Companies are adding AI to global trade workflows at a time when tariff rules are becoming harder to interpret, maintain and apply at speed. The technology can accelerate research and identify anomalies, but its value depends heavily on something less visible: whether the underlying trade systems can keep pace with rapidly changing duties, classifications and exemptions.
Research commissioned by QAD found a striking gap between AI adoption and tariff automation. Ninety-nine percent of surveyed trade, supply chain and compliance executives said their organizations were using or planning to use AI for foreign-trade zone operations, while only 20% reported having software that automatically updates when tariffs change. The same research found 67% had experienced negative FTZ audit findings during the previous 11 months.
That disconnect matters because the workload facing trade teams has expanded substantially. AI can help interpret regulatory material and investigate classifications, but it cannot compensate for a system of record that lacks current tariff rules or cannot consistently apply them to transactions.
The problem starts with the growing number of variables behind a tariff determination. A product’s primary classification may be only the first layer. Country of origin, entry date, additional duties, exclusions and Chapter 99 provisions can all affect what ultimately has to be declared.
The U.S. tariff landscape in 2026 illustrates the difficulty. Trade teams have had to accommodate changes involving Section 232, Section 301 and other tariff authorities while tracking different effective dates and exemptions. Each new measure can create another relationship between a product classification, its origin and the additional tariff provisions that apply.
The result is effectively a continuously maintained rules map across products, countries, tariff programs and exceptions. As the number of combinations increases, manual maintenance becomes harder to sustain without creating delays or errors.
Tariff Changes Are Compressing the Response Window
The speed of regulatory change adds another layer of pressure.
On July 23, the Office of the U.S. Trade Representative announced Section 301 tariffs covering 60 economies over their treatment of imports produced with forced labor. The additional duties, generally set at 10% or 12.5% subject to exemptions and other provisions, became effective at 12:01 a.m. on July 24.
That left companies with little time between final publication and implementation. CBP guidance issued on July 23 provided the Chapter 99 classifications and filing instructions required for entries beginning the following day.
The timing was particularly significant because the new duties arrived as temporary Section 122 tariffs expired. Those tariffs had followed the Supreme Court’s invalidation of earlier tariffs imposed under emergency powers. The July Section 301 measures therefore became another major adjustment that trade systems had to incorporate without an extended transition period.
This exposes an important distinction in the role of AI.
Much of tariff calculation is deterministic. Once classification, origin, entry date and the relevant additional-duty rules are established, software should be able to apply the required logic consistently. Generative AI can assist with researching regulatory changes, reviewing documentation and identifying issues, but the calculation itself depends on structured rules and reliable data.
That makes the system of record critical. If tariff provisions, effective dates and exclusions cannot be captured quickly and accurately, an AI layer has an unstable foundation from which to work.
The QAD findings therefore point to a broader technology issue. Near-universal interest in AI is developing alongside much lower levels of automatic tariff updating. Investment in the intelligence layer may be advancing faster than modernization of the systems responsible for executing compliance decisions.
AI Needs Institutional Trade Knowledge
General-purpose AI introduces another consideration because trade compliance depends heavily on provenance.
Classification decisions, valuation positions, rulings and origin determinations need to be supported by authoritative information and defensible reasoning. A general-purpose model can accelerate research, but its response does not automatically become a reliable compliance record.
A stronger model is to connect AI workflows with a company’s established compliance history. Previous classification decisions, customs rulings, product information and valuation positions can provide controlled context against which new transactions are reviewed.
That institutional memory also creates opportunities for pattern detection.
Instead of asking AI simply to produce a classification or summarize a regulation, systems can use it to identify transactions that depart from established patterns. A different supplier, unexpected origin, unusual declared value or classification inconsistent with similar products can be surfaced for review.
This changes where automation creates value. The objective is not necessarily to automate the final compliance judgment. It is to narrow the field of transactions requiring human attention while giving reviewers better information when an exception appears.
CBP’s position reinforces the importance of that distinction. In a January 2026 headquarters ruling, the agency examined the use of automated tools in preparing customs-entry information. CBP reiterated that identifying entry-related data and making decisions about information intended for filing can constitute customs business, including when OCR technology performs parts of that work.
That creates a boundary for agentic systems. Automation can collect information, compare transactions, identify exceptions and prepare work for review, but organizations still need to determine where licensed or accountable human judgment must remain in the process.
The European regulatory framework similarly places emphasis on oversight for systems classified as high risk. Article 14 of the EU AI Act requires high-risk AI systems to support effective human oversight, including the ability for authorized people to understand limitations, interpret outputs and override or stop a system where appropriate.
For trade technology, that makes auditability as important as automation. Companies need to know what information an AI system used, which rules were applied, where an exception was raised and who approved the final decision.
The more useful architecture is therefore not AI operating independently of the trade platform. It is AI connected to current tariff data, established compliance decisions and a system capable of applying changing rules consistently, with human review positioned where judgment and accountability are required.
Tariff Speed Changes the Value of Good Data
As tariff implementation windows compress, companies may need to treat classification, origin and duty data with the same discipline applied to other critical transaction data. A tariff rule captured days late can affect landed cost, sourcing decisions, broker instructions and customs entries before the error becomes visible. That raises the value of knowing not only whether a system has the latest rule, but when it was updated, which transactions were affected and whether earlier entries need to be reviewed. AI can accelerate that work, but the quality and timing of the underlying tariff data will determine how much of that speed is actually useful.