Logistics / Supply ChainNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.

Bulk HS-code classification for your whole item master

Added 2026-10-10 · Last updated 2026-10-14

Role

Logistics & trade compliance · Typical users: trade compliance / customs, purchasing, ERP master-data owner

The problem

Hundreds or thousands of part numbers in the ERP have no HS code, or an old one nobody can justify. Every urgent export or import then waits while someone digs through datasheets and the tariff, and two people often pick two different codes for the same item. It is slow work: at Ingram Micro's German branch, classifying one new item by hand could take up to 30 minutes, for about 1,250 new items a month. A wrong code means wrong duties, lost CUSMA/USMCA preference, delays, and penalties or reassessments later.

Step by step

Export the item master: part number, description, material and composition, function/use, specs, drawing or datasheet link, whether it is a part of a machine (and of which machine), BOM parent, supplier, country of origin, and any existing HS code.
Clean and group: remove duplicates, group similar items (same family, same material) so one decision can be reused, and flag items whose description is too vague (e.g. 'BRACKET 3') for enrichment first.
Run AI in batches of 50–200 rows: for each item, a 6-digit HS code plus the Canadian 10-digit and/or US 10-digit HTS line, the GRI reasoning (heading, notes, why not the alternatives), a confidence level and the questions it needs answered.
Check against official sources: look up the proposed line in the CBSA Customs Tariff or USITC HTS, and search CBP CROSS for binding rulings on similar products. Rulings are the best examples of how customs reasons.
Triage: high-confidence items get a quick review by your trade-compliance person; low-confidence, 'parts of machines' (Section XVI notes), sets, and multi-material items go to a licensed customs broker or specialist. For high-value or recurring doubtful items, consider an advance ruling.
Check the money: for each changed or new code, compare duty rates (MFN vs CUSMA/USMCA preference) and any special tariffs or surtaxes, and confirm the origin rules still work, since a different HS code can change the rule of origin.
Write back to the ERP: approved code, date, who approved, source version of the tariff, and the reasoning text, so an auditor can see why each code was chosen.
Keep it alive: classify every new part at creation, re-check when the tariff changes (the HS is revised every few years), and spot-check a sample of AI codes each quarter.

Tools

traide AI (traideCompliance): used at ZwickRoell, MINDA and Ingram Micro
PowerClass.ai (Squadra / PowerSuite.ai): used at Kaemingk
Other AI classification tools: Gaia Dynamics (Canadian 10-digit), iCustoms.ai, and the classification modules of global trade platforms
Any general LLM (ChatGPT, Claude, Copilot) with the prompt below for a first pass on a small batch
Free official references: CBSA Customs Tariff (https://www.cbsa-asfc.gc.ca/trade-commerce/tariff-tarif/menu-eng.html), USITC HTS search (https://hts.usitc.gov/), CBP CROSS rulings database (https://rulings.cbp.gov/home)

Result

ZwickRoell (materials-testing machines, 1,800 employees, Germany): in production since July 2025 for single and bulk classification, over 3,000 products classified, 84% of AI suggestions adopted, each with a full justification and source references (traide AI case study, July 2026). MINDA Industrieanlagen (conveyor and plant engineering, 500–999 employees): 50% less time spent on daily classification; bulk classification of master data fed into the ERP (traide AI, June 2025). Ingram Micro Distribution (Germany, an IT distributor/importer rather than a manufacturer): about 1,250 new items a month; classification time down from about 30 minutes to 5–8 minutes per item; up to 80% of AI suggestions accepted without rework; savings of about half a full-time employee (traide AI, Oct 2025). Kaemingk (seasonal decoration products, selling to 90+ countries): about 9,000 products to classify every six months; Squadra's PowerClass.ai now classifies about 18,000 products a year in batches of about 3,000, with about 90% accuracy (the source writes '±90%') and an average of 675 hours of manual classification saved per batch (case published on the PowerSuite.ai blog, 21 Jan 2025; PowerClass.ai is a Squadra / PowerSuite product). Reality check: in an independent 2025 benchmark built from 200 US CBP binding rulings, the best general model (GPT-5-Thinking) got only 25% of 10-digit codes fully right and 55.5% at 6 digits; a fine-tuned model reached 40% and 57.5%. Without your own specs and a human review step, raw chatbot codes are not good enough. All company figures come from the vendors' published case studies.

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

Legal responsibility for the code stays with the importer (and exporter on export declarations), not the AI or the software vendor.
General chatbots get many full 10-digit codes wrong: in an independent benchmark on 200 CBP rulings, the best general model was fully right only 25% of the time. Use AI to speed up research and documentation, not as the final answer.
Classification follows the General Rules of Interpretation in order (GRI 1 first: heading texts and section/chapter notes). A plausible-sounding explanation is not proof; check the notes.
A code change can change duties, CUSMA/USMCA eligibility, anti-dumping or surtax exposure and export-control checks. If you find past errors, ask a broker about corrections or voluntary disclosure.
Company results here are published by the vendors that sold the tools. Test on 50–100 items you already know the answer to before trusting a tool's accuracy on your data.
Do not paste confidential drawings or customer data into a public chatbot; use an approved enterprise tool.

Try it yourself

You are a senior customs classification specialist (Canada CBSA and US CBP), expert in the Harmonized System and the General Rules of Interpretation (GRI 1 to 6), section and chapter notes, and the WCO Explanatory Notes.

Classify this batch of items from our item master. Import country: [Canada / US / both]. Typical country of origin: [e.g. China, Mexico, US].

CSV (columns: part_number, description, material, function_or_use, specs, used_in_machine, current_hs_code):
[paste up to 100 rows]

For EACH row, return one CSV row with these columns:
part_number | hs6 | ca_tariff_10 (if Canada) | us_hts_10 (if US) | heading_text | gri_reasoning (GRI applied, the heading and section/chapter notes that decide it, and why the closest alternative heading was rejected) | alternative_code | confidence (High / Medium / Low) | questions (the exact facts needed to confirm, e.g. material %, power source, whether it is solely or principally used with a given machine) | current_code_check (Agree / Disagree / None)

Then give:
1. A short list of items grouped by family that can share the same decision.
2. Every Low-confidence or 'parts of machines' item, flagged for a licensed customs broker.
3. Any row where the description is too vague to classify, with what to add.
4. For changed codes: what to verify about duty rate and CUSMA/USMCA origin rules.

Rules: use only the data I give you; never invent specs. If a fact is missing, lower the confidence and ask the question instead of guessing. Mark every 8/10-digit line 'to verify in the CBSA Customs Tariff or USITC HTS'. Suggest a search term for the CBP CROSS rulings database (rulings.cbp.gov) for each Medium or Low item. This is a draft for review by a qualified person; do not change any system.
Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Want more like this? Join the newsletter