Customs broker vs classification software: cost, accuracy and when to switch

Most importers do not choose between a customs broker and classification software. They drift into paying for both without deciding which one owns the work, and the bill shows up as per-line charges nobody reviews. This is the decision written out: what each costs on a real catalog, how the accuracy actually compares, who carries the legal exposure either way, and the volume at which the arithmetic flips.

Last updated September 2026.

Should I use a customs broker or classification software?

Use software for the first pass on your catalog and a broker for filing entries and for the handful of lines that are genuinely contested. They solve different problems. A broker is licensed under 19 CFR Part 111 to transact customs business on your behalf and is paid mainly to get entries filed correctly and on time. Classification software is paid to produce a defensible code for every SKU at a cost that does not scale with catalog size. Framing it as a replacement question is what leads people to buy the wrong thing.

The reason this now needs a decision, rather than drifting, is volume. CBP suspended the $800 de minimis exemption for every mode other than the international postal network effective June 24, 2026, and for the postal network effective July 24, 2026. Shipments that used to clear on a manifest with no classification at all now need formal or informal entry with a full 10-digit HTSUS number. A brand that classified twelve products last year is classifying four hundred lines this year, and per-line pricing that was invisible at twelve is a budget conversation at four hundred.

What does a customs broker charge for classification?

Around $5 to $15 for each HTS line beyond the three to five folded into the base entry fee, and $50 to $200 per commodity for a formal classification review on something new. Full-service entry clearance itself runs roughly $100 to $250, with entries carrying many lines or partner-government-agency checks landing higher. Brokers set their own rates, so these are the bands US brokers publish rather than a tariff.

The row that matters is the third from the bottom. Broker classification cost tracks how many distinct products you import; software cost does not. That is the entire economic argument, and it is why the answer differs so sharply between a company importing eight stable SKUs and a company importing eight hundred seasonal ones. The full breakdown of HTS classification cost runs the same numbers across consultants and binding rulings as well.

How accurate is AI classification compared to a customs broker?

Close enough that accuracy is no longer the deciding factor, and the failure modes differ. An experienced broker who knows your commodity is very good on that commodity and merely adequate outside it. Software is uniformly decent across every chapter and does not get tired on line 340 of a spreadsheet. Vendors publish first-pass accuracy figures in the low to mid nineties; treat every one of them as marketing until you have tested it on your own products, because accuracy measured on easy consumer goods tells you nothing about your mixed-material kits.

What genuinely separates the two is consistency. A broker's classification decisions live in the head of whichever staff member handled that entry. Turnover, holiday cover or a change of brokerage produces a different code for the same product, and inconsistent codes for identical goods across entries is one of the oldest audit triggers there is. Software applies the same logic every time, which is worth more than a percentage point of accuracy when CBP asks why the same widget was classified two ways in March and September.

The practical test is not the code. It is whether you can see the reasoning. A tool that returns a 10-digit number and a confidence score cannot be checked by you, so it cannot be corrected by you. A tool that shows which General Rule of Interpretation it applied and which chapter or section note it relied on lets a human catch the one case in twenty where the input was ambiguous. Ask for that in a demo and watch how the vendor reacts.

Who is liable if the HTS code is wrong?

You are, in every scenario. 19 USC 1484 places the duty of reasonable care on the importer of record, acting in person or by an agent. Hiring a broker does not transfer it, and neither does buying software. Penalties under 19 USC 1592 can attach even when the wrong code would have produced identical duty, because the violation is the false statement, not the underpayment.

This changes how you should read vendor claims on both sides. A broker's licence is real value: they carry professional obligations, they are accountable for the filings they make, and a good one will tell you when a classification is a stretch. But no broker indemnifies you against a 1592 penalty, and no software vendor does either. Only a ruling issued under 19 CFR Part 177 binds CBP, and under 19 CFR 177.1(b) even an oral opinion from CBP's own staff does not. Where a call is close and the annual duty at stake is meaningful, the right move is neither vendor: request a binding ruling, which costs nothing and takes about 30 days.

At what volume does classification software beat a broker?

Roughly at the point where you import more than 40 or 50 distinct products a year, or where new SKUs arrive faster than once a month. Below that, broker line charges are noise and the relationship is worth more than the savings. Above it, the per-line meter starts running against you and, more importantly, classification stops being an occasional task and becomes a step inside your product launch process.

Run the arithmetic on your own numbers rather than trusting a threshold. Take a brand with 300 SKUs turning over roughly 40 percent a year, so 120 new classifications. At $50 for a light review on each, that is $6,000 a year in review fees before a single entry is filed, and it does not include the per-line charges on the entries themselves. The same brand's tenth SKU and its three hundredth cost the same in software, which is zero at the margin.

Then run the number that dwarfs both. A cotton knit t-shirt enters at 16.5 percent while the same shirt in man-made fiber enters at 32 percent, and the chief weight rule that decides which applies is a plurality test rather than a majority test. On $200,000 of landed value that single distinction is about $31,000 a year. Classification spend is rarely the expensive mistake. The rate you are quietly paying on a line nobody rechecked usually is.

What a customs broker does that software cannot

Quite a lot, and any vendor telling you otherwise is selling. A broker transmits entries through ACE, posts and manages your customs bond, handles partner-government-agency filings for FDA, USDA, EPA and the rest, responds to CBP requests for information, works protests and post-summary corrections, and knows the operational quirks of the ports you actually use. None of that is classification, and none of it is something a classification tool touches. If you were hoping software would let you drop your broker entirely, read whether you need a customs broker at all before you make that call.

A broker is also a human you can put a hard question to. On genuinely contested goods, sets, novel products, anything where two headings both look defensible, a licensed opinion with reasoning behind it is worth paying for. That is a different service from stamping a code onto a routine reorder, and it should be priced differently.

What software does that a broker cannot

Three things, all of them about scale and record keeping rather than expertise. It classifies your whole catalog in an afternoon instead of over a quarter. It keeps the reasoning attached to the SKU, so the answer survives staff turnover and can be produced when someone asks why. And it prices the duty before you commit to a purchase order, which turns classification from a compliance chore into a sourcing input.

That last point is where the money is, and it is underused. If you know the landed rate on two candidate suppliers before you order, classification pays for itself in margin decisions rather than in avoided penalties. The same product master that drives your duty exposure increasingly drives obligations your broker never sees either, from supplier due diligence under the forced-labor rules to emissions reporting on imported goods, which makes clean product data worth more than the customs use case alone.

The hybrid model most importers end up with

Software does the first pass on the whole catalog and holds the reasoning. Your broker reviews the exceptions, files the entries and handles anything contested. You send them 300 pre-classified lines with the basis attached instead of a spreadsheet of product names, which is a materially different conversation and usually a cheaper one, because you are buying review time rather than research time.

Brokers are generally fine with this, and the good ones prefer it. Reviewing a proposed classification with stated reasoning takes a fraction of the time of building one from a product title, and it moves the relationship toward the work they would rather be doing. If you sit on the broker side of that arrangement, the same workflow from your seat is covered on the customs broker software page.

How to switch without breaking your entries

Do not reclassify everything at once. Pull 25 SKUs that caused an argument in the past year, classify them fresh without showing the tool the code you currently use, and compare. Agreement tells you the tool is safe on your goods. Disagreement is the useful result: work out which side leaned on the better rule, and price the gap over a year of volume before deciding who was right. Keep the incumbent code on live entries until you have that answer.

Then feed the new codes in as products come up for reorder rather than in a single sweep, and keep both the code and the written basis with the SKU record. Entry records have to be retrievable for five years and the HTS number itself sits on CBP's records list, so the reasoning is not optional paperwork. It is the thing that makes the classification defensible later.

If you are running this comparison because an enterprise renewal is coming up rather than because of a broker invoice, the tradeoffs are slightly different, and the Descartes CustomsInfo alternative comparison covers what a full trade content suite gives you that a focused classifier does not. For the tooling side of the decision on its own, the tariff classification software page lays out what to test in a demo.

Where to start

Take the SKU that caused your last classification argument and classify it in the box at the top of this page. If the reasoning matches what your broker told you, you have just confirmed the code for free. If it does not, you have found a line worth a phone call, and possibly a ruling request. Either outcome is worth five minutes, and neither requires you to change anything about how your entries get filed.