A number, with anywhere from six to twelve or even seventeen digits, decides how much duty, and potentially import tax, a product pays, whether it needs a license, whether it qualifies for a trade agreement rate and how long it sits at the border. Too many products cross borders with that number picked in a hurry, by someone who isn’t a classification specialist, from a description written for marketing. Sometimes it is literally copied and pasted from the line above, even if the products are not the same.
What the HS code is
The Harmonized Commodity Description and Coding System, the HS, is maintained by the World Customs Organization (WCO) and used by more than 200 countries and economies, covering over 98% of world trade. It organises goods into 21 sections and 97 chapters, then into four-digit headings and six-digit subheadings. The 2022 edition contains just over 5,600 subheadings.
The first six digits are shared internationally. After that, countries add their own digits for tariff and statistical purposes. The US uses 10-digit HTSUS codes, the EU uses the 8-digit Combined Nomenclature with 10-digit TARIC codes, although Germany uses 11, and the UK uses 10-digit commodity codes for imports. A product can carry the same six-digit subheading in London and Chicago and still have different full codes and different duty rates.
The WCO revises the nomenclature roughly every five years. The next edition, HS 2028, takes effect on 1 January 2028, with changes driven by new technology, shifting trade flows and environmental and social concerns. Codes that are correct today will need checking again.
How classification works
Classification follows six General Interpretative Rules (GIRs), applied in order. Rule 1 says classification is determined by the terms of the headings and any relevant section or chapter notes, not by the titles. Later rules deal with incomplete or unassembled articles, mixtures, goods that could fall into more than one heading, packaging and the step down to subheading level. Section and chapter notes, the WCO Explanatory Notes and national rulings shape the answer.
The difficulty is in the detail. A smartwatch, a heated jacket, a gift set with a candle and a mug, a phone case with a card slot: each needs a judgement on material, function and essential character. A classifier who reads “cotton” when the fabric is 51% polyester, or who misses a chapter note excluding the product, ends up with a code that looks right and isn’t.
What happens when the code is wrong?
Money leaks in both directions. Underpay and the customs authority can come back for the difference, plus interest and penalties, often years after import. In the US, penalties under 19 U.S.C. 1592 scale with culpability, from negligence through gross negligence to fraud, and the statute of limitations runs to five years. In the EU, customs debt can generally be recovered for up to three years after it arose. Overpay and the importer hands money to the treasury that it may not get back, or prices itself out of a market without realising.
Shipments stop. Mismatched descriptions and codes trigger queries, holds and physical inspections. In high-volume parcel flows, one bad code applied across a product line multiplies into thousands of stuck shipments.
Compliance exposure grows. HS codes drive prohibited and restricted screening, sanctions checks, anti-dumping measures and the additional duties now layered onto US imports. A misclassified item can slip past a control it should have triggered. Trusted trader status, such as AEO in the EU and UK, depends partly on a clean classification record.
Trade agreement benefits slip away. Rules of origin are written around tariff classification, so a wrong code can mean claiming a preference that isn’t valid, or missing one that is.
Customers feel it. For retailers selling cross-border, the HS code feeds the landed cost at checkout. Get it wrong and the customer faces an unexpected charge on delivery, refuses the parcel, and the seller pays for the return.
Why general purpose AI falls short
Large language models, like ChatGPT and Claude, are good at producing plausible answers, and a plausible HS code is a dangerous thing. Generic chatbots don’t apply the GIRs consistently, can mix up national extensions, and may return codes that don’t exist in the destination tariff. They also leave no audit trail a customs authority would accept. We put this to the test in our earlier piece on using ChatGPT to classify goods – read more here – and the results showed why classification needs purpose-built tools.
Classification built for the job
Hurricane’s classification engines were built for global trade intelligence, not adapted to it. Our fist fully automated 10-digit classifier Zephyr has classified goods at scale for postal operators, carriers and retailers since 2019. We were the first company to offer full automated 10-digit classification for both import and export.
Our latest iteration, Classification v2, pushes accuracy further across more niche areas, and operate within the B2B and B2G arenas. In side-by-side tests run by customers against competing classifiers and general-purpose LLMs, Hurricane’s engines have consistently come out on top. Our solutions classify to full national tariff level, cope with poor or sparse product descriptions, and return results fast enough for real-time checkout and label creation to be unaffected.
Accurate classification is the first data point in the chain. Duty calculation, prohibited and restricted screening and export control checks depend on it. Get the code right at source and the rest of the process has something solid to work from.
Read more about our classification solutions by clicking here.
Or why not Contact Us to find out how we can help you get your HS Classification accuracy where it needs to be for digital first customs clearance.