Brokers, Customs Navigate Through Early Uses of AI

“We’re going to see a lot more guidance from the government, but I think it’s going to be at a very high level,” said customs broker-consultant Tom Gould, a member of CBP’s Commercial Customs Operations Advisory Committee. “In other words, I don’t see the government saying, in order to use the AI to classify, you have to have A, B and C. They're going to be more about, [if] any applications use AI, [you] need to consider these types of concepts. That to me seems like the direction that the government's going to go. And I think that's appropriate. I think we’re still far away from the government coming in and putting out guidelines.”

For its part, CBP recognizes that customs brokers use AI, and while it hasn’t issued specific guidelines governing its use, a spokesperson has said customs brokers should not rely on AI alone to do their work.

“CBP understands that Artificial Intelligence (AI) tools exist in the trade industry and brokers may be relying on those tools in varying degrees for classification purposes,” an agency spokesperson said. “A licensed customs broker must exercise responsible supervision and control over their customs business, which can include classification. If a broker is relying solely on AI output for tariff classification and for purposes of filing the entry of merchandise with CBP, he/she likely is not exercising responsible supervision and control by failing to verify proper classification.”

How customs brokers -- and CBP and other partner government agencies -- use AI will likely continue to evolve as the technology evolves, the experts said.

“There is definitely an ongoing discussion as to whether AI and/or automation is going to supplant the customs broker, the intermediary, etc. I think a lot of individuals right now are considering the fact that while it's a useful tool, you still need the customs broker, you still need the intermediary to play a role, to manage that data and make the final decisions as to what data is going to be provided to the government agencies that enforce international trade, particularly CBP,” Lenny Feldman of Sandler Travis said.

Using Automation and AI

Within customs, automation is an emerging technology to speed up data entry. The documents that automation tools could scan range from commercial invoices to bills of lading, as well as packing lists and PDFs of emails from the thousands of documents received from different suppliers, importers, carriers and forwarders.

“One way that the intermediaries, specifically the brokers, have found automation to be very productive and useful for them is when you have thousands and thousands of pages of documents, and you have a tool through OCR [optical character recognition] or another type of technology that is able to take the data, which for years was those paper documents, and digitize them so that you're able to manipulate more data,” Feldman said.

A step beyond automation is AI, which seeks to find trends within the data being analyzed.

“These emerging technologies can rapidly and continuously process vast amounts of historical classification data, suggest accurate HTS codes, automatically populate customs documentation, and flag potential compliance issues before they become problems,” said Victor Borges of trade compliance AI startup Gaia Dynamics. "This automation not only reduces the risk of human error but also allows brokers to handle higher volumes of transactions while focusing their expertise on more complex cases that require nuanced interpretation or special handling.”

AI eventually might be able to do these types of classification functions, but humans will still need to oversee the process, according to Gould. For instance, a commercial invoice might have a description of a men’s T-shirt, but a broker will know that CBP needs more information for it to be classified properly. That information might be details about the shirt's fabric and how it was designed or made, which can be collected only by contacting the company directly.

“While AI excels at handling most situations and can process information faster than humans, just as a self-driving car might struggle with unexpected road conditions or unusual traffic patterns, AI in customs brokerage might misclassify unique products or miss crucial regulatory nuances,” Borges said. “That's why maintaining human oversight and expertise remains critical -- AI should be viewed as an enhancer of human capabilities rather than a replacement for professional judgment.”

In response to these potential uses of AI by customs brokers, a CBP spokesperson said, “Classification for purposes of identifying merchandise being entered into the United States and determining the appropriate rates of duty, taxes, and fees, requires expertise that must be provided by licensed customs brokers.”

There hasn’t been any federal guidance yet defining specific procedures for using AI in customs, but some parameters have been set.

In December 2023, CBP ruled in HQ H326926 that if a foreign, unlicensed company reviews shipping documents, determines and extrapolates the necessary entry-related data and keys the data into the broker’s API software system, then that would be impermissibly conducting customs business, according to Feldman.

“A very important part of the analysis is, who does the data belong to? Who owns that data? And the question would be really, is it the importer? Because the importer is able to do what they want with their data. You even have nonresident importers who are able to be self-filers,” Feldman said.

Feldman continued, “However, if you are acting just as a third party -- you’re not an importer, you’re not related to the importer, you’re not a customs broker -- that’s where you have to be careful so that you're not bumping up against those requirements, because then really you’re overseeing. You’re in charge. You’re only owning that data as a third party.”

How Government Agencies Use AI

As government regulators consider if and how to regulate the use of AI in customs and trade, federal agencies are also grappling with how to use AI to improve import compliance.

“I think CBP is already exploring automation and artificial intelligence. I think they’re looking at everything, from trying to determine the country of origin and the proper value of de minimis, to trade patterns and thresholds,” Feldman said. “There’s a lot of discussion in the trade community, within customs, as to how AI can really detect forced labor in supply chains, because there you really have to go several tiers into the supply chain to make sure that your imported products are not wholly or partially containing forced labor. And artificial intelligence does provide a tool. It provides indicators. I see AI as a really good indicator as to where these risk areas may exist and where CBP needs to devote their attention so that the haystack starts to become smaller in which to find the needle that they're looking for.”

In the de minimis space, one area where CBP is struggling is how to identify multiple packages going to the same person. Another is pinpointing situations where the data for entry is insufficient, and so a package shouldn’t be let into the U.S. AI might be able to come into play in these areas, Gould said. AI also could provide supply chain data at a granular level and enable better risk management.

“If there's one package in the morning going to Jim Jones and another one in the afternoon that's going to James Jones at the same address, is that father and son or is that the same person? Or if something is going to ABC boutique in Portland, Maine, and another package going to ABC boutique in Portland, Oregon, is that the same company, or are those two separate companies?” Gould said. “That kind of decision-making is stuff that Customs is struggling with now, but maybe artificial intelligence can help them start to do a better job of answering those types of questions.”

Both customs brokers and government might leverage AI to automatically research relevant SKU details that humans might miss.

AI could enable the trade community and Customs to analyze and assess more meaningful data at the SKU level, according to Feldman. “Right now, you look at an account and you see all this different merchandise, and you see an entry for all this different merchandise, different types of SKUs and products coming in. Artificial intelligence should be able to allow the trade community, the service providers and the government, to really drill down more on the data as it pertains to that item.”

Another use of AI is in compliance screening, efficiently analyzing shipment details against restricted party lists and identifying regulatory requirements in real time, Borges said.

“This real-time regulatory monitoring capability is becoming increasingly critical, especially as we approach potential significant policy shifts under the incoming administration. With expectations of substantial changes to trade policies, tariffs, and regulations, AI's ability to continuously update its knowledge base and instantly adapt to new requirements will be invaluable for businesses navigating these changes,” Borges said. However, AI works best as an augmentation tool in partnership with experienced professionals who provide final oversight and decision-making, he added.

Other areas where customs authorities could leverage AI are risk assessment and targeting, image analysis of container X-rays, valuation verification against market benchmarks, and document authenticity verification, according to Borges.

But as AI use becomes more sophisticated within government and among customs brokers, regulators will have to watch out for AI hallucinations, which occur when the AI application makes incorrect or illogical conclusions.

Hallucinations are a “critical concern given the high stakes of customs compliance,” and so “the next frontier will likely focus on creating robust validation standards and explainable/auditable AI systems that can auto-reflect and document their decision-making process for audit purposes,” Borges said.

Meanwhile, CBP's new requirement calling for customs brokers to earn continuing education credits to maintain their licenses ensures that brokers will always be up-to-date with the latest developments, and as such, they will not be supplanted by AI, according to Gould.

“Those brokers who are going to survive are going to be the brokers that are going to be acting more as advisers. They’re going to be the ones working with the companies to make sure that the companies understand that things are being done right,” Gould said.

Furthermore, “each PGA is looking more and more at their mission or what they need to do, and they’re looking at using technology to allow them to do a better job of their mission -- which, again, is going to require advisers on the trade side to help those companies that import products subject to different PGAs do a better job of getting the right data to the right agency at the right time so that the agency can make their decision and not stop the shipment when it's moving,” Gould said.