HTS Classification API vs Manual Customs Classification (2026)

An API classifies products into HTS codes in under a minute (typically 15-35 seconds) with CBP ruling evidence, while manual classification takes hours and scales poorly. Here is when each approach works best.

The Scale Problem

The US Harmonized Tariff Schedule contains roughly 20,000 ten-digit codes across 99 chapters. The schedule is revised multiple times per year by the USITC, with Section 301 and Section 232 tariff actions adding further complexity. Since 2018, over 30 rounds of tariff modifications have altered rates on thousands of codes.

Manual classification requires a specialist who:

• Understands the product's material composition, function, and end-use
• Knows General Rules of Interpretation (GRI 1-6) and how they interact
• Searches CBP rulings for precedent on similar products
• Tracks tariff revisions and additional duty programs
• Stays current on Section 301/232 exclusions and their expiration dates

Training a customs classification specialist takes 2-3 years. The US has a structural shortage of them. Meanwhile, e-commerce has pushed SKU counts from thousands to millions per importer. A mid-size company importing 5,000 distinct products cannot afford to manually classify and re-classify every item every time tariff schedules change.

Side-by-Side Comparison

Where Manual Still Wins

API classification is not a universal replacement for human expertise. Manual classification is still superior for:

• Truly novel products. A product that has no analogous CBP ruling and doesn't fit neatly into existing tariff headings requires GRI analysis and judgment calls that benefit from experience. Think: a new category of wearable medical device that combines electronics, textiles, and sensors.
• Binding ruling requests. If you need a legally binding classification from CBP, a human must prepare and file the ruling request through the eRuling system. The API cannot do this.
• Protests and disputes. When CBP challenges a classification, the protest process involves legal arguments, prior disclosure, and negotiation. This requires a licensed customs broker or trade attorney.
• Classification of sets and kits. GRI 3 (sets put up for retail sale) involves subjective "essential character" determinations where experienced classifiers have an edge.

Where API Wins Decisively

For the majority of classification work (the repetitive, high-volume, well-precedented cases), an API is the better tool:

• Volume. Classify 10,000 SKUs in about a day of unattended calls instead of over several weeks. Re-classify the entire catalog when tariff schedules change, instead of hoping nothing was affected.
• Consistency. The same method and the same evidence sources on every query, with the rationale (and the cited ruling, when one applies) returned alongside, so any variation is visible and checkable. This matters for CBP audits where inconsistent classification across shipments raises flags.
• Cost at scale. At $0.05-0.10 per classification versus $100-200, the economics are not close. A catalog of 5,000 products costs $250-500 via API versus $500,000-1,000,000 through manual classification.
• Speed. A product manager can get an HTS code and a duty estimate in under a minute, not days. That makes duty estimates practical for sourcing decisions and pricing models.
• Audit trail. Every classification response includes the CBP ruling cited (when precedent exists), legal notes applied, confidence level, and Census figures for duty actually collected on past imports. This structured evidence is immediately usable for reasonable care documentation.

How htsapi.dev Combines Both Approaches

The htsapi.dev classification endpoint is not a simple lookup table or keyword matcher. It uses an agent-based architecture that mirrors how an experienced customs specialist works:

1. Ruling precedent first. The agent searches 220K+ CBP ruling letters for similar products. When a ruling clearly applies, it follows authoritative precedent.
2. GRI reasoning. When precedent is thin, the agent reads Section/Chapter/Heading notes and applies the General Rules of Interpretation to pick the right heading.
3. Schedule verification. Older rulings often cite codes that have since been restructured. The agent verifies every committed code against the current 2026 schedule and finds the current equivalent.
4. Clarification. When candidates diverge (e.g., the product could be mechanical or electric), the API returns a targeted question instead of guessing.
5. Evidence packaging. The response includes cited rulings (with links to rulings.cbp.gov), relevant legal notes, and Census Bureau figures for duty actually collected on past imports.

The result: AI does the heavy lifting (searching, matching, reasoning) while showing its evidence. The human reviews the evidence, confirms the classification, and escalates edge cases to a specialist. Best of both approaches.

Real Examples

Cost Comparison

The cost difference between manual and API classification is significant at any volume, but the breakeven is almost immediate.

The breakeven point: roughly 1 classification. Even a single API call at $0.08 versus a broker's $150 per-item research fee saves money. The real question is not cost: it is whether your product is common enough for API classification to be accurate, or novel enough to warrant manual research.

Getting Started

Try it free on the web

Visit htsapi.dev and type a product description into the demo. No account required. You will see the HTS code, duty figures, CBP ruling citations, and confidence level, typically within 15-35 seconds.

API access

Buy a credit pack to get an API key. Each successful /classify call costs 1 credit. All other endpoints (HTS search, ruling search, the AD/CVD screen) are free with a key.

Python example

import requests

resp = requests.post(
    "https://htsapi.dev/v1/classify",
    headers={"X-API-Key": "YOUR_API_KEY"},
    json={
        "description": "stainless steel insulated water bottle 500ml",
        "country_of_origin": "China"
    }
)
data = resp.json()

top = data["candidates"][0]
print(f"HTS: {top['hts_code']}")
print(f"Duty: {top['general_duty']}")
print(f"Confidence: {top['confidence']}")

# Census: duty collected on past imports (historical average)
if top.get("census_duties"):
    census = top["census_duties"]
    print(f"Effective rate: {census['total_effective_rate_pct']}%")

# Check if the API needs more info
if data.get("clarification"):
    print(f"Question: {data['clarification']}")

Frequently Asked Questions