Penny
Chen,
in the refund

Penny Chen runs a company that gives people money the government already owes them. That sentence sounds like a scam. It is, instead, one of the oldest and most ignored corners of American trade law. The program is called duty drawback. It lets a company that paid import tariffs claw the money back when those goods are later re-exported, returned, or destroyed. It has existed, in one form or another, since 1789. Almost nobody uses it.

Chen is the co-founder and CEO of Pax, a San Francisco startup in Y Combinator's Summer 2024 batch. Pax is what she calls an AI-native broker: software that reads a shipping document, understands the rule behind it, matches an import to an export, and files the refund claim with U.S. Customs. The pitch she keeps coming back to is blunt. "Many companies don't even know that the tariffs they pay are refundable."

The numbers behind that sentence are the reason Pax exists. U.S. companies spend more than $100 billion a year on import taxes. Roughly $15 billion in refunds are eligible to be reclaimed each year. About 80% of it - call it $10 to $15 billion - simply evaporates, unclaimed, year after year. Not because the money is hard to get on paper, but because the paperwork is a nightmare and the people who handle it have no incentive to bother with small accounts.

From my point of view, tariff refunds are actually a very interesting mathematical problem. - Penny Chen, CEO of Pax

That line is the whole person in one breath. Most people look at a customs filing and see tedium. Chen looks at it and sees a combinatorial optimization problem: thousands of imports, thousands of exports, a thicket of regulations, and somewhere in the middle the single arrangement that legally returns the most cash. Finding it by hand is impossible. Finding it with the right algorithm is just engineering.

This is the gap Pax drives a truck through. The old way of doing duty drawback is a brokerage business: high-touch, expensive, and allergic to small clients. The math never worked on a modest refund, because a claim could take nine to twelve months to process and a human had to babysit every line. Pax compresses that to a few weeks and, because the heavy lifting is automated, can serve a sub-$50M e-commerce brand as happily as a Fortune 500 importer. Chen's framing: "We automate extensively with AI and software to lower the cost to serve, enabling us to work with companies of all sizes."

She is careful, though, about where the AI gets to play. This is regulated work, and a wrong claim to Customs is not a typo you shrug off. "We operate in a highly regulated space," she has said. "We only use them where there's error tolerance or where a manual review process exists." The large language models read and extract the unstructured documents - bills of lading, invoices - cutting what she describes as 99% of the manual processing time. The refund itself is computed deterministically, by optimization, not by a model guessing. The result, she claims, is about 15% more refund on average than the industry's best software.

Traditional service providers simply cannot do this. They can't offer you the optimal refund because it's impossible for them to iterate through all combinations. - Penny Chen

Before the refund, the algorithms

Chen did not arrive at customs law by accident. She earned a PhD from MIT, finishing in 2022, in the world of optimization and algorithm design. Before that, she came up through National Taiwan University with a background in mechanical engineering. The throughline across all of it is the same instinct: take a messy real-world system and find the structure that makes it solvable.

She spent the better part of a decade as a research scientist applying that instinct to supply chains - eight-plus years building algorithms for pricing, forecasting, fulfillment, and capacity planning. At Amazon she worked on the machinery that keeps a planetary logistics network from buckling. At Flexport she worked on freight. And it was at Flexport, knee-deep in the plumbing of global trade, that she first ran into duty drawback - and could not unsee how much money was quietly going unclaimed.

The co-founder, Chris Le, is the other half of the engine. He is a second-time founder and former engineer at Amazon, Brex, and TikTok, where he built the e-commerce merchant systems from zero to one. He and Chen connected through Y Combinator's co-founder matching platform and bonded over the same thing: a stubborn appetite for unglamorous, practical supply-chain problems. Le, for the record, lists his conversation topics as boxing, meditation, and Vancouver. Chen lists hers as optimization, saunas, and Taiwan.

A trade war is good for the refund business

Pax launched publicly in August 2024 with a tagline that does all the work: "TurboTax for import duty rebate." Make the rebate a few clicks instead of a few months. Then the macro climate did Chen a strange favor. As U.S. tariffs escalated through 2025 - new duties stacking onto tens of billions of dollars of goods - the value of a refund specialist shot up. The company reported tripling its revenue in that stretch and has said it processed roughly $10 million in refunds for clients.

In April 2025 that traction turned into a $4.5 million seed round led by Initialized Capital, with Sancus, Basis Set, Soma Capital, General Catalyst, Transpose, ZVC1, and a roster of Flexport angels alongside. The software is ABI-certified, which means Pax can submit claims directly to Customs and Border Protection rather than routing them through someone else. For a company whose entire thesis is removing the middleman, owning that pipe matters.

What Chen is really selling is access. Duty drawback was never secret. It was just gated - behind expertise, behind fees, behind a process so slow that nobody small could justify it. Her wager is that software dissolves the gate. If the cost to file a claim drops far enough, the program stops being a perk for multinationals and becomes a default for anyone who imports. That is the version of the future Pax is built for: importers of every size reclaiming money that was always, legally, theirs.

It is not a flashy mission. There is no consumer app, no viral moment, no product you would mention at a party. There is a customs rulebook, a pile of shipping documents, and an optimization problem with a billion-dollar answer. Chen seems entirely content with that. She is, by every account, not in it for the spotlight - she is in it because there is a real problem, the math is interesting, and the money is just sitting there.

Quote, unquote