Trava Thinks Importers Are Leaving Billions on the Table

The Macro: Trade Compliance Is a Mess Nobody Talks About

I want you to think about how many physical products cross international borders every single day. Millions of shipments, each one requiring customs documentation, each one assigned a tariff classification code that determines how much duty gets paid. Now think about how that classification process actually works at most companies. It’s a mix of spreadsheets, legacy customs broker software, tribal knowledge from a compliance person who’s been doing this for 25 years, and a lot of guessing.

The result is predictable. Importers overpay on duties constantly. The estimates vary, but industry groups put the number in the billions annually for U.S. importers alone. And it cuts both ways. Overpay and you’re giving away margin. Underpay and you’re looking at penalties, audits, and potential seizures. The Harmonized Tariff Schedule has over 10,000 six-digit subheadings, and the U.S. adds even more granularity on top of that. Getting the right code matters, and it’s genuinely hard.

The existing tools in this space are not great. Descartes, Amber Road (now E2open), and Integration Point all offer trade compliance platforms, but they’re enterprise software in the most unflattering sense of the word. Long implementation cycles, rigid workflows, expensive seats. Smaller importers often just rely entirely on their customs broker, which works until the broker makes a mistake and the importer eats the cost.

What’s shifted recently is that large language models turned out to be surprisingly good at classification tasks. Not perfect, but good enough to flag errors, suggest alternatives, and audit existing entries at a speed no human team can match. The question is whether anyone can build a product around that capability that actually fits into how importers work day to day.

The Micro: Agents That Read Your Customs Entries

Trava, backed by Y Combinator (W25), is building AI agents specifically for trade compliance. The founder, Pushkar Lanka, came from Plaid where he ran the Privacy and Risk engineering org, and before that was an early engineer on Meta’s blockchain project. That background in financial infrastructure and risk systems maps pretty directly onto what trade compliance actually requires: parsing structured data, applying complex rule sets, and flagging exceptions.

The product focuses on two core jobs. First, auditing existing customs entries to find classification errors and duty overpayments. If you’re an importer doing thousands of entries per year, even a small error rate compounds into real money. Having an AI agent review every single entry against the tariff schedule and your historical data is the kind of task that was always theoretically possible but practically too expensive to do manually.

Second, automating the classification process itself. Instead of a compliance analyst spending 30 minutes researching the right HTS code for a new product, Trava’s agents can propose a classification with supporting reasoning. The analyst still reviews and approves, but the heavy lifting of searching through tariff schedules, reviewing rulings, and checking precedent is handled by the agent.

The tagline on the site is “Protect Your Margin on Every Import,” which tells you exactly who they’re selling to. This isn’t a pitch about regulatory elegance. It’s a pitch about money. Every misclassified entry is either margin you’re giving away or risk you’re accumulating.

What I find interesting about the approach is that it’s narrow enough to actually work. Trade compliance is one of those domains where the rules are well-documented (even if they’re complicated), the data is structured (customs entry forms follow standard formats), and the cost of errors is quantifiable. That’s a much better setup for AI agents than vague “automate your workflow” promises.

The competitive landscape is worth watching. Customs brokers like Livingston and C.H. Robinson have their own technology platforms. Flexport has been investing in automation. And the enterprise trade compliance vendors aren’t going to sit still. But most of those players are optimizing around their existing business models, not rethinking the classification workflow from scratch.

The Verdict

I think Trava picked a smart problem. Trade compliance is technical enough to create a moat, expensive enough to justify real pricing, and underserved enough that “better than the status quo” is a low bar to clear.

At 30 days, I’d want to know: how accurate are the classification suggestions? In trade compliance, a 95% accuracy rate sounds good until you realize that the 5% you get wrong could mean a CBP audit. The trust calibration here matters more than in most software categories.

At 60 days: are importers actually letting the agents touch their entries, or are they using Trava as an audit-after-the-fact tool? The difference between those two use cases is enormous in terms of product value and pricing power.

At 90 days: can this scale beyond the mid-market? Enterprise importers with tens of thousands of entries per month are where the real revenue is, but they’re also the ones with the most entrenched customs broker relationships and the least appetite for switching costs.

The trade compliance space is big, boring, and broken. That’s usually where the best enterprise software companies get built. Trava has the right background and the right angle. Now it’s about execution.