XONTAINERLogistics ecosystem
Where does the time go in customs work?
An import or export file usually spends most of its time on collecting documents before the declaration: the commercial invoice, packing list, CMR or bill of lading, origin papers and others arrive from different people in different formats. The broker types these fields into the declaration screen, completes gaps by phone and tracks deadlines in a separate list.
AI’s most direct contribution is this preparation layer: reading the document, placing fields into a draft and flagging inconsistencies, for example a gross-weight difference between the invoice and the packing list.
What does AI do?
Three jobs can be handed over safely:
- Extracting fields from documents: consignor, consignee, goods description, quantity, weight, value and currency move into the draft.
- Suggesting a tariff code: a likely classification is offered from the goods description; it is a suggestion, not a decision, and is shown with its reasoning.
- Flagging inconsistencies: differences between documents and missing fields are put in front of the broker.
Every output is marked as a draft. The broker compares it with the documents and the rule catalogue, corrects it and approves it.
What does AI not do?
It does not file, sign or represent anyone. Customs brokerage is a licensed profession in most systems; in Türkiye it operates under the Customs Law No. 4458. Software or a software company does not act as a broker. Legal responsibility stays with the broker and the declarant.
It also does not apply a rule that has no source. In a good system every rule is tied to a source, and an unknown case shows as “unknown”, not as zero or as a guess.
The right division of work: suggestion, rule, approval
A sound customs flow has three layers. AI produces a suggestion. A rule catalogue tests the suggestion with sourced rules, for example a missing mandatory field or an incompatible unit. A person, the broker, makes the final decision and approves. Nothing is written to the official system before that approval.
This split is not only a legal preference; it also sets the cost of error. A wrong tariff class can come back as extra duty and penalties, and human approval is the step that carries that risk.
Deadline tracking: discipline before AI
Transit and temporary procedures carry unavoidable deadlines. In an MRN-centred file structure every declaration number is tied to its own deadline counter, and T1 deadlines turn yellow and then red as they approach. This is a data-model job rather than an AI job, but once AI takes over drafting and document load, the broker has more time for deadlines.
Rules are moving
On the EU side, the customs reform framework Regulation (EU) 2026/2108 has been adopted with phased application planned, and there are planned dates for a central EU customs data hub. In Türkiye declarations are filed through BİLGE, and development of the system with AI has been announced publicly, but detail and dates are not known. For this reason customs software should not bury legislation in code; it should keep a sourced rule catalogue.
Common mistakes
- Carrying an AI suggestion into a declaration without approval.
- Keeping the rule catalogue without sources; when legislation changes it is not clear which rule rests on what.
- Leaving deadline tracking to AI while neglecting the data model.
- Treating a software company as a broker; brokerage is a licensed profession.