Pax Democratizes Duty Drawback with AI

Diving deeper into

Penny Chen, CEO of Pax, on building AI-powered tariff refunds

Interview
Right now, drawback is a privileged tool primarily for large enterprises.
Analyzed 3 sources

The real bottleneck in duty drawback is not the law, it is the service model. Claims require matching messy import and export records, line by line, across invoices, bills of lading, HTS codes, and ERP exports. Legacy providers do this with analyst labor and old software, which makes small claims uneconomic. That is why drawback has mostly been captured by large importers with big refund pools and compliance teams.

  • Traditional providers often take nine to twelve months to process a claim, and many will not prioritize claims below roughly $100K in refunds. That screening function effectively excludes smaller importers, even when they are legally eligible for refunds.
  • The workflow is concrete and ugly. A broker has to pull PDFs, CSVs, invoices, and customs records, normalize them into structured fields, then decide which imports can be matched to which exports under the rules. Pax uses LLMs for extraction and deterministic algorithms for the refund calculation.
  • This is where Pax sits relative to the market. Charter Brokerage still controls about 30% share and processes about $1B of refunds annually, while Flexport is building drawback into a broader digital customs and forwarding stack. Pax is attacking the narrow step where manual work has kept the category enterprise only.

If software keeps cutting claim prep time and raising refund accuracy, drawback should move from a specialist service into a standard feature of customs brokerage and trade software. That would widen the market from Fortune 500 importers to the long tail of SMB brands, manufacturers, and brokers that today leave refund dollars unclaimed.