stack of documents is prepared behind it: commercial invoice, packing list, bill of lading, certificate of origin, movement document, insurance policy and customs declaration. These documents must be fully consistent with each other. If the total weight on the invoice and the weight on the packing list do not match, if the amount on the invoice and the value on the declaration conflict, if the buyer-seller titles differ from document to document, the customs process stops: the shipment is suspended, a penalty is charged, or even the goods are returned. In this unit, we will discuss how to use artificial intelligence as an assistant that accelerates this document traffic and checks consistency, and where its limits are.
Let's clarify the terms first. Commercial invoice: It is the official sales document issued by the seller to the buyer, showing the type, quantity, unit and total price of the goods, delivery and payment method; It forms the basis of value at customs. Proforma invoice: It is a draft invoice in the nature of an offer given before the sale. Packing list: lists how much goods are in which box, number of boxes, gross/net weight and volume. Bill of Lading (B/L): is the transport document showing that the carrier has received the goods and will deliver them upon arrival; It is a valuable instrument in sea transportation. Customs declaration: It is the document in which the import/export of the goods is officially declared to the customs. ETGB (Electronic Commerce Customs Declaration): is a simplified declaration used in small-scale e-export shipments.
Step by step: the place of AI in the document workflow
It's helpful to think of document automation in four steps.
Step 1 — Data collection. The raw information you need: type and quantity of goods, unit price, weights, party information, delivery and payment method. This information is generally scattered in orders, contracts and supplier correspondence. The AI is very fast at extracting a structured table from this raw text you provide.
Step 2 — Producing a draft. Generating invoice, proforma or packing list draft from structured data. Here the AI fills a template; It maintains the format, required fields, and language. But the draft he produced is not an official document; An official document is produced from your system/program.
Step 3 — Cross-consistency check. This is AI's most valuable contribution. Giving the invoice, packing list and bill of lading to YZ and asking "Do the quantity, number of parcels, gross/net weight, amount, party titles and delivery method match? If not, where is the discrepancy?" Within seconds, AI marks the weight difference of 3 kilograms or more than a parcel that the human eye misses.
Step 4 — Expert approval and official action. Declaration and customs procedures are the responsibility of the customs consultant. Nothing the AI produces is a substitute for official declaration; It paves the way for the consultant as a draft, checklist and consistency report.
Caution: AI can produce a “draft” of an invoice or statement; But the responsibility for official documents, signatures and declarations belongs to the person and the customs consultant. Sending the draft thinking it is an official document is like signing an unfilled form.
Inter-document consistency: critical checkpoints
The table below summarizes the areas that most frequently conflict between documents and cause problems at customs:
controlled area
In which documents?
Typical error
result
Total quantity/piece
Invoice, packing list, bill of lading
Writing different numbers
Rejection at customs, counting
Gross/net weight
Packing list, bill of lading
Not holding weight
weighing, delay, penalty
Goods value (amount)
Invoice, declaration
Amount mismatch
valuation review
Party title/address
All documents
different spelling
document rejection
Delivery method (Incoterms)
invoice, contract
Conflict between documents
Cost/risk dispute
country of origin
Invoice, certificate of origin
Incompatible origin
Preferential tariff loss
Number of boxes/packages
Packing list, bill of lading
Missing/excess parcels
Count, minutes
Checking these areas regularly with every shipment solves most of the problems before they reach customs.
Four copyable templates
1) Extracting structured table from raw data:
Your role: foreign trade operations assistant. An invoice data table emerges from the following messy order text. Columns: goods type | quantity | unit | unit price | total | Origin. Mark the fields you see as missing or contradictory as "INCOMPLETE/SUSPICIOUS", do not fill them in yourself. Keep currency as is.Text:[paste order/correspondence text here — non-hidden]
2) Cross consistency check:
Your role: document consistency checker. I give you the data of 3 documents: INVOICE, PACKING LIST, BILL OF LADING. Compare the following fields and show "satisfied / DOES NOT hold" in a table: total amount, number of parcels, gross weight, net weight, goods value, buyer's title, seller's title, delivery method, origin. Write the difference numerically for each line that does NOT hold. The result is not a definitive statement; Final control belongs to the customs consultant.INVOICE: [...]CHECKING LIST: [...]BILL OF LADING: [...]
3) Proforma invoice draft:
Your role: export document assistant. Prepare a draft PROFORMA INVOICE in English with the following information. Fill in the mandatory fields: seller/buyer, date, validity, description of goods, quantity, unit/total price, method of delivery (Incoterms), method of payment, origin, estimated dispatch time. I will verify the numbers; Do not guess any amount. Information: [anonymous/sample data]
4) Document deficiency checklist:
Your role: customs process assistant. Destination: [import/export], country: [...], mode of transport: [sea/air/land]. Make a CHECKLIST of documents that may be required for this shipment (invoice, packing list, bill of lading, certificate of origin/circulation, insurance, special permit/compliance documents if necessary). Write "why is it needed" and "where to buy / who approves" next to each document. This is a reminder list; The consultant determines the final document requirement.
Weak prompt / Strong prompt
Weak prompt:
Check these documents, are there any problems?
This prompt, written without providing documentation or telling which areas to compare, produces a superficial and useless "generally looks good" answer; It does not capture the small numerical differences that are really dangerous.
Powerful prompt:
Your role: document consistency checker. Compare the net weight, gross weight, number of boxes and total quantity in the INVOICE and PACKING LIST data below. For each field, give "holds/does not hold + numerical difference". Also check whether the buyer's title is exactly the same in the two documents. INVOICE: net 980 kg, gross 1.050 kg, 42 boxes, 2.100 pieces... PACKING LIST: net 980 kg, gross 1.050 kg, 40 boxes, 2.100 pieces...
This prompt gives clear fields, clear data, and clear output format; As a result, a concrete and correctable finding emerges, such as "the number of parcels does NOT match: invoice 42, packing list 40, difference 2 parcels".
three mini cases
Case 1 — Caught parcel difference. A forwarder company started comparing invoices and packing lists with AI for every shipment. In the first month, 7 out of 60 shipments were found to have a box number or weight mismatch. These were corrected before reaching customs; An estimated 3 customs reports and delays were avoided. AI marked the difference, the expert did the correction and verification.
Case 2 — Error mistaking the draft for the official. A new employee sent the English proforma draft produced by YZ to the customer without checking it. There was one unit price (below the actual price) left in the draft that the AI had set as an "example". The customer thought this price was binding and a dispute arose. Lesson: Every number in the AI draft must be human-verified before being sent; The draft is not an official document.
Case 3 — Amount discrepancy. In an import company, there was a small difference (120 USD due to rounding) between the total value on the invoice and the value entered in the declaration. The AI flagged this on cross-checking; The consultant corrected the difference. If it was not corrected, valuation review and possible punishment would be on the agenda. Small difference, big problem.
Common mistakes
- Mistaking the AI draft for an official document. The outline is only preliminary; The official document is generated from your system and the expert approves it.
- Skip the cross check. Small differences between documents lead to the most expensive customs problems; Compare with each shipment.
- Making AI predict missing data. The model can fit empty space; Mark the missing items as "MISSING" and do not fill them in.
- Providing confidential price and customer information as is. Run the required fields for the check with anonymous/sample data.
- Sending numbers to the AI without verifying them. Every amount, weight and quantity must be confirmed by human eyes.
Tip: When having the AI do the document checking, always say "write the numerical difference". "Does it hold?" The model easily answers "yes" to the question; When you say "show the difference with a number" it reveals the real contradiction.
In summary
Artificial intelligence is very powerful in two places in foreign trade documents: producing rapid drafts from raw data and checking cross-consistency between documents. Catching the amount, weight, value and title differences between the invoice, packing list and bill of lading early prevents most of the customs rejections and penalties. However, the draft that the AI produces is not an official document; Responsibility for declaration and signature belongs to the customs consultant. Every number is verified, confidential data is anonymized, omissions are not made up.
Application task
Get invoice and packing list data of a real (or sample) shipment, anonymize confidential information. Have the AI compare with the “cross-consistency check” template above. Check the "DOES NOT SATISFY" lines in the output one by one and note whether they are correct. Then create a "document missing checklist" and compare it with your own process: Is there a document that the AI reminded you that you missed?
checklist
- [ ] I anonymized the document data and gave it to AI.
- [ ] I made an amount/weight/value/title comparison between the invoice, packing list and bill of lading.
- [ ] I wanted the output from the AI in the form of "write the numerical difference", not "does it hold?"
- [ ] I have independently checked each marked difference myself.
- [ ] I verified all the numbers in the draft before sending it.
- [ ] I put the customs consultant's approval for the official document and declaration into the workflow.