How can AI help with customs and HS code classification for apparel?
AI can suggest HS codes from product data and flag risky classifications, but fibre content, construction and legal responsibility still need human checks.
KEY TAKEAWAYS Summary by the editors
- The Harmonized System is a six-digit product nomenclature used by over 200 countries and economies as the basis for customs tariffs and trade statistics, and the WCO updates it every 5 to 6 years.
- For apparel, the main determinants of classification are knitted or woven construction (Chapters 61 and 62), the predominant fibre by weight, and the garment's gender category.
- AI can propose codes and flag missing or inconsistent product data, but small changes in declared fibre percentages can change the code, so inputs must be verified.
- Common errors include classifying knitted garments as woven and relying on supplier fibre declarations without independent testing.
- The importer, not the software, remains responsible for the declared classification, so a documented human review step is needed.
How can AI help with HS code classification for apparel?
AI can read product descriptions, bills of materials and technical data, propose a likely Harmonized System code, and flag records where data is missing or contradictory. It is a drafting and checking tool. Responsibility for the declared classification stays with the importer or its customs broker.
This how-to covers what determines apparel codes, how to prepare the data and how to build a review process.
What is the HS and why does classification matter?
The Harmonized System is a multipurpose international product nomenclature developed by the World Customs Organization. Each commodity group has a six-digit code, and over 200 countries and economies use it as a basis for customs tariffs and trade statistics. The WCO's HS Committee prepares amendments every 5 to 6 years. Countries add further digits for their own tariffs, so the full code used at the border differs by market.
The code determines duty rate, import conditions and statistical reporting. A wrong code can mean underpaid or overpaid duty, delays and penalties.
What decides the code for a garment?
A customs compliance guide summarises the main factors for textiles. Fibre content follows the chief weight rule, under which a textile mixture is classified as if consisting wholly of the fibre that predominates by weight. Where no fibre exceeds 50%, the essential character rule applies. Small shifts in fibre percentages can change the code, so declared composition should be verified, especially near a threshold.
| Factor | Why it matters | Data AI needs |
|---|---|---|
| Knitted or woven | Chapter 61 covers knitted or crocheted apparel and Chapter 62 woven apparel | Construction method, fabric specification |
| Fibre content | The predominant fibre by weight drives the subheading | Verified composition by weight, including the shell fabric |
| Gender category | Men's and boys' garments are separated from women's and girls' at subheading level | Design, sizing and retail presentation |
| Garment type | Heading depends on item such as jacket, shirt or trousers | Clear product description and images |
| Sets and composites | Matching garments worn together may classify as a set | Packaging and sales configuration |
The same guide lists common pitfalls: classifying knitted garments such as polo shirts as woven, ignoring exceptions for metal threads and elastomeric yarns, classifying down-filled jackets by the fill fibre rather than the shell fabric, labelling solid-dyed fabric as printed, and relying on supplier fibre declarations without independent testing.
What product data does an AI classifier need?
- A precise description of the garment, including construction, closures and intended wearer.
- Verified fibre composition by weight for the outer fabric and for linings or fillings where relevant.
- Images and technical drawings, so that the model or reviewer can check the description.
- Country of origin and destination, since national subheadings and duty rules differ.
- Previous declarations and any binding rulings, to build consistency across seasons.
If product data sits in several systems with different attribute names, consolidating it is usually the first task, and often the largest.
How should a classification workflow with AI be built?
- Collect product data at style level, with fibre composition confirmed by test reports or reliable supplier documents.
- Let the model propose a code with its reasoning and the data fields it relied on.
- Flag records where fibre content is near a threshold, construction is unclear or a sensitive category applies.
- Route flagged records to a trained customs specialist, and approve all new product types manually.
- Store the decision, the evidence and the approver against the style, and review a sample regularly.
- Feed corrections back, so that recurring errors are caught earlier.
Consistency is a benefit in its own right. A model that applies the same logic to every style can expose cases where similar products were declared under different codes in different seasons.
Training and calibration matter as well. Classification models perform best when they have been checked against a company's own history of reviewed declarations, since product descriptions and naming conventions differ between brands. Starting with a sample of styles where the correct code is known, and measuring how often the model agrees, gives an early view of reliability. Disagreements are valuable, because they either reveal model weaknesses or expose historic inconsistencies in declarations.
It is also worth separating routine from difficult cases. Repeat styles with stable composition can be handled with light review, while new constructions, unusual fibre blends, sets and composite items deserve specialist attention. Many teams set thresholds, such as fibre shares close to a boundary, which automatically route a style to a specialist. This keeps expert time focused where the risk of error and the financial effect are greatest.
Finally, keep in mind that rules differ by market. A code proposed for one destination may need adjustment for another, because national subheadings and duty regimes vary beyond the first six digits. Product data should therefore be kept in a form that can be reused across markets, with the market-specific decision recorded separately.
What is changing in customs, and what are the limits?
Customs data requirements are shifting. The European Commission's customs reform proposes an EU Customs Data Hub, with dates running from 2028 for e-commerce consignments to 2038 when it becomes mandatory, and it will abolish the EUR 150 duty exemption. It also plans to reduce duty categories for common low-value e-commerce goods from thousands to four. Those dates and details may change as the proposal progresses, so teams should check current status.
Tariffs also raise the stakes. McKinsey's State of Fashion 2026 reports that tariffs are the top hurdle cited by executives in its survey, so classification accuracy has direct cost consequences. The limits of AI are the quality of input data and the lack of legal accountability. A model cannot resolve an ambiguous composition, and it cannot be questioned by customs authorities. Treat it as a way to prepare, check and document, with a qualified person making the final call.
Frequently asked questions
What is an HS code?
It is a six-digit code from the Harmonized System, a nomenclature developed by the World Customs Organization and used by over 200 countries and economies for customs tariffs and trade statistics. Countries add digits for national use.
How are knitted and woven garments classified?
Chapter 61 covers knitted or crocheted apparel and Chapter 62 covers woven apparel. The same type of garment can have different headings and duty rates depending on construction.
Can AI choose the correct HS code for me?
It can propose a code and flag doubtful records, but it depends on accurate product data and does not carry legal responsibility. A qualified person should approve the declaration.
Why does fibre content matter for customs classification?
Under the chief weight rule, a textile mixture is classified by the fibre that predominates by weight. Small changes in declared percentages can therefore change the code, which is why composition should be verified.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
