Why the Customs Broker Is Becoming the CPA of Tariffs

In the 1980s, a US customs broker was, in Tom Gould's words, "a guy with a bicycle and a clipboard, sitting in a port, writing stuff down." The job was data entry. Then came typewriters, computers, the ACE system, and successive layers of automation. Today's customs broker is most of the way through a different transformation, one the industry has not yet fully named. It runs from data-entry clerk to strategic advisor, the CPA equivalent for the tariff schedule.

On Episode 4 of the Trade and Tech podcast, Gaia Dynamics' Chief Strategy and Compliance Officer Tom Gould and Caspian Co-founder and CEO Justin Sherlock made the case for what that transformation requires, what it makes harder, and where AI fits in. Sherlock brings an unusual combination to the conversation: investment banking at Barclays, private equity at TA Associates, senior director at Flexport Capital, and a licensed customs broker. Caspian calls itself the first AI global trade advisor, starting with duty drawback and expanding into IEEPA refunds, protests, post-summary corrections, and reconciliation.

How the Customs Broker's Role Is Shifting From Clerk to Strategic Advisor

The clerical work that defined the broker's job for forty years is being automated. What replaces it is not "no broker", it is a different broker.

"Where I see the role of the broker today is really changing because the automation of the clerical tasks is happening as we speak," Gould said. "A customs broker needs to be the strategic advisor to companies as to how they can take advantage of the tariff rules. Kind of like a CPA is an advisor on how to strategically take advantage of the tax rules."

The math behind that framing has shifted decisively. With average effective US tariff rates climbing into double digits across most product categories, tariff strategy now matters more to corporate margin than tax strategy in many import-heavy businesses. "We're starting to see companies realizing that strategizing over taxes can sometimes have a negative impact on tariffs, to the point where by paying less taxes they're actually paying more in tariffs," Gould said.

Sherlock added the talent dimension. The scope of professional services work is being elevated in ways that are hard for every firm to absorb. "The expectation is rising of what an individual person is capable of producing. That's a real challenge to up-skill your entire workforce, to adopt new tooling, and to go into the office every day and do high-leverage work instead of doing the paperwork you might be accustomed to."

The shift is visible in Caspian's own sales calls. Two years ago, the questions being asked were "have you heard of duty drawback?" Today the conversation starts with systems questions: how are you doing returns management, where is your sales data stored, are your forwarders filing on a product-level basis per entry line? The customer arrives with prior knowledge from off-the-shelf AI tools and the broker has to elevate immediately or lose the deal.

Sherlock cited roughly 500,000 US importers and exporters as the addressable population. Fewer than 10% have in-house trade compliance teams. The rest need accessible expertise fast and cannot afford a Big Four engagement. Whether the delivery vehicle is Caspian, Gaia's classification and compliance platform, or a broker firm running modern tooling, the gap between what most importers need and what they can buy in 2026 is what trade technology is racing to fill.

What AI Gets Right, What It Gets Wrong, and the Last-Mile Problem

Sherlock gave the clearest articulation of the AI limit across four episodes of the podcast.

"You can get to 80%, you can get to 90%. Getting to 100%, having a product that operates with 99.999% accuracy on any given regulation when the regulations are changing so quickly, is very challenging. There is a final mile of service delivery that will be human in the loop for a very, very long time."

His framing of Caspian's value proposition follows directly. "Any software company is capable of getting to 80% or 90%. What's really strategic is when you compare great technology with super-humans who are experts to deliver the outcome to the customer."

Gould added the regulator-versus-industry definitional gap as a persistent source of classification complexity. "When you talk to people in the apparel industry, they look at a garment and they say that's a t-shirt. When you talk to somebody in customs responsible for regulating the import of t-shirts, they might say, that's not a t-shirt, that's a pullover." AI can surface and often resolve those splits, but the final call and the legal exposure sit with the licensed broker.

Which is exactly what CBP signaled in its most recent ruling on AI in customs business. Gould read the ruling as a clarification, not a new restriction. "Customs' longstanding position is that somebody doing customs business on behalf of another has to be a licensed customs broker. As long as the customs broker is still making the final decisions, customs is okay with it."

Sherlock added a less-discussed wrinkle. CBP has rulings on offshore labour performing data processing in customs work and now a parallel ruling on OCR and AI doing the same. Neither has been aggressively enforced in isolation. "I think they'll be used as part of enforcement action for other things going wrong in investigations," he said. The implication for vendors and importers is clear. Assume these rules will be enforced eventually, and make sure a licensed broker is genuinely in the loop and not just nominally on the org chart.

The Risks No One Is Pricing Right

The episode closed with two questions that turn the usual AI-risk conversation on its head.

The first was what is fundamentally hard to automate beyond the accuracy problem. Gould's answer was counterintuitive: communication between systems, not between humans. His analogue was the protocol problem in banking before Plaid and in healthcare before MyChart. Trade has the same problem, made worse by a CBP decision in the 1980s to maintain the legacy ACE communication protocol. The result is message sets of 80 characters, all uppercase, with a limited character set. That constraint is still live today.

Sherlock added the operational reality. At a recent NCBFAA conference, the core problem was validations that may or may not run inside entry-filing software, may or may not run inside ACE, and may or may not align with current regulations. "These three elements are disconnected: the regulations, ACE, and the entry filing. In some cases, two of the three are correct and the other one's wrong. It's never the same set." Building software that survives across all three, in a year when the HTS schedule changed thirty-one times, is the operating constraint trade technology has to solve. Gaia's tariff classification governance guide outlines what brokers and importers should build before the next regulatory shift forces it.

The second question was about the balance between innovation and risk. Sherlock's answer was the line of the episode. "There's risk for adopting new technology, but there's also risk for doing nothing. The regulator has an OpenAI and a Cloud contract. They're using this technology every day to audit all of us in the industry. So what's riskier is actually a really interesting question."

Sherlock's lightning-round close put the same point three ways:

• Misconception about global trade: that we will revert to a free-trade economy

• Underestimated trend: the adaptability of younger workers entering an environment with very low entry-level demand

• Advice to importers: store and centralize 100% of your trade documents and customs entry data in your own systems

On that last point, Stefanutti and Gould agreed, reluctantly. Document custody is the prerequisite for every refund, protest, audit defense, and supply-chain shift importers will run over the next twelve to eighteen months. The teams that own their data move. The teams that do not wait on someone else.

Conclusion

The customs broker's transformation is not a future scenario. It is happening now in sales calls, hiring decisions, and conference conversations. The clerical layer is being automated. The strategic advisory layer is being built. The brokers and compliance teams that close that gap with the right technology will capture the next decade of value in this market.

The working principle is consistent across every theme in this episode. AI does the structural work: narrowing codes, surfacing rulings, flagging compliance layers. A licensed broker reviews, decides, and signs off. The documentation trail makes that pairing visible to CBP. And the data has to live in systems the importer owns.

See the Platform Built for This Shift

Gaia Dynamics builds AI specifically for customs and trade compliance, designed for the operating environment Gould and Sherlock described. The platform supports the advisory shift directly:

• HTS classification from 18,000+ codes narrowed to the most relevant candidates in seconds, with audit trails for CBP review

• Automatic layering of Section 232, Section 301, IEEPA, AD/CVD, and PGA flags on top of the base HTS code

• Historical ruling lookup that surfaces CBP precedents faster than manual Federal Register searches

• Importer and exporter compliance workflows built for teams that need strategic advisory output, not just raw classification

Standalone deployment gets value in days. The licensed broker stays in the decision loop. The documentation trail is there when CBP comes asking.

Explore Gaia Dynamics and see how the platform supports the broker's shift from data entry to strategic advice.

Listen to the Full Episode

Listen to Episode 4 of the Trade and Tech podcast for the full discussion, including Justin Sherlock's view on why the customs market has only recently emerged as a category, the cross-industry parallels with Plaid and MyChart, and the talent shifts heads of compliance are now planning around.

Frequently Asked Questions

Is the customs broker role being eliminated by AI? 

No, the clerical layer is being automated, but the licensed-broker accountability CBP requires is not going away. The role is shifting into strategic advisory work: classification strategy, tariff engineering, and refund and protest workflows. Brokers who treat AI as a productivity tool are positioned well. Brokers who treat it as a threat are not.

Did the recent CBP ruling on AI in customs change what is allowed? 

In Tom Gould's reading, no. The ruling clarified that customs business on behalf of a third party requires a licensed broker in the decision loop. AI can do the structural work (classification suggestions, ruling research, data extraction) but a licensed broker must sign off and maintain documentation consistent with reasonable care.

What is the single most actionable takeaway for an importer? 

Store and centralize 100% of your trade documents and customs entry data in your own systems. Document custody is the foundation for refunds, protests, audit defense, and supply-chain restructuring. Importers who own their data move. Importers who depend on others to retrieve it wait.

Where should importers start if they want to work effectively with AI-driven customs brokers?

Start with data ownership. Centralize trade documents, entry filings, product data, and historical classifications in systems you control. AI and brokers both rely on clean, complete records to produce accurate classifications, defend audits, and unlock refunds. Without structured data, even the best tools and advisors operate with gaps.