Short answer: HTS classification software comes in five kinds: lookup and search tools that find tariff text, rules-based engines that apply coded decision logic, machine-learning suggesters that predict a code from a description, expert-reviewed services that pair an engine with human classifiers, and classification modules inside global trade management (GTM) and ERP suites that store and maintain codes. Choose by five tests: what the tool does when a fact is missing, what record it keeps of its reasoning, how it tracks HTSUS revisions, when a person reviews its output, and how results reach your systems. Price and demo accuracy come after those.
Vendors in all five categories use the same words (“AI”, “automated”, “audit-ready”), so product pages are hard to compare. This guide maps the categories, says what each does well and badly, lists the questions that decide the fit, and ends with a short test you can run on any tool. It does not rank vendors.
What does “classification software” cover?
Classification software is any tool that helps a team decide, record or maintain the tariff codes for its products. Deciding means getting from a product description to a code in the HTSUS (US Harmonized Tariff Schedule). Recording means keeping the reasons. Maintaining means finding and fixing codes when the schedule, the rulings or the product change. No software takes over the legal duty: under 19 USC 1484 the importer of record must use reasonable care when it declares classification and rate, whatever tool produced the code.
| Category | What it does | Strong at | Weak at |
|---|---|---|---|
| Lookup and search tools | Find tariff lines, rates and rulings by keyword | Free, official text | No decision, no record of why |
| Rules-based engines | Apply coded questions and legal notes by product category | Repeatable answers, asks for missing facts | Only covers the categories modeled; upkeep when notes change |
| Machine-learning suggesters | Predict a code from description text | Speed and coverage on common goods | Learn past errors; answers can drift between versions |
| Expert-reviewed services | Engine plus classifiers who check uncertain items | Hard products, documented reasoning | Turnaround for reviewed items; depends on reviewer quality |
| GTM and ERP modules | Store codes, apply content updates, feed filings | Maintenance across many SKUs and countries | Proposing codes for new products is not what the documentation cited below describes |
What do lookup and search tools do?
Lookup tools search the official text and return lines, rates and rulings; they do not decide anything. The Census Bureau’s Schedule B search does the same job for export statistical codes. CBP’s CROSS database searches rulings.
Lookup tools are the right choice for confirming a code someone already chose and for reading the exact text of a heading or note. Their limit is matching. HTSUS lines are written in legal terms, so a search on a retail product name can miss the right line or return a plausible wrong one, and nothing in the result records why a code was picked.
How do rules-based engines work?
A rules-based engine encodes classification logic as questions and conditions for each product category: knit or woven, fiber content by weight, principal function, whether a part is suitable solely for one machine. The engine walks the questions, applies the relevant Section and Chapter Notes, and returns a code with the path it took.
The strength of a rules engine is repeatability. The same answers always produce the same code and the same explanation, so an auditor can retrace the path a year later. A good engine also stops and asks when an answer it needs is missing. The weakness is coverage and upkeep. Every category has to be modeled, and every change to the nomenclature has to be reflected; HS 2028, which takes effect on January 1, 2028, brings 299 sets of amendments, and an engine built on HS 2022 logic will need work across many chapters.
What are machine-learning suggesters good for?
A machine-learning suggester predicts a code from the words in a product description, using a model trained on past classifications, shipping data or tariff text. Large language models are one kind; text-similarity models trained on declarations are another. Suggesters are fast and handle a wide range of common goods without anyone writing rules.
Three weaknesses follow from how they learn. A model trained on past declarations learns the errors in those declarations along with the right answers. A model’s answer for the same input can change when the vendor retrains or updates it, and some generative tools vary between runs. And a generated explanation can read well while citing a note that does not say what the explanation claims, or a ruling that does not exist. If you use a suggester, open every ruling it cites in CROSS and check every note it quotes. Our article on reliable tariff classification workflows compares how teams combine these approaches.
What is an expert-reviewed service?
An expert-reviewed service runs an engine on every item and sends the uncertain ones to human classifiers before results are returned. The questions to ask are which items go to a person (by confidence score, by product category, by duty at stake), who the reviewers are and how much classification work they have done, how long reviewed items take, and what the reviewer adds to the record.
TariffWolf sits in this category. Its engine is deterministic and does not rely on public LLMs to classify, so the same question returns the same answer and justification. It asks product-specific clarifying questions when technical data is thin, cites the WCO Explanatory Notes, Section and Chapter Notes, the GRIs applied and the CBP or EU rulings behind each 6-digit suggestion, and sends low-confidence and dual-use SKUs to an in-house expert automatically. It stops at the 6-digit HS level; the importer’s classifier or broker chooses the last four digits of the HTS line.
Where does classification fit in GTM and ERP suites?
GTM and ERP suites are where codes live once they are decided: they hold the code against each product, load tariff content from data providers and push codes into filings and invoices. SAP’s own SAP Global Trade Services training material describes numbering schemes for each nomenclature, each tied to a legal regulation, updated by uploading files from a data provider. When a code is retired, the system can reclassify automatically where one old code maps to one new code; where it maps to several, a user chooses. The same material notes that in SAP S/4HANA the material master no longer has a commodity code field. Oracle’s Global Trade Management documentation describes product classification codes with effective and expiration dates, a view of Section and Chapter Notes, and duty data retrieved from a third-party provider for HTS codes.
Both pages describe storing, maintaining and applying codes that a user enters or that arrive from a data provider; neither describes the suite proposing a code for a new product, so ask the vendor about any classification-assistance features before you assume either way. If your company already runs one of these suites, the question to settle is what feeds it: who decides the code, and where the reasoning is stored.
Which questions decide the fit?
Six questions separate tools that look alike in a demo. Ask each vendor to answer them on your products, not on its examples.
What does the tool do when a fact is missing?
The tool should ask for the missing fact or mark the item for review. A knit top with no fiber content, a chemical with no composition, a part with no stated use: each has a heading that turns on the missing fact. A tool that returns a confident code anyway is guessing.
What record does it keep?
The record should show the code, the GRIs applied, the notes and rulings relied on, the alternatives rejected, who reviewed it and when. That record is what you hand over when CBP sends a Form CF-28 (Request for Information) about an entry, and it should be exportable per SKU.
How does it track HTSUS revisions?
The tool should store which revision each code was checked against and list the stored codes a new revision affects. The 2026 HTSUS reached Revision 21 on October 9, 2026, so a tool that only updates its tables without telling you which of your codes changed leaves the re-check to you.
When does a person review the output?
Know the routing rule before you buy: which confidence level, which categories and which duty amounts send an item to a person, and whether that person is yours or the vendor’s.
How do results reach your systems?
Options are a web interface, file upload and download, spreadsheet add-ins, an API with batch endpoints, and connectors to an ERP or GTM suite.
Who is doing customs business?
If the tool or its provider will supply full codes for goods that others are entering, check the broker licensing line. In HQ H350722 (January 16, 2026), CBP said an unlicensed platform’s AI tool may give 6-digit suggestions or general classification information with a meaningfully implemented disclaimer, but deriving subheadings beyond six digits for goods that will be entered is customs business, which requires a broker’s license.
How can you test a tool before buying?
A short test on your own products tells you more than any feature list. This version takes a classifier about two days:
- Pick 30 products: 10 easy single-material items, 10 hard items with answers from CBP rulings, and 10 items with the deciding fact removed.
- Run each tool on all 30, then run it again a week later.
- Score codes at 6, 8 and 10 digits separately, and count the wrong answers given with high confidence.
- Count how many of the 10 incomplete items produced a question or a review flag.
- Read the reasoning for the five hardest items and open every ruling cited.
- Export the full record for one SKU and ask whether you would send it in answer to a Form CF-28.
The detailed method, with scoring sheets, is in how to test HTS classification tools on hard products.
Related reading
- Using AI to classify imports: who is responsible when the code is wrong
- How to classify a product from description to HTS code
- Best Free ECCN Classification Tool in 2026
- Free ECCN Classification Tool in 2026: What You Actually Get
Frequently asked questions
What is HTS classification software?
HTS classification software is any tool that helps decide, record or maintain the HTSUS codes for a company’s products. It ranges from free lookup tools to rules engines, machine-learning suggesters, expert-reviewed services and classification modules in GTM and ERP suites. The importer of record stays responsible for the code under 19 USC 1484.
What is the difference between GTM software and HTS classification software?
GTM software manages trade data and processes across many countries: it stores codes, loads tariff content, and feeds filings and documents. Classification software focuses on reaching the code and recording why. Many companies use both, with classification software or a classifier deciding the code and the GTM or ERP system holding it.
Do I need classification software or a customs broker?
They do different jobs. A licensed customs broker conducts customs business, including filing entries for you, while classification software helps you or your broker reach and document codes. Low-volume importers often rely on their broker; companies with thousands of SKUs usually need a tool and a review process of their own.
Can AI classification software be used without human review?
Not for codes where a wrong answer is costly. Machine-learning tools can return confident wrong answers and cite notes or rulings incorrectly, so route uncertain, high-duty and hard-category items to a classifier, and spot-check the rest.
How should HTS software handle tariff revisions?
HTS software should record which HTSUS revision each code was checked against and list the stored codes that a new revision changes or deletes. Where one old code splits into several new ones, a person has to choose; a tool should flag those cases instead of picking one.
The category you need follows from your volume and your hardest products. A few dozen new items a year can be handled with lookup tools, a broker and the occasional binding ruling. Thousands of SKUs with parts, sets and chemicals need an engine that asks before it guesses, a person on the uncertain items, and a record you can hand to CBP.
If you want to see how an expert-reviewed engine handles your hardest products, the TariffWolf platform page explains how the engine, the clarifying questions, the cited sources and the expert review fit together, and where its 6-digit result passes to your classifier. Bring a handful of difficult SKUs and compare the reasoning with what your current process produces.
Pick a time on the calendar that loads next — no email back-and-forth.