Bulk AI HS Classification: How AI Meets the Complexity of HS Codes at Scale

Today, HS code classification still remains mainly manual. This ties directly to the very nature of the process: the requirement for a deep, nuanced understanding of both legal frameworks and real-world context.

Dubai chocolate? Milk chocolate outer coating. Pistachio cream/butter filling. Kataifi (finely shredded filo-pastry strands). Tahini (sesame) paste. The winner? Classified as 'Chocolate and other food preparations containing cocoa, other, in the form of tablets, bars or sticks, filled'.

The bar is high. Product descriptions are frequently incomplete, inconsistent, or lost in translation. Our experts are constantly required to apply GRI, interpret complex customs rulings, recognize data gaps, and infer intent from experience, all while keeping pace with the evolving 2026 regulatory landscape.

This explains why AI adoption in logistics and customs has stayed static. Recent Eurostat data shows that while AI adoption grew across many sectors between 2024 and 2025, logistics recorded no change at all, remaining the lowest measured sector at just 6.1%. Beyond common skepticism, the challenging nature of the customs field has traditionally favored human judgment and expertise.

But technology, especially AI automation, is here to stay. And improving steadily too. The challenges that once restricted AI adoption may now be the reasons that encourage it. The EU is pioneering this shift: the 2026 Customs Reform entails large-scale digitalization, with EU Customs Data Hub creating new grounds for structured, automated data.

The question is no longer whether AI has a role in customs classification. It's now how to make it work reliably at scale.

Manual HS Classification: Where Human Expertise Carries the Load

HS code classification is, at its core, a reasoning task. For every product, a classifier must navigate a rigorous sequence of judgments:

• What is this product, precisely?
• What is its primary function?
• Do the General Rules of Interpretation (GRI) point to multiple headings?
• Are there Binding Tariff Information (BTI) rulings that override standard logic?

For a single product, this is manageable. For a team working through thousands of SKUs, such as a new product catalogue, a market expansion, or an inventory review ahead of the 2026 tariff and annual CN updates, the complexity compounds.

When a specialist reaches their 50th product of the day, accuracy risks slipping. Fragmented spreadsheets and outdated internal databases only add layers of friction. This makes manual bulk classification more than just slow. It makes it risky. The compounding demand of making hundreds of careful decisions in sequence, with the same accuracy required at line five hundred as at line one, places immense pressure on the very expertise the process depends on.

The EU: A Digital-First Customs Landscape

The 2026 EU Customs Reform represents the most significant overhaul of European trade infrastructure in decades. At its heart is the EU Customs Data Hub, a centralized platform designed to create a "single point of truth". Though it is set to come online from 2028, the directional shift towards large-scale digitalization and centralization starts in 2026.

This shift demands more than just faster filing; it demands data governance. The 2026 reform includes substantial nomenclature revisions, particularly for batteries, renewable energy technologies, and advanced materials. This means catalogues classified in 2025 may need review if their product categories are affected by the 2026 CN update.

Furthermore, with the removal of the €150 duty-free threshold for e-commerce, the volume of required classifications is set to increase significantly. Annual CN updates and the data requirements coming with the 2026 reform all point to one conclusion: product data must be structured, reusable, and legally defensible. AI-assisted processes are now a natural alignment with the EU's own digital infrastructure.

How Digicust Handles Bulk HS Code Classification

The difference between processing files and classifying goods in bulk lies in the reasoning. Digicust applies the same legal, GRI-based logic to every line item that a specialist would apply to one, consistently and at volume.

Step 1: Upload Your Catalogue

Submit your Excel or CSV file containing article numbers, descriptions, and SKUs. There is no volume limit; the system is built to process thousands of products simultaneously. Input quality matters: a well-described product typically reaches higher confidence, while vague descriptions are flagged for expert review rather than forced into a wrong code.

A useful product description typically includes the material composition, primary function, and intended use. Commercial descriptions, technical datasheets, and product URLs are all accepted. The more context the system has, the higher the initial confidence.

Step 2: AI Reasoning Chain

For each product, Digicust executes a specialist's workflow:

• BTI Database Search: Cross-referencing legal precedents and existing rulings.
• GRI Reasoning: Systematically working through the General Rules of Interpretation.
• Web Research: If data is missing, the AI searches for technical specifications to fill the gaps where available.
• Dual-Use Check: Automatically identifies potential export controls (e.g., BAFA in Germany) and provides ECCN suggestions (available as part of Digicust's broader export control workflow).

Step 3: High-Precision Review

High-confidence classifications can be approved in bulk. Lower-confidence items are flagged for human specialists. This ensures your experts' time is concentrated where judgment is actually needed. Not distributed thinly across routine classifications.

Step 4: Export and Integrate

Completed classifications can be pushed directly to ERP systems or major platforms like DAKOSY, Scope, and CargoSoft. For those who prefer a familiar environment, Digicust's Excel Add-in allows you to classify directly within your spreadsheet.

Accessing Digicust: Simple, Transparent Pricing

Digicust offers several access points depending on your volume and workflow needs.

• AI Tariff Classification Lite: €1.00 per classification. Prepaid, pay-as-you-go, no complex credit systems. Ideal for small batches or testing.
• Standard Plans: Credit-based (0.5-3 credits per case). API access included for seamless system integration.
• Excel Add-in: Available as a downloadable add-in for direct spreadsheet workflows, classify entire columns without leaving Excel.

It's not manual against AI. It's AI-supported expertise

The manual approach to bulk classification has a ceiling. It's not an expertise ceiling, but a resources one. It's a question of what happens to accuracy and consistency when human expertise is stretched across hundreds of decisions made sequentially, under time pressure, from incomplete data.

The AI-powered workflow doesn't replace that expertise. It removes the conditions that degrade it. What the business gets back is an audit-ready catalogue that holds the same standard at line five hundred as it did at line one.

Amid the annual CN updates and broader demands of the 2026 reform, that consistency is what makes compliance at catalogue scale sustainable.

Do you need bulk AI HS classification? Four signals worth checking

Not every business needs AI automation immediately. But for those that need it, the urgency tends to arrive before the solution is ready. These four signals are worth checking against your own operation. For those that apply, the case for a structured classification process becomes practical rather than theoretical.

• Reactive Growth: If you only classify when a shipment is booked or a customs issue surfaces, your catalogue likely contains hidden gaps and unreviewed assumptions.
• Fragmented Data: When descriptions and historical codes live across multiple ERPs and local business units, inconsistency becomes difficult to avoid.
• Classification Debt: If your codes were assigned years ago by different people without a structured review cycle, you are likely carrying silent compliance risks.
• The 2026 Regulatory Trigger: With the current EU reforms and CN updates affecting batteries, energy storage, renewable energies, and advanced materials, doing it the old way becomes increasingly difficult to sustain for large catalogues. The removal of the €150 customs duty exemption threshold, which is set to take effect in 2026, adds further classification volume to an already demanding environment for e-commerce businesses specifically.

So, what to take away?

HS code classification has always demanded expertise. What's changing is the environment that expertise operates in: more products, more markets, more regulatory updates, and now a structural overhaul of EU customs infrastructure that raises the bar further still.

AI HS code classification doesn't lower that bar. It gives the people responsible for clearing it the conditions to actually do so consistently, at scale, and without the attrition that comes from applying careful judgment to everything equally.

For businesses working through a catalogue review, preparing for new market entry, or getting ahead of the 2026 reforms, it's worth seeing what that looks like in practice.

Talk to the Digicust team, or book a demo to see the workflow firsthand.