Whatnot has acquired Shaped, a machine-learning company specializing in real-time recommendation and search, to accelerate personalization across its livestream shopping marketplace. Terms were not disclosed. Shaped's founder and CEO Tullie Murrell will join Whatnot alongside nearly a dozen engineers and AI researchers, and will lead a newly formed Applied AI Research group.
The deal lands as Whatnot sits at an $11 billion valuation off a $225 million Series F earlier this year, having added 20 million buyers over the past 12 months and pushed sellers past one billion cumulative orders since launching in 2019. It is also expanding aggressively into new verticals, with more than 35 categories added last year and 45 more during the first half of 2025, from art to golf to vinyl.
The core problem Shaped is meant to solve is latency. Whatnot has spent six years compressing the time between a shopper's behavior and a personalized recommendation, moving from roughly a day to a matter of minutes. Integrating Shaped's stack is intended to push that closer to real time — a harder engineering target for live auctions than for a static product catalog.
Key facts
- 01Whatnot acquired Shaped, a real-time recommendation and search ML company, to speed up personalization across its live-shopping marketplace.
- 02Shaped founder and CEO Tullie Murrell joins with nearly a dozen engineers and AI researchers to lead a new Applied AI Research group.
- 03Whatnot processes more than 500,000 hours of live video and millions of real-time interactions per week to train its recommenders.
- 04The deal follows a $225M Series F earlier this year that valued Whatnot at more than $11B, with 20 million buyers added over the past year.
- 05Sellers on Whatnot have crossed one billion orders since the platform launched in 2019.
Emmanuel Fuentes, Whatnot's VP of data and AI, framed live commerce as a distinct recommendations problem. Inventory turns over in seconds as auctions close, shows begin and end continuously, and buyer intent shifts mid-stream. A recommender tuned for stable e-commerce catalogs is, in his framing, the wrong tool.
The scale of the training signal is the moat Whatnot is trying to widen. The company says its systems ingest more than 500,000 hours of live video per week alongside millions of real-time interactions — clicks, bids, watches, purchases — and feed that back into its models continuously. Shaped's technology, which pairs customer data with large language models and machine learning for search and discovery, plugs directly into that flywheel.
“That speed matters because live commerce is a uniquely hard recommendation problem. Inventory changes by the second, shows start and end continuously, and buyer intent shifts throughout a show.”— Emmanuel Fuentes, VP, Data and AI at Whatnot
Shaped came out of the applied-ML world with a customer roster that included Outdoorsy and QVC, another live-commerce operator. Murrell previously worked at Meta before founding the company, and the team's expertise sits squarely in the recommendation-and-ranking discipline that platforms like TikTok and Instagram have used to dominate attention.
The competitive frame matters. Resale and marketplace incumbents including eBay and Poshmark are racing to bolt generative AI and improved recommenders onto legacy catalogs. Whatnot's argument is that it already owns the harder surface — live, ephemeral, high-velocity inventory — and only needs to make its ranking layer faster and more personalized to widen the gap. Acquiring a specialist team is a cleaner path than hiring one from scratch.
There are real questions the acquisition does not answer. Integration risk on recommender systems is high; swapping ranking infrastructure on a live marketplace without degrading conversion is a delicate operation, and Whatnot has not disclosed a timeline for when Shaped's technology will be running in production. The Applied AI Research group is also a new organizational construct, and its output will take quarters, not weeks, to show up in gross merchandise value.
For the broader AI market, this is another data point in a clear pattern: consumer platforms with proprietary interaction data are quietly buying specialist ML teams rather than relying solely on foundation-model APIs. The value Whatnot is paying for is not a chatbot layer — it is a ranking engine tuned for a data shape that no general-purpose model provider can replicate. Expect more marketplaces to make similar moves, and expect the small pool of experienced recommendation-systems teams to keep getting more expensive.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
