Can you trust an AI with your customs declaration?

Spoiler alert! Yes.

You can trust AI with your customs declarations…

as long as you never ask it to carry the responsibility. That distinction is what separates a useful customs AI from a risky one.

Customs is not a field where “mostly right” is good enough. A wrong commodity code can mean underpaid duty, lead to a post-clearance audit or result in a penalty. A missing certificate can hold a shipment at the border for days. So, when brokers ask us whether they can trust an AI with their declarations, it is the right question to ask. The honest answer depends on what you are trusting it to do.

Compliance doesn’t care who typed the data

The law is clear about who answers for a declaration. Under Title I, Chapter 2, Article 15 of the Union Customs Code, whoever lodges a declaration is responsible for the accuracy and completeness of its content and the authenticity of supporting documents. The UK applies the same principle post-Brexit. Customs authorities don’t ask whether a person or a piece of software entered the data. They ask who submitted it.

That makes three things non-negotiable for any AI in customs:

• Accuracy you can verify. Every suggested classification, value or origin claim must be traceable to the source document it came from.
• An audit trail. When an auditor asks why a code was chosen two years ago, you need a record of what was proposed, what was checked and who approved it.
• Consistency. Rules applied the same way on the thousandth declaration of the week as on the first.

An AI that produces answers without showing its working fails all three, however clever it may be. An AI built to support verification passes them and often does better than a tired human at 6 p.m. on a Friday.

Accountability stays where it belongs

Because liability sits with the declarant, accountability cannot be delegated to an algorithm. A broker acting as indirect representative is jointly liable for the debt. A trader using direct representation still answers for the data it supplied. No software license changes that.

That reality should shape how any customs AI is built. A good tool works on a simple principle: the AI does the groundwork; the agent makes the call. It reads invoices and packing lists, validates commodity codes, checks supporting documents, and flags data-quality issues. It should never decide on its own that an ambiguous product description is good enough. Instead, it raises the question, shows the evidence behind it and leaves the answer to a qualified person. It also records who reviewed and approved each declaration before it goes out.

That split is not a limitation. It is what makes the output defensible. When every judgment call has a named person behind it, your compliance position is as strong as it was under a fully manual process, and usually stronger, because fewer errors slip through unnoticed.

The human in the loop in practice: EORI UK

EORI (UK), a customs brokerage working with AI platform Customaite, shows what this looks like day to day.

The customs brokerage handles thousands of declarations every week. Before Customaite, its process was fully manual: line-by-line commodity code checks, document validation and data corrections. Reliable, but slow, and entirely dependent on human attention.

As Trade Solutions Manager Robert Keegan put it, the core problem was “rising volume combined with inconsistent trader documentation.” Agents lost hours to incomplete invoices and vague product descriptions, and SLA pressure kept growing.

Customaite took over the repetitive layers: commodity code validation, document checks and data quality flags. Agents kept control of the judgment calls and exceptions. Keegan compares the shift to trading a hand saw for a power tool. The craftsperson still does the work, just faster and more precisely.

One real shipment makes the point. A packing list error had swapped the gross and net weights. Instead of an agent finding it only after working through the whole declaration, Customaite flagged the mismatch early. The agent then clarified it with the exporter before submission. The AI spotted the problem; a person resolved it.

The results:

• Noticeably more consistent SLA performance
• Higher throughput without extra headcount
• Agents handling more work with less pressure, and more time for exceptions and customer queries

There was an unexpected benefit too. Within weeks, the system exposed long-standing data quality issues from certain traders that had gone unnoticed under the manual process. Automation did not lower the bar on compliance. It raised it.

So, can you trust it?

Trust an AI with the work, not with the responsibility. Let it read every line, cross-check every document and flag every inconsistency, tirelessly and consistently. Keep the decisions, and the signature, with your people.

When you evaluate any customs AI, ask three questions. Can I see why it made each suggestion? Does it flag uncertainty rather than hide it? Does a qualified agent approve what gets submitted? If the answer to all three is yes, the AI is not a compliance risk. It is a compliance asset.

That’s how the model EORI UK runs today: repetitive work is automated, and judgment stays human.

This article was first published on LogisticsMatter.