How AI Handles HTS Classification: What Customs Brokers Need to Know
AI and HTS Classification: Why the Real Complexity Isn't the 10-Digit Code
Everyone in customs brokerage talks about HTS classification like the hard part is figuring out the 10-digit product code. It isn't. The hard part - and the part where entries actually go wrong in 2026 - is everything that stacks on top of it.
The Classification Problem Nobody's Talking About
Every import into the United States needs a Harmonized Tariff Schedule (HTS) code. These 10-digit codes determine base duty rates, trade program eligibility, and regulatory requirements. The HTSUS contains over 18,000 subheadings across 99 chapters, and selecting the right one requires deep product knowledge, familiarity with the General Rules of Interpretation (GRI), and awareness of how CBP has historically classified similar goods.
But here's what has changed: the base code is no longer where most of the compliance risk lives. In today's tariff environment, a single product can carry its base duty rate plus Section 232 tariffs, Section 301 tariffs, IEEPA duties, and antidumping/countervailing duties (AD/CVD) - all layered through Chapter 99 provisions that shift with every new executive order and proclamation.
A broker who nails the 10-digit code but misses a Chapter 99 subheading is still filing a wrong entry. And with tariff policy changing as fast as it has in 2025 and 2026, the Chapter 99 layer is where the real exposure sits.
The 10-Digit Code Is the Broker's Job
Let's be clear about something: product classification is skilled work that belongs in the hands of licensed customs brokers. Determining whether a stainless steel bolt falls under 7318.15.20 or 7318.15.80 requires understanding material composition, function, GRI hierarchy, chapter notes, and relevant CBP binding rulings. That judgment is exactly what a licensed broker is trained and authorized to do.
AI is not going to replace that. And any system claiming to classify products from scratch based on a text description alone should be approached with serious skepticism - especially after CBP's recent guidance on where technology crosses the line into customs business.
Where AI adds real value is making sure that classification expertise scales and doesn't stay trapped in one person's head.
The Parts Library: Turning Broker Expertise Into a Compounding Asset
The most valuable thing a brokerage builds over time is institutional classification knowledge. An experienced broker knows that a particular customer's "Model X housing assembly" is classified under 8473.30, that it qualifies for USMCA preferential treatment, and that it's not subject to Section 301. That knowledge took years to develop.
The problem is that it usually lives in someone's memory, a spreadsheet, or a sticky note on a monitor. When that broker is out sick, on vacation, or leaves the company, the knowledge walks out with them.
A parts library changes that equation. Every time a broker classifies a product, reviews it, and confirms or corrects the code, that decision gets stored. The next time the same product appears - from the same importer or a different one - the system already has the answer. No reclassification needed. No second-guessing. The broker's original judgment carries forward automatically.
Over time, this creates an institutional knowledge base that makes your entire team faster. New hires have access to every classification decision the team has ever made. Experienced brokers stop wasting time reclassifying products they've seen a hundred times. And the data gets cleaner with every correction, because the library learns from the brokers who use it.
This is fundamentally different from an AI tool that tries to guess a code from a product description. It's your team's expertise, systematized and compounding.
Where AI Actually Earns Its Keep: Chapter 99 and Tariff Stacking
Once the base HTS code is established - whether through the parts library or the broker's own classification work - the next question is: what additional duties apply? This is where AI delivers outsized value, because the Chapter 99 layer is both complex and constantly changing.
Consider a single steel import in April 2026. The broker needs to determine:
- Does the product fall under the new Section 232 five-tier tariff structure?
- Which tier applies based on the metal's composition and country of origin?
- Is there a Section 301 tariff stacking on top of 232?
- Do IEEPA duties apply based on the country of export?
- Are there active AD/CVD orders on this product from this origin?
- Does the importer qualify for any exclusions or trade program preferences (GSP, USMCA, other FTAs)?
Each of these questions maps to a specific Chapter 99 subheading (9903.xx.xx) that must appear on the entry. Miss one and you're under-collecting duties. Add a wrong one and you're overpaying. And the answer changes every time a new proclamation drops or an exclusion expires.
This is not a job for memory and spreadsheets. The combinatorial complexity of origin, product type, trade program eligibility, and active tariff orders makes it almost impossible to track manually at scale - especially when the rules can change overnight.
AI systems that maintain live connections to HTSUS data and tariff updates can resolve this automatically. When a new executive order adds duties on a specific set of HTS codes, the system reflects those changes in real time. When an exclusion expires, it stops applying it. The broker reviews the output, but they're not manually cross-referencing Federal Register notices against every line item.
How This Fits Into the Entry Workflow
Classification - both the base code from your parts library and the Chapter 99 stacking - is a means to an end. The end is a compliant, accurately filed customs entry.
This is why classification tools that operate in isolation miss the point. A standalone lookup tool that tells you the HTS code still leaves the broker to manually build the 7501, key in the duty calculations, check PGA requirements, and format everything for ABI submission. That is where most of the processing time actually goes.
When classification happens inside the same system that builds your entry, the base HTS code flows directly from the parts library into the entry summary. Chapter 99 provisions are stacked automatically based on origin and product. Duty calculations populate without re-keying. PGA flags are triggered by the classification. The entry is ready for broker review rather than broker construction.
The difference is not marginal. Brokerages running integrated workflows consistently process 3-4x more entries with the same headcount, because their brokers spend time reviewing and approving instead of typing and looking things up.
The Human-in-the-Loop Is Non-Negotiable
None of this works without broker oversight. A parts library match on a routine product with a known classification is straightforward. A new product the system has never seen gets flagged for full manual classification by a licensed broker. An entry with unusual origin or trade program circumstances gets surfaced for review.
This is not a philosophical position - it is a practical compliance requirement. If CBP questions a classification or a duty calculation, the answer needs to be: "Our licensed broker reviewed and confirmed this based on established classification history and current tariff provisions." The system provides the data, the speed, and the accuracy. The broker provides the judgment and the accountability.
Dynamic Tariffs Demand Dynamic Systems
The tariff landscape in 2026 is not the tariff landscape of 2019. Section 232 now has a five-tier structure for steel, aluminum, and copper based on composition and origin. Section 301 investigations are expanding to cover new economies. IEEPA duties are being applied and challenged in court simultaneously. AD/CVD orders are stacking in ways that require line-level origin tracking on steel entries.
Any system that requires a broker to manually update duty rates from a spreadsheet is already behind. The brokerages that are keeping up are the ones where tariff changes flow automatically into their entry workflows - where a new proclamation is reflected in their Chapter 99 logic before the next entry ships, not after a compliance manager reads the Federal Register over the weekend.
Classification That Compounds
The future of HTS classification technology is not about replacing the broker's judgment on 10-digit product codes. It is about two things: making sure that judgment scales through a living parts library, and automating the Chapter 99 tariff stacking layer that has become the most complex and highest-risk part of every customs entry.
The brokerages that get this right build a compounding advantage. Every classification decision strengthens the parts library. Every tariff update is reflected automatically. Every entry gets filed faster and more accurately than the last one. The brokers on the team do more interesting work, handle more volume, and spend less time on repetitive data entry.
That is what AI-assisted classification actually looks like in practice.
