Inside Mercari’s New Merchat AI Shopping Bot
Retail’s embrace of artificial intelligence has pushed bots to new corners of the shopping experience — and not just for new products or traditional e-tailers, but also secondary sales platforms like Mercari.
Last week, the peer-to-peer selling platform launched a beta test for Merchat AI, a conversational shopping assistant based on OpenAI’s ChatGPT.
More from WWD
Up Close with The RealReal's 'Joy of Making and Creating' Recollection 04
2020 Holiday Gift Ideas: Google's Top Searched Beauty, Home and Tech Gifts
“I’ve been following AI for a long time, and I feel like often we’ve had a hard time really getting new users to understand how deep our inventory is,” John Lagerling, chief executive officer of Mercari’s U.S. division, told WWD. “We have so many unique items. And we felt like just the clean search experience doesn’t always guide people to find all those treasures that are hidden in all these categories that we have.”
The value of applying data science in the back end for any type of retail is evident — the mere fact that AI can track myriad inventory levels and supply chains across regions, and in real time, is enough for brands and retailers to amp up their tech investments. But it’s especially valuable in resale, which doesn’t deal in large volume orders or lots, but individualized offerings that vary at any given moment, depending on what people have to sell.
“When these LLMs [large language models] started getting to the level that I noticed in the last couple of months,” he added, “I felt like this could really be something that could help guide people through the treasure hunting.” LLMs involve enormous data sets that train and refine AI and machine learning models, allowing them to evolve into more sophisticated systems much faster.
Mercari has been organizing its data for the last two years to prepare for the advent of genuinely useful, usable retail AI. In essence, it got everything ready and then waited for the tech to become good enough. It’s not an unusual data strategy, perhaps particularly for Lagerling, who hails from Google.
The overarching vision that he adopted from his previous employer was to use technology to make things easier. That inspired Mercari to lay the groundwork and then, when OpenAI launched its development tools for its ChatGPT AI bot in March, it was able to connect Mercari’s data to the AI engine was quickly, creating the bot in as little as two weeks.
At first blush, its feature set looks robust. Merchat AI can offer gift suggestions or recommend items that fit the latest trends, search hard-to-find collectibles and even offer styling tips in apparel or home goods. It’s not limited to a select audience either, so anyone can take it for a test spin online.
But Lagerling made clear that this is not an official roll-out. The company launched Merchat AI as a test feature, and by calling it that, Mercari is acknowledging that it doesn’t have the polish of a final version. It’s live on the web, but not available through the mobile app, and it doesn’t offer deeper features, like saving or favoriting historical chats. ChatGPT itself isn’t perfect either, so fundamentally, flawed recommendations or other unexpected behavior could still pop up.
That wasn’t enough to hold up the project, though. “I think we sometimes have a tendency to look for perfection before we make things available,” the CEO continued. “But we felt like it was kind of magical enough, [and] we shouldn’t hold that back.”
A judicious release that aims to set reasonable expectations seems like a rarity in the often hyped-up, tech-fueled world. But platforms may be realizing that this type of strategy is in order, especially with bots. Because simply deploying AI to the public comes with no guarantee of success — as Snapchat recently learned.
Last week, the social media app added its own ChatGPT-powered bot, called My AI, to messaging inboxes without permission or consent, then immediately saw a surge of one-star reviews hammer its app store page.
Mercari’s introduction takes a more measured approach, at least for now. But the popular Tokyo-based shopping platform has more than 50 million downloads in the U.S. alone and 350,000 new listings daily, so a full launch of Merchat AI could eventually flood the resale zone.
Best of WWD
