Skip to main content
DoneThat

AI Adoption GuideManufacturingShip

Autonomous Customs Documentation

Agentic system classifies HS codes and drafts commercial invoices, certificates of origin, and export declarations from shipment data without manual entry.

Manufacturing processPlanSourceMakeInspectPackShipServiceReturn

By Don, DoneThat’s AI coach · updated

What autonomous customs documentation covers

Autonomous customs documentation is the ship-stage workflow that turns finished shipment records into reviewable export paperwork. An agentic system reads commercial and logistics data for an outbound order, proposes Harmonized System (HS) classifications, and drafts the commercial invoice, certificate of origin, and export declaration the shipping clerk and freight broker need before the goods leave the dock.

The outcome is speed, not unsupervised filing. The system collapses the copy-paste and lookup work that dominates export desks: matching SKUs to tariff lines, aligning invoice values with sales orders, filling origin statements, and assembling declaration fields that must stay consistent across documents. The licensed broker or filing party still submits the official entry. The clerk still owns completeness and exception handling.

This page is written for export-compliance leads and shipping clerk leads who already run a manufacturing ship process and need drafts that are fast enough for same-day cutoffs without inventing facts the ERP never stored.

Inputs the agent needs before it drafts

The agent starts from shipment data that already exists in the order-to-ship chain: sold-to and ship-to parties, line items with product identifiers and descriptions, quantities and units of measure, commercial values and currency, Incoterms, country of origin, destination country, and carrier or booking references when they are available.

HS classification depends on product attributes, not only a short catalog name. Useful inputs include material composition, function, technical specs, and any prior classification decisions the company has already validated. Certificates of origin need manufacturer and origin evidence that matches preferential or non-preferential rules the trade lane requires. Export declarations need license flags, ECCN or equivalent controls where applicable, and consistent party and goods data across the packet.

When commercial data is incomplete, the system returns empty drafts rather than guessing. Missing unit price, ambiguous origin, unresolved product identity, or destination fields that conflict with the booking are stop conditions. Empty output is the correct failure mode: a blank packet forces the clerk to fix the source record instead of reviewing a plausible but wrong invoice that would fail at the broker desk or at the border.

How the agent classifies and drafts the packet

Work typically runs in a fixed sequence. First the agent normalizes line items against the product master and shipment snapshot so descriptions, quantities, and values align. Next it proposes an HS code per line, citing the product attributes and any historical classification the company already approved for that SKU or family. Confidence stays lower when the description is thin or when multiple headings remain plausible; those lines surface for human classification review before the rest of the packet proceeds.

With classifications in place, the agent drafts the commercial invoice: parties, Incoterms, line descriptions, HS codes, quantities, unit and extended values, and totals that reconcile to the commercial record. It then drafts the certificate of origin using the same origin and manufacturer facts, without inventing preferential claims the data does not support. Finally it assembles an export declaration draft that mirrors the invoice and booking identifiers so the broker receives one coherent packet.

Throughout, the agent treats cross-document consistency as a first-class check. An HS code that appears on the invoice but not on the declaration draft is a defect. A value that drifts between sales order and commercial invoice is a defect. Origin language that exceeds what the master data can prove is a defect. The clerk reviews a packet that either holds together or clearly lists the gaps to close.

Where the clerk and broker stay in control

Autonomy stops at drafting and triage. Shipping clerks confirm classifications on ambiguous lines, resolve empty outputs by completing commercial data, and release only packets that match the physical shipment. Export-compliance owners set the classification policy, preference rules, and which destinations or product families always require secondary review.

The broker still files. Descartes, SAP Global Trade Services (GTS), and Flexport-style platforms remain the systems of record or submission path for many manufacturers; the agent feeds them structured drafts and exception lists rather than replacing licensed filing. That separation keeps regulatory accountability clear: the agent accelerates preparation, the broker and the company remain responsible for what is declared.

Hand-off should be explicit. The clerk packages approved drafts with shipment identifiers and open questions. The broker validates against carrier and customs requirements for the lane, then files. Rejected drafts return with field-level reasons so the agent and clerk fix the source data instead of retyping the whole packet.

Operating with Descartes, SAP GTS, and Flexport

Vendors in this space already hold parts of the trade stack. Descartes often sits in compliance connectivity and filing networks. SAP GTS holds master data, screening, and customs process configuration inside the ERP estate. Flexport and similar platforms combine freight execution with document workflows for shippers who outsource more of the move.

An agentic drafting layer is most useful when it reads from the same commercial sources those systems trust and writes drafts back into the path clerks already use for review. That means product and party masters stay authoritative in GTS or the ERP, booking and milestone data stay authoritative in the TMS or freight platform, and the agent does not maintain a parallel truth for HS codes or invoice values.

Integration discipline matters more than model cleverness. Prefer identifiers over free text. Prefer empty over inferred origin. Prefer broker review queues over silent auto-file toggles. When a vendor already generates a document template, the agent should populate and validate fields against shipment data rather than inventing a second document format the desk must reconcile by hand.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first