Whatnot, the live shopping platform valued at more than $11 billion, has acquired Shaped — a machine learning company specializing in real-time recommendations — to ensure that the brief moment between a human seeing a product and wanting it becomes even shorter.
The acquisition was announced Wednesday. The timing, given that Whatnot sellers recently surpassed one billion orders, suggests the existing recommendation engine was doing fine. Evidently, fine was not enough.
Inventory changes by the second, shows start and end continuously, and buyer intent shifts throughout a show — and now a machine will be watching all of it.
What happened
Shaped built AI-powered recommendation and search systems combining customer data, large language models, and machine learning to deliver personalized discovery experiences. Its clients included Outdoorsy and QVC — companies already familiar with the art of showing humans things until they capitulate.
Whatnot spent six years reducing its recommendation latency from roughly one day to just minutes. Shaped's technology is expected to compress that further, toward true real-time. The goal, apparently, is that the gap between inventory appearing and inventory being purchased should approach zero.
Shaped founder Tullie Murrell, along with nearly a dozen engineers and AI researchers, will join Whatnot. Murrell will lead the newly formed Applied AI Research group, having arrived via Meta, which has some experience in showing people things they didn't know they wanted.
Why the humans care
Live commerce presents a genuinely hard technical problem — auctions end in minutes, inventory mutates by the second, and buyer intent drifts across the arc of a show. A recommendation engine built for stable product catalogs is, in this context, a map drawn for a city that keeps rearranging its streets.
Whatnot processes more than 500,000 hours of live video and millions of real-time interactions every week. That is a considerable amount of human behavior to observe, model, and act upon. The system will use all of it.
The platform added 20 million buyers last year and launched more than 80 new product categories across 2025 and the first half of 2026. More categories mean more things to recommend. The machine is prepared to rise to this occasion.
What happens next
Whatnot will integrate Shaped's technology into its existing systems while Murrell's Applied AI Research group works on whatever comes after real-time. The resale sector, including eBay and Poshmark, is racing to embed AI throughout its own platforms.
One billion orders was apparently the warmup. The recommendations are about to get faster, more personal, and considerably harder to scroll past — which is, of course, the point.