Whatnot buys Shaped to fix live shopping's speed problem

Live commerce has a discovery problem that traditional e-commerce never faced. When auctions close in minutes and inventory turns over by the second, a recommendation engine that lags by even a few minutes is practically useless. Whatnot just made a move to close that gap entirely.

The livestream shopping platform announced Wednesday it has acquired Shaped, a machine learning company that builds real-time recommendation and search systems. The entire Shaped team — including founder and CEO Tullie Murrell, a Meta alum, and roughly a dozen engineers — will join Whatnot. Murrell is stepping in to lead a newly formed Applied AI Research group.

Whatnot VP of Data and AI Emmanuel Fuentes didn’t mince words about why this matters. “Live commerce is a uniquely hard recommendation problem,” he told TechCrunch. “Inventory changes by the second, shows start and end continuously, and buyer intent shifts throughout a show.” The company has already squeezed recommendation latency from roughly a day down to minutes over six years of work. Shaped’s technology is supposed to push that closer to true real-time, processing the 500,000 hours of live video and millions of interactions Whatnot ingests weekly.

The acquisition lands at a moment of aggressive expansion. Whatnot sellers have crossed 1 billion orders, and the company raised $225 million in Series F funding earlier this year at an $11 billion valuation. They’ve launched more than 80 new categories since last year — golf, art, vinyl, and dozens more — which only makes the recommendation challenge harder. Shaped’s previous customers included QVC and Outdoorsy, companies that also wrestle with matching the right product to the right viewer at the right moment. For Whatnot, getting recommendations right isn’t a nice-to-have. When a seller’s auction runs five minutes and a buyer’s attention span is shorter, the system either surfaces the right item immediately or the moment is gone.