Mercari's Switch to GPT-4o Mini Is Quietly Fueling Millions of Sales

Mercari is doing something quietly brilliant with its AI: making it invisible. The Japanese marketplace giant has moved its core listing features to GPT-4o mini, and the results are the kind of numbers that make product managers swoon — several hundred AI-assisted listings are now generated every single minute. That scale matters. We’re talking about a feature set that launched in September 2024 and took just two months to go from concept to final release, a timeline that would make most enterprise tech teams spit out their coffee.

The initial AI Assistant, launched in October 2023, used a clever hybrid approach. GPT-4 would analyze top-performing listings offline to understand what makes a listing sell, while the faster GPT-3.5 Turbo handled real-time suggestions for users whose items were languishing. The company saw a statistically significant bump in average sales per user. But the real acceleration came when they decided to rip out the friction from listing creation entirely. Max Frenzel, Head of AI/LLM at Mercari, captured the philosophy bluntly: “We believed we could make significantly larger improvements to the listing experience itself. So we decided to try to massively simplify it.”

The result is AI Listing Support. Sellers upload a few photos, and GPT-4o mini handles the rest — categorizing the item, writing a title, and generating a full description. The pivot from GPT-4 to GPT-4o mini wasn’t about chasing a shiny new model; it was a cold, hard calculation about cost-efficiency and latency. The faster, cheaper model actually improved the product because it enabled multi-image processing, which dramatically boosted the quality of the AI’s output. You get better descriptions because the model can see the sneakers from three angles instead of one. Less experienced sellers, the ones who might have abandoned a listing halfway through, now report higher confidence in creating professional-looking pages.

What’s striking is the cultural machinery behind this. Frenzel describes a convergence of top-down pressure and bottom-up experimentation. Leadership from the Group CEO down made it clear they wouldn’t miss the AI wave, while developers on the ground started building Slack bots and internal tools with AI on their own initiative. The company is now building a centralized hub to collect all that grassroots innovation. Looking ahead, Mercari’s AI/LLM team is focused on a three-pronged strategy: flagship features, reusable AI building blocks for other product teams, and deeper exploration across the entire ecosystem. This isn’t a chatbot slapped onto a marketplace. It’s AI as infrastructure, and it’s already working at a scale that would make most retailers envious.