What is Commers?
The problem, the user, and the value proposition of an AI customs broker.
The problem
Every shipment entering the US needs a customs entry filed with CBP. Today that means a licensed customs broker manually:
- Reading the commercial invoice + packing list + bill of lading
- Classifying each line item to a 10-digit HTS code (26,000+ possible codes, ambiguous rules, binding precedent in 250K+ CBP rulings)
- Screening the importer / consignee / supplier against ~30K sanctioned entities across 8 US government lists
- Computing duties — base rate + Section 301 / IEEPA / Section 232 additional duties + MPF + HMF
- Filing the ABI/ACE message with CBP and managing the response
This is slow, error-prone, and expensive. Misclassification is the #1 cause of CBP penalties; a single wrong HTS code can mean years of back-duties. Brokers charge $50-150+ per entry and the expertise is scarce.
Who it's for
Commers serves three sides of the same entry:
- Customs brokers — the licensed customs broker (LCB) operates the workspace and co-signs every entry. The AI clears the routine work and augments the team; it never replaces the broker.
- Importers — mid-market companies filing roughly 500–5,000 entries a year. They are the paying customer and want speed, transparency, and lower cost than a traditional per-entry broker or an in-house customs team.
- Customs authorities — where Commers is headed: the same grounded, auditable classification and screening that makes a broker faster is what a customs authority needs to review entries at scale.
The trust principle
A licensed broker is on record and signs every entry. The AI proposes; the broker approves. High-confidence lines can auto-approve to speed the queue, but Commers never auto-files to CBP — transmission is always a deliberate broker action. (Until direct ACE transmission ships, the broker files the finished entry.)
The value proposition
Customs clearance, tariff intelligence, and denied-party screening — one AI workspace, one supplier graph, one audit trail. Broker-operated, importer-priced, authority-ready.
Concretely, Commers:
- Classifies line items to HTS codes automatically, with calibrated confidence and citable rationale (cites the actual CBP ruling it relied on)
- Screens every party against the full US sanctions-list set in <1 second
- Computes landed cost including the full additional-duty stack
- Routes only the uncertain cases to the broker — high-confidence lines auto-approve, ambiguous ones get co-sign; the broker transmits
- Learns from every broker correction — the classifier gets measurably better as the importer's book of entries grows
Why now
- AI got good enough — frontier models can classify HTS codes accurately when grounded on the right reference data; grounding each classification on live CBP rulings measurably improves accuracy over an ungrounded model
- Tariffs got complicated — Section 301, IEEPA, Section 232 actions now affect ~$370B of imports; the additional-duty math is beyond what a spreadsheet handles
- The data is open — HTSUS, CBP CROSS rulings, and US sanctions lists are all freely available; the moat is in the AI + the workflow + the feedback loop
Positioning
| Legacy brokers | Customs software (e.g. legacy ABI tools) | Commers | |
|---|---|---|---|
| HTS classification | Manual | Manual / rule-based | AI, grounded + self-improving |
| Screening | Manual / 3rd-party | Bolt-on | Built-in, every entry |
| Duty calculation | Spreadsheet | Partial | Full 301/IEEPA/232 stack |
| Feedback loop | None | None | Every correction trains the model |
| UX | Email + PDFs | 1990s green-screen | Modern broker workspace |
| Built for | Brokers | Brokers | Brokers + importers, authority-ready |
The moat
Not the data (it's public) and not a single AI model. The moat is the closed feedback loop: every broker override on every entry becomes a training example specific to that importer's product taxonomy. The more entries flow through, the more accurate the classifier gets — measurably, tracked in an eval database. A new entrant starts from zero; Commers compounds.
Status
Commers is live in production:
- Workspace live at app.commers.ai — an invite-only broker workspace running on real cleared entries.
- Public tariff tracker live at commers.ai/knowledge; docs at commers.ai/docs.
- Data layer real: 26.6K HTS codes, CBP's rulings database, 31K sanctioned entities, 9 tariff programs.
- AI live: classification, invoice extraction, denied-party screening, landed-cost simulation, duty drawback, and a closed feedback loop — all on real entries. Classification is grounded on live CBP rulings, which measurably lifts accuracy; schema validation means zero hallucinated codes; confidence is calibrated and every classification cites the ruling it relied on.
The one step still on the roadmap is direct ACE transmission to CBP. Until that ships, a licensed broker files the finished entry.
Put your own book of entries through Commers.