The Intelligent Border: How AI Is Redefining Customs Risk Management
By Marek Retelski, Head of Global Sales at Webb Fontaine
Customs authorities are struggling to keep pace with global trade. Cross-border e-commerce is driving up daily shipment volumes, and de minimis rules (the thresholds below which low-value shipments can be exempted from duties and controls) are being continually rewritten around the world. Yet none of these changes fully keeps pace with current trade volumes.
Today, tens of millions of low-value parcels move through customs every day, and the number keeps climbing.
This gives criminal networks ample opportunity to exploit the gap, splitting shipments and altering documents to bypass controls.
Why Legacy Systems Can’t Keep Up
The reason those controls can’t keep up comes down to how they’re built. Most customs systems are now at least partly digital, but many still run on rigid, rule-based processes.
If a risk score crosses a threshold, the shipment gets flagged. However, when there is a policy change, engineers must rewrite the rules in code, which is burdensome and time-consuming.
Machine learning-based risk scoring has improved detection, but it’s still governed by the rules people set. Either way, a policy update means manual intervention, and a lag between a new threat appearing and the system catching up.
That rigidity shows up in two places.
- First, in fragmentation. Customs officers often check several separate platforms before they can make one decision.
- Second, in documentation. Invoices, packing lists and certificates vary by trader and by country, and legacy systems can only process structured declarations. Everything else gets reviewed by hand, which is where errors and fraud slip through.
A Volume Problem With No Single Fix
At today’s volumes, manual review isn’t just too slow; it has become unsustainable.
In 2024, US Customs processed 1.36 billion de minimis shipments, a tenfold increase over the previous decade, and they accounted for around 92% of all US import entries by volume.
The EU recorded almost 5.9 billion low-value items in 2025, up from 4.6 billion the previous year.
The US and the EU have both responded to this growth by withdrawing de minimis relief rather than building better screening. Washington ended duty-free treatment for nearly all countries in August 2025, and Brussels replaced its own exemption with a temporary €3 duty per item from 1 July 2026.
Unfortunately, withdrawing de minimis relief doesn’t reduce the volume. It routes every one of those parcels into formal entry instead, and legitimate low-value trade now carries the same friction as everything else.
Argentina went the other way. It raised its courier ceiling from $1,000 to $3,000 in late 2024 and exempted $400 per shipment on up to five personal imports a year, and courier imports promptly rose 274% to $894 million. That trade-off is plain to see: facilitation on that scale only works if every parcel can be assessed for risk, and that is not something a manual process or a rule-rewrite cycle can deliver. For the many authorities that have not withdrawn de minimis, and for those that want to keep facilitating low-value trade without leaving it open to abuse, the answer isn’t fewer parcels. It’s a risk assessment on every one of them.
Where AI-Native Platforms Change the Equation
This is where AI-native platforms can make a difference, because they address the two failure points above directly.
They pull scattered data into a single case file per shipment, closing the fragmentation gap. By combining large language models with machine learning, they can interpret both structured and unstructured documents, surfacing discrepancies that would otherwise wait for manual review. They also analyse transaction histories, trader behaviour and external data to identify emerging fraud patterns sooner.
By contrast, a rules-based system waits for someone to write a new rule.
None of this, however, removes the need for human judgement. Complex or high-stakes cases still need an officer’s eye. The best model splits the load. Automated systems clear routine work; officers focus on what’s ambiguous. Legitimate trade moves faster, and enforcement targets the risks that matter. And because a customs decision can be questioned by an officer or appealed by a trader, the system must show the reasoning behind a flag, not just the flag itself.
Webb Fontaine Zerø in Practice
Webb Fontaine Zero, launched earlier this year at the WCO Technology Conference Abu Dhabi, puts these principles into practice, embedding AI across the customs workflow rather than adding it as a layer on top.
In practice, that means an officer can describe a policy change in plain language and see it reflected in the operational system in minutes rather than months.
The Investment That Compounds
That’s the distinction that matters. Adding AI to old infrastructure treats the symptom. The systems that will keep pace are the ones designed around AI and data governance from the start. This is especially critical for regions pursuing deeper trade integration: as the African Continental Free Trade Area (AfCFTA) pushes member states toward shared trade infrastructure, and ASEAN advances its single-window integration, only systems that are interoperable across borders while keeping each country’s data under its own control will hold up.
Common reference points already exist in the WCO Data Model and the SAFE Framework of Standards, and building to them from the outset costs far less than retrofitting them later. Getting there takes investment, governance and change management, but the cost of propping up legacy infrastructure is only moving in one direction.
Trade will keep getting faster and more complex. The customs authorities that keep pace won’t be the ones bolting AI onto systems built for another era. They’ll be the ones who rebuilt around it.
About The Author
SUBSCRIBE TO OUR NEWSLETTER
