Pax
reclaims the
refund.

There is a pile of money in your shipping records. Almost nobody goes and gets it.

Here is a thing about the United States customs system that sounds made up but is not. If you import goods, pay a tariff on them, and then export those goods again - or turn them into something you export, or destroy them - the government will give most of that tariff back. Up to 99% of it. This has been the law, in one form or another, since roughly the first tariff acts of the 1790s. It is called duty drawback, and it is one of the oldest refund programs in American commerce.

The catch, and there is always a catch, is that claiming it is miserable. You have to match specific imports to specific exports across piles of bills of lading, commercial invoices, and PDFs that were never designed to talk to each other. You have to classify everything correctly under the Harmonized Tariff Schedule. You have to keep records that will survive a customs audit years later. Legacy brokers who do this work take 10-20% of the refund, move at the speed of a fax machine, and - crucially - won't touch your claim if it's under $100,000, because the paperwork costs them more than the fee.

So the money sits there. By Pax's accounting, about 80% of eligible refunds go unclaimed each year, which works out to roughly $15 billion annually. That is not a rounding error. That is a business.

“TurboTax for duty drawback,” which is a more interesting sentence than it sounds.

Pax is an AI-native broker. The word "broker" matters here - it doesn't just sell software and wish you luck, it does the claim. But the word "AI-native" matters just as much, because the whole reason Pax can do the claim profitably is that software does the parts that used to require a room full of analysts.

The founders describe the product as "TurboTax for duty drawback": you hand over your trade documents, and the platform runs the process end-to-end - collection, extraction, calculation, and electronic filing with U.S. Customs and Border Protection. Large language models read the unstructured documents. Deterministic optimization algorithms - not the language models - do the actual refund math. That distinction is the entire company, and CEO Penny Chen is emphatic about it.

The moat isn't the AI. It's the cost to serve.

A lot of AI companies want to tell you the model is the magic. Pax's version is more honest and, frankly, more durable. The insight is that duty drawback eligibility was never the bottleneck - plenty of companies qualify. The bottleneck was that serving them cost too much, so brokers only served the whales. Automate the extraction and the calculation, drive the cost to serve toward zero, and suddenly you can profitably file the small claims everyone else rejected.

Chen puts it plainly: "We automate extensively with AI and software to lower the cost to serve, enabling us to work with companies of all sizes." That's the unlock. Not a smarter refund - a cheaper one to produce, offered to a market the incumbents had written off.

The incumbent worth naming is Charter Brokerage, owned by Berkshire Hathaway, which reportedly processes around $1 billion in refunds a year and holds something like 30% of the market. Pax is not trying to out-Charter Charter on the biggest accounts. It's going after the long tail Charter won't get out of bed for - and using software so the economics work.

Same refund. Very different wait.

From a folder of PDFs to a filing with CBP.

Who this is actually for.

Two builders who saw the refund side of the tariff story.

Two Taiwanese founders looked at America's tariff chaos and, instead of writing another explainer about how much importing costs, asked the quieter question: what about the money on the way back out? Pax is the answer. The team is small - around 20 people - and hires with an unusual twist: paid work trials instead of whiteboard interviews, betting that demonstrated work beats a good interview performance.

A seed round to go collect the unclaimed billions.

There is a nice logic to a fintech built entirely on recovering money that already exists. Pax doesn't lend, doesn't trade, doesn't speculate. It goes and gets duties the government is holding and hands them back. In a market where tariffs keep climbing, the refund side of the ledger only gets bigger - which is a strange, contrarian tailwind: the more painful importing becomes, the more valuable it is to have someone squeeze every dollar back out of the export side.

Penny Chen on the “duty drawback riches.”

A boring program, a big number, and software that finally makes the math cheap.

The most interesting thing about Pax is how unglamorous the underlying opportunity is. Nobody was going to make a movie about duty drawback. But that's exactly why the $15 billion sat there for so long - it was too tedious to chase and too small, per claim, for the incumbents to bother. Pax's wager is that AI plus deterministic math changes the arithmetic of who's worth serving, and that the answer is now "basically everyone."

Whether Pax becomes the "trade compliance control tower" it aspires to - expanding from drawback into audits, foreign-trade zones, reconciliation, and post-entry compliance - is the open question. For now it's doing the specific, verifiable, unsexy thing: matching your imports to your exports and getting your money back. If that works, the boring version of the story is also the good one.