Customs AI Can Speed Declarations, but Brokers Must Keep Approval

Customs AI can take on document checks and repetitive data review, but responsibility for a declaration stays with the person or business that submits it. That division matters when an incorrect commodity code, missing certificate or inconsistent shipment detail can delay clearance or lead to a later audit or penalty.

For brokers and traders, the practical question isn’t simply whether an AI tool can produce a plausible answer. It’s whether staff can inspect the evidence behind a suggestion, see when the system is uncertain and approve the final declaration themselves. A tool that hides its reasoning can make a mistake harder to catch. One that surfaces its checks and unresolved questions can help agents focus their attention where judgment is needed.

Responsibility stays with the declarant

Under Title I, Chapter 2, Article 15 of the Union Customs Code, the person lodging a declaration is responsible for its accuracy and completeness, as well as the authenticity of supporting documents. The article says the UK applies the same principle after Brexit. The responsibility follows the submitted declaration, regardless of whether a person or software entered the information.

Representation arrangements affect who carries exposure. A broker acting as an indirect representative is jointly liable for the customs debt. With direct representation, the trader remains responsible for the data it supplied. A software licence doesn’t transfer those obligations to the system’s provider.

That leaves a clear role for automation: prepare and check information, while a qualified person resolves ambiguous cases and approves what goes out. The tool can read invoices and packing lists, validate commodity codes, check that supporting documents are present and flag data quality problems. When a product description doesn’t support a confident classification, it should ask for review and show the information behind its suggestion.

That review step gives the agent a chance to correct source data, request clarification or decide that the information needs further examination before submission. It also creates a record of who reviewed and approved the declaration. For a brokerage handling many entries, those records can help make the review process visible after the work is complete.

Checks that catch problems earlier

EORI, a UK customs brokerage, handles thousands of declarations each week. Before using the AI platform Customaite, its process relied on manual, line-by-line commodity code checks, document validation and data corrections. According to the account, the work was dependable but slow and relied heavily on agents maintaining attention across repeated checks.

Trade Solutions Manager Robert Keegan described the pressure as “rising volume combined with inconsistent trader documentation.” Incomplete invoices and vague product descriptions consumed agents’ time, while service-level agreement pressure increased. Those conditions make early, visible checks useful: an agent can spend less time locating routine discrepancies and more time resolving information that needs a human decision.

At EORI, Customaite took on commodity code validation, document checks and data quality flags. Agents retained control of judgment calls and exceptions. Keegan likened the change to replacing a hand saw with a power tool: the person still does the craft work, but can do it faster and more precisely.

One shipment illustrates the division. Its packing list had the gross and net weights switched. Customaite flagged the mismatch before the agent had worked through the entire declaration. The agent then contacted the exporter to clarify the figures before submission. The system identified the inconsistency; a person established what the correct information should be.

The distinction is operationally important. A flag is useful when it directs attention to a specific problem, but it doesn’t establish which value is correct. That requires a source or a person able to confirm the details. Teams considering similar tools can examine whether alerts explain the discrepancy, whether uncertainty is made visible and whether approval remains with a qualified agent.

Data quality becomes easier to see

Within weeks, the system also surfaced longstanding data quality issues linked to certain traders, which the brokerage says had gone unnoticed in its manual process. That finding points to a second use for automated checks: recurring errors can become visible across day-to-day declarations instead of appearing only when an agent happens to catch them.

For operations teams, repeated flags can help direct follow-up toward the documents or data that need clarification. The source doesn’t describe a particular corrective programme, but the workflow makes the problems easier to identify and discuss with the traders supplying the information. Automation can raise the consistency of checks; people still need to resolve the underlying gaps.

The experience at EORI suggests a workable test for customs AI: can agents see why the system made a suggestion, does it flag uncertainty, and does a qualified person approve the submitted declaration? If those steps are built into the process, software can reduce repetitive review while keeping decisions and accountability with staff.

Jana De Borger, a trade compliance and customs software specialist based in Belgium, wrote the source article. She works at Customaite and previously held a senior sales role at AEB focused on customs and trade compliance software and services. The EORI example she describes puts the operating principle plainly: automate repeatable checks, keep human review for judgment, and make sure each declaration has a responsible approver.