AI in Trade Compliance: Use Cases, Agentic AI and Human Oversight

Trade compliance has traditionally been viewed as an administrative function operating in the background of supply chain operations. But the recent pace of change in tariffs, sanctions, trade restrictions and compliance digitisation means that businesses and customs teams cannot treat this work as purely reactive anymore.

Trade compliance is shifting from a "check-the-box" function towards a more strategic capability.

At the same time, the volume and complexity of modern trade are putting greater pressure on human expertise. Customs and compliance professionals must work across changing tariff measures, sanctions lists, export controls and geopolitical restrictions, often while processing large volumes of product, supplier and shipment data.

AI can change this equation in trade compliance. Its value, however, goes beyond reducing manual work or producing answers faster. When designed effectively, AI can help surface relevant information, explain the basis for a recommendation, identify inconsistencies and exceptions, and involve a professional when human judgement is required.

What AI in trade compliance means

AI in trade compliance is not simply another name for customs software or existing workflow automation.

Many trade compliance systems have long relied on structured databases, deterministic rules and predefined workflows. AI, however, adds a different set of capabilities that are harder to achieve through predefined rules alone. These include interpreting unstructured documents, analysing product descriptions and connecting information that does not arrive in a standard format.

More advanced AI systems can also operate with a degree of autonomy, identifying which information, data sources or tools are needed to complete particular steps in a compliance workflow.

Where AI can add value in trade compliance

Many specific trade compliance processes require professionals to gather information from different sources, interpret incomplete or unstructured data, compare it against regulatory requirements, and identify cases that need further investigation.

AI can support this work across several areas.

Tariff classification

AI can analyse product specifications and technical documentation, compare relevant product characteristics against tariff information and classification rules, and suggest possible HS and/or national tariff classifications.

Document extraction and cross-validation

Commercial invoices, packing lists and transport documents may contain information in different formats or occasionally contain inconsistencies. AI can extract data across these disparate, unstructured files and flag discrepancies, such as quantity mismatches, missing required fields like units, or conflicting item descriptions, before customs declarations are submitted.

Sanctions and restricted-party screening

AI can support the investigation of potential screening matches by analysing names, addresses and other entity information across available sources. Rather than treating every potential match equally, it can help highlight and provide additional context for exception cases that require human review.

Export control

Determining whether a shipment requires an export licence depends on more than just the product itself. AI helps bring together information about products, parties, destinations and intended use to support export-control assessments and identify cases that may require further investigation or licensing review.

Regulatory research

AI can help professionals locate and work through relevant tariff information, rulings, regulatory guidance and other sources without manually searching across multiple systems.

Audit and consistency checking

AI can compare current cases against product master data, previous classifications, supporting documents and internal instructions to identify inconsistencies, missing information or decisions that differ from established records.

Across these use cases, AI is typically designed to help bring together scattered information that would otherwise need to be collected and checked manually, while directing professional attention towards inconsistencies, uncertainty and exceptions.

Generic AI and agentic AI in trade compliance

Not all AI works in the same way. Generally, AI can assist with many of the individual tasks described above, but there is an important difference between asking a general-purpose AI for compliance information and using an AI agent embedded within a trade compliance workflow.

Generic or general-purpose AI tools typically respond to a specific request using the information available in the prompt or conversation. A professional might provide a product description and ask for possible tariff classifications, upload a document for analysis, or ask the AI to explain a regulatory requirement. While useful analysis might be provided, the AI does not necessarily know the company's products, previous decisions, internal procedures or the specific case being processed. In many cases, every prompt is a separate set of data and instructions.

Agentic AI, however, can operate with greater context, coordination and autonomy. When integrated with authorised regulatory sources, master data, case information, company instructions and compliance tools, the AI agent can determine, within its configured workflow and permissions, which information, tools or actions are needed to move a task forward rather than simply returning an answer.

For tariff classification, for example, an AI agent could retrieve existing product data, inspect supporting documents, consult the relevant tariff system, evaluate possible classifications, compare the result with historical decisions and flag inconsistencies for professional review. Within an integrated workflow, the resulting information can then be carried forward into the relevant customs process rather than manually transferred from an isolated AI conversation.

Still, greater autonomy does not make incomplete information reliable or eliminate regulatory uncertainty. Part of the value of agentic AI lies in recognising missing or conflicting information, identifying exceptions and bringing cases that require judgement to human attention.

As AI takes over more routine processing and becomes a sharper tool for navigating trade compliance complexity, human expertise can become more important, not less.

Human-in-the-loop automation in trade compliance

Human-in-the-loop (HITL) automation is sometimes reduced to two opposite scenarios. In one, AI runs the workflow and a professional simply clicks "Approve". In the other, the professional performs almost all of the work while the system handles isolated tasks only when instructed.

In customs and trade compliance, real human oversight is about intentional division of work. Advanced AI can handle much of the routine processing and screening, while professionals retain oversight of the process.

When information conflicts, uncertainty is high or a case requires greater judgement, the workflow should bring the exception to human attention rather than simply pushing it forward. Professionals can then review the underlying information, reasoning and alternatives before deciding how to proceed.

This shifts human expertise towards where it adds the most value: interpreting exceptions, understanding consequences and making judgement calls that cannot be reduced to routine processing.

Process can be automated with AI. Accountability remains with people and organisations.

What explainable trade compliance AI should show

Human review is only meaningful if the professional can understand what the AI has done. A recommendation without its underlying information or reasoning gives the reviewer little basis to verify, challenge or correct it.

Explainable trade compliance AI should therefore make relevant decision context visible, including:

• Source information: the documents, product data and other inputs used in the analysis.
• Regulatory references: the tariff information, rules, measures, rulings or other sources relevant to the recommendation.
• Reasoning: how the available information led to the proposed result.
• Alternatives: other classifications or interpretations considered where relevant.
• Missing or conflicting information: gaps or inconsistencies that could affect the outcome.
• Uncertainty and review signals: where available information strongly supports a recommendation and where additional review may be necessary.
• Decision history: previous decisions, corrections and human input relevant to the case.

Explainability therefore supports more than transparency. It gives professionals the information they need to verify the AI's work, investigate exceptions and exercise their own judgement before consequential decisions move forward.

Real-world example: AI-assisted classification with human-in-the-loop

In one recent tariff-classification case, a user was working with a piece of clothing described as being made from "bamboo fibre" under a historical classification that had been used for several years.

Digicust AI challenged the existing assumption based on textile-specific classification considerations. Commercially marketed bamboo fabric can be made from regenerated cellulose, such as viscose, rather than mechanically processed natural bamboo fibre. This distinction can affect how the product is classified.

Rather than accepting the recommendation automatically, the user questioned it and reviewed the issue further. The classification was subsequently checked again with the customs authority, which confirmed the revised classification.

The workflow was:

The value of the AI in this case was a second pair of eyes, rather than simply producing a different tariff code. It identified an assumption worth reconsidering, provided the reasoning behind that recommendation and gave the professional a basis for further review. Human judgement and subsequent verification then determined the outcome.

Bamboo Fibres: Man-made or Natural Fibres?

✓ Clear result: Bamboo fibres are man-made fibres (viscose/rayon)

Yes, “bamboo fibre” underwear is made of man-made fibres. Here is the reasoning:

• Mechanism: While bamboo is a plant, the so-called “bamboo fibre” fabric is chemically processed. The bamboo pulp is dissolved and spun into viscose/rayon (regenerated cellulose fibres). This is a chemical process, not a mechanical one.
• Legal classification: In both the EU and the USA, these fibres are classified as regenerated cellulose fibres (man-made fibres), not as natural fibres.
• Only mechanically obtained bamboo fibres (very rare, similar to linen) would be natural fibres. This is not the case with regular “bamboo underwear”.

! Consequence for the Customs Tariff Number

The existing HS code needs to be checked:

Position	Item	Previous classification	Correct classification
	Bamboo fibre men's underwear	possibly cotton?	6107 1200 000 – Briefs/underpants made of man-made fibres, knitted
	Cotton men's underwear	Cotton ✓	6107 1100 000 – made of cotton
	Polyester men's underwear	Man-made fibres ✓	6107 1200 000 – made of man-made fibres

i

Both customs rates (12%) are identical, so the customs burden does not change – but the correct tariff number for bamboo viscose underpants is 610712000000 (man-made fibres), not 610711000000 (cotton).

!

Note
This question concerns the anonymised historical case (CH0370326-1), not the present case (CH0370326-2). The positions (running shorts 610343000000) are not affected by this.

A step further: turning human feedback into reusable instructions

Human-in-the-loop can also extend beyond reviewing individual cases. When a user identifies an issue or establishes a preferred operating rule, that feedback can be recorded as an instruction for future workflows.

In the same case, the user specified how the AI agent should handle cases where no net weight is available in the underlying documents. The instruction was updated and made available for future cases, rather than remaining a one-off correction.

Instruction change

Applied

The user wants a fallback of 85% of gross weight when no net weight is available in the documents, instead of leaving the field empty.

Set net mass = 85% of gross mass.

✓Rule updated. The existing instruction “Never set gross = net” has been supplemented:

“If no net weight is specified in the documents, use 85% of the gross weight as tare weight.”

Applies immediately to all future cases.

How Digicust approaches AI-assisted trade compliance

Digicust AI Suite brings AI agents into existing customs and trade compliance workflows, connecting documents, product and master data, regulatory information, internal instructions and human review within a structured process.

The platform supports workflows including:

• AI tariff classification: analysing product information against relevant tariff data and supporting expert review.
• Document processing: extracting and validating information across invoices, packing lists, transport documents and other records.
• Sanctions and restricted-party screening: screening party information and supporting the review of potential matches and exceptions.
• Export control: bringing together relevant product, party, destination and end-use information for compliance checks.
• Audit and consistency checking: identifying discrepancies, missing information and decisions that may require further investigation.
• Reusable knowledge: incorporating relevant expert instructions and previous decisions into future workflows.
• System integration: connecting the resulting customs data with existing ERP, TMS, declaration and broker systems.

The objective of the Digicust AI Suite is to give customs professionals a structured, AI-assisted workflow in which routine processing can be automated, exceptions can be surfaced, and expert judgement can be applied where it matters.

Conclusion

AI is expanding what can be automated across trade compliance, from tariff classification and document processing to screening, export control and consistency checks. But greater automation does not reduce the importance of professional expertise.

The real opportunity lies in combining AI's ability to process and connect information at scale with human judgement where uncertainty, exceptions and accountability matter most. As AI becomes more capable and agentic, effective human oversight and explainability become more important, not less.

FAQ

What is AI in trade compliance?

AI in trade compliance refers to the use of AI to support tasks such as tariff classification, document processing, sanctions screening, export-control checks, regulatory research and compliance review.

How is agentic AI different from generic AI?

Generic AI typically responds to individual requests using the information provided to it. Agentic AI can operate with greater context and autonomy, using available data sources and tools to coordinate multiple steps within a defined workflow.

Can AI automate trade compliance?

AI can automate or support many routine trade compliance processes, but not every case can be reduced to automation. Ambiguous information, exceptions and higher-risk decisions may still require professional review and judgement.

What is human-in-the-loop in trade compliance?

Human-in-the-loop (HITL) combines automated processing with professional oversight. Routine work can be handled by AI while exceptions, uncertainty and cases requiring greater judgement are brought to professionals for review.

Why is explainable AI important in trade compliance?

Professionals need enough information to understand and evaluate AI-assisted recommendations. Relevant source data, regulatory references, alternatives, uncertainty and decision history can help reviewers verify results and exercise their own judgement.

Can AI perform tariff classification?

AI can analyse product information, consult relevant tariff data and assist in evaluating possible classifications. Final treatment should reflect the applicable tariff rules, available product information and appropriate professional review.

How does Digicust use AI for trade compliance?

Digicust uses customs-specific AI agents alongside documents, master data, regulatory information, internal instructions and workflow tools to support processes including tariff classification, document processing, screening, export control, consistency checking and downstream customs workflows.