Classify every product against HTS in minutes

Run every product through AI for HTS classification, validate the codes against the USITC API, and output a classified product file ready for customs submission.

The prompt

I want to classify a product list for customs filing. Can you build me a flow that pulls product descriptions from a CSV, sends each item to OpenAI to generate candidate HTS codes, validates them against the USITC API, and outputs a classified product file ready for submission?

Just copy and paste the prompt into a new Parabola flow to get started.

What Parabola builds

A workflow with six steps you can edit:

1. Read the product list. From a CSV, a NetSuite export, a PLM extract, or a shared file drop. The flow takes whatever the trade team can hand off.

2. Generate candidate HTS codes through AI. Each product description goes through the AI step. The output is one to three candidate codes per item with reasoning the customs team can review.

3. Validate every code against USITC. The flow hits the public USITC API to confirm the code exists, pull the duty rate, and flag any code that's been retired or replaced.

4. Apply your override list. Some products have a previously approved HTS that overrides any new suggestion. The flow respects the override file.

5. Flag low-confidence rows. Anything below your confidence threshold lands in an exception list so a customs broker can review before submission.

6. Output the classification file. Clean version for customs filing, exception list for review, and a change log if any prior classifications shifted.

Why teams stop doing this manually

HTS classification is a judgment call dressed up as a lookup. Every new SKU needs a code. The customs team can spend twenty minutes on a complicated item, looking up the chapter, narrowing the heading, picking the right subheading, then double-checking the rate. Multiply that by a catalog of a few thousand items and the trade team is in the work for weeks.

The manual version is one classifier on Excel with the USITC website open in another tab. Read the product. Pick the chapter. Drill into the headings. Cross-check the GRI rules. Write the code, the rate, the rationale. Move to the next product. It works for a slow week. It does not scale to a brand launching a hundred new SKUs in a season, or a forwarder onboarding a new client with five thousand items in their catalog.

The cost of getting this wrong is regulatory. An overclassified product carries an inflated duty bill. An underclassified one creates customs exposure that can sit on the books for years. A misclassified SKU can hold up an entry at the port. Each of these traces back to one classification call that didn't get the time it needed.

How it works

Step 1. Paste the prompt.

Open Parabola, paste the prompt in section 2, and let it ask follow-up questions about your product catalog, your existing classifications, and your confidence threshold for auto-approval.

Step 2. Connect your data.

Source file for the product list, USITC API for validation, optional override file for previously approved codes. ERP connection if you want the codes written back to the product master.

Step 3. Run it on every new batch.

The flow runs when a new product list lands, generates candidates, validates against USITC, and outputs the classified file. The customs team reviews the exception list and signs off on the rest.

FAQ

Does the AI step replace a licensed customs broker?

No. It accelerates the first pass. The flow generates candidates and the customs broker still owns the final classification call. Exceptions and low-confidence rows route to a human every time.

How does the USITC validation work?

The flow hits the USITC public API for every candidate code. It confirms the code is active, pulls the current duty rate, and flags codes that have been retired or replaced. Validation runs on every batch so retired codes never make it to filing.

What if my products require country-specific classifications beyond US HTS?

Add the country-specific schedule as another validation step. The flow can run the same product list against EU CN codes, Canadian HS codes, or any other harmonized schedule you have access to.

How does the flow handle products with multiple components?

The AI step can split the product into components for AI review, apply GRI rules to determine the essential character classification, and surface the reasoning. Complex items still get human review, but the flow does the prep work.

Can the flow learn from prior classifications?

Yes. Feed it your historical classifications and the AI uses them as a reference. Approved codes get applied to similar new products. The customs team only reviews the genuinely new ones.
Classify a catalog in hours, not weeks.
Paste the prompt, point it at your product list, and let the AI plus USITC pipeline do the prep work.
Start for free