Whatnot buys Shaped to sharpen real-time AI recommendations for live shopping

Whatnot has acquired Shaped, a machine learning company focused on real-time search and recommendation systems, in a move designed to make product discovery faster and more personalized on the live-shopping platform. The deal matters because live commerce changes by the second, and Whatnot says Shaped’s technology will help it recommend the right items while auctions, inventory and buyer interest are all in motion.

The acquisition underscores how seriously Whatnot is treating AI as core infrastructure, not just a feature. As the marketplace expands into more categories and attracts millions of buyers, the company is trying to solve one of live shopping’s most difficult technical problems: surfacing relevant products instantly in a setting where listings can appear, disappear and sell out during a stream.

Why live shopping is a harder recommendation problem than traditional e-commerce

Live shopping does not behave like a normal online storefront, and that difference is exactly why Whatnot is investing in specialized recommendation technology. On a conventional retail site, inventory is usually stable enough for search systems to index products in advance and update them periodically. On Whatnot, inventory is dynamic, auctions are time-bound, and buyer intent can change while a seller is still on camera.

That creates a moving target for personalization. A shopper may join a livestream looking for trading cards, then pivot to collectibles, apparel or another category within minutes. The platform has to interpret those signals quickly, while also accounting for what is actually available right now.

Whatnot says it has spent years refining its recommendation stack to work in that environment. Emmanuel Fuentes, the company’s vice president of data and AI, said the system has already cut recommendation latency from about a day to only a few minutes. The Shaped acquisition is meant to bring that window even closer to real time.

Fuentes said the combination of Whatnot’s systems and Shaped’s technology should make recommendations faster, more responsive and more personalized, adding that live commerce is unusually difficult because inventory shifts second by second and shopper intent changes during every show.

What Shaped brings to Whatnot

Shaped built AI tools for companies that want to improve search, discovery and recommendation quality using a mix of customer data, machine learning and large language models. In practical terms, that means it helped businesses predict what users are most likely to want, then present those items in a way that feels relevant rather than generic.

The company’s customer list included brands such as Outdoorsy and QVC, which suggests its tools were already being used in commerce environments where relevance and timing matter. For Whatnot, that background is especially valuable because live retail depends on matching people to products instantly, not simply ranking static catalog results.

As part of the acquisition, Shaped founder and chief executive Tullie Murrell will join Whatnot, along with nearly a dozen engineers and AI researchers. Murrell is set to lead a newly created Applied AI Research group, giving Whatnot a dedicated team to push recommendation and discovery work deeper into product development.

Murrell’s résumé also hints at why Whatnot wanted him. Before founding Shaped, he worked at Meta, where large-scale recommendation systems are a major part of the company’s infrastructure. That kind of experience is useful in a platform where speed, ranking quality and personalization can directly affect sales.

How Whatnot is using AI across a fast-growing marketplace

Whatnot says its systems already process more than 500,000 hours of live video every week, along with millions of real-time interactions. Those signals feed its recommendation engine, which the company continues to tune as it expands into new categories and serves a larger audience.

The logic behind the deal is straightforward: the more the marketplace grows, the harder it becomes to surface the right product at the right moment. AI can help by reading behavior in context, identifying patterns across streams and adjusting suggestions as the market changes in real time.

That matters particularly in livestream environments, where a seller can move from one lot to another without warning and where a hot item may disappear before a shopper even finishes scrolling. Better recommendations can improve the odds that a buyer sees something relevant before the moment passes.

What the company says its systems already handle

According to Whatnot, the platform’s recommendation infrastructure is not just processing product clicks. It is also absorbing a massive volume of live-video and interaction data that helps inform what users might want next.

• More than 500,000 hours of live video processed each week
• Millions of real-time interactions analyzed weekly
• Recommendation latency reduced from roughly one day to a few minutes
• Aiming to move recommendations even closer to live delivery

What the acquisition means for Whatnot’s business strategy

The purchase comes during a period of rapid expansion for Whatnot, which launched in 2019 and has since built itself into one of the best-known names in live commerce. The company recently said sellers have crossed one billion orders, a milestone that points to the scale of its marketplace and the volume of transactions flowing through the platform.

Whatnot also raised $225 million in Series F funding earlier this year, a round that valued the company at more than $11 billion. The company said it added 20 million buyers over the past year, giving it both more scale and more data to feed into recommendation systems.

That growth has been accompanied by an aggressive category expansion. Whatnot launched more than 35 new categories last year, including art, golf and vinyl, then added more than 45 more in the first half of 2025. New subcategories are still rolling out each month, which increases the complexity of the discovery problem but also broadens the company’s revenue potential.

Why category growth increases the value of personalization

Each new category creates a new set of buyer behaviors, seller dynamics and product patterns. A recommendation model that works for collectibles may not behave the same way for apparel, sports memorabilia or home goods.

That is one reason live-commerce platforms are increasingly leaning on more advanced AI systems. As the range of products expands, generic ranking becomes less effective and personalized discovery becomes more important. A system that can infer intent quickly across different types of inventory can help keep users engaged and help sellers find buyers faster.

Who is Tullie Murrell and why does he matter here?

Tullie Murrell is the Shaped founder stepping into a leadership role at Whatnot, and his background is likely one reason the company viewed the acquisition as more than a simple technology purchase. He is moving over with a team that includes engineers and AI researchers, which suggests Whatnot wanted both the product know-how and the people who built it.

Murrell will lead a new Applied AI Research group inside Whatnot. That is a meaningful signal: rather than outsourcing recommendation work to an external vendor, the company is bringing the capability in-house and creating a unit specifically tasked with improving it.

This kind of integration is increasingly common among consumer platforms that depend on algorithmic discovery. The best recommendation systems are often shaped by constant iteration, direct access to proprietary data and fast feedback loops between product teams and researchers. Owning the team and the technology can accelerate all three.

How does this fit into the broader AI race in resale and commerce?

Whatnot’s move reflects a broader trend across resale and commerce platforms, where AI is becoming a competitive necessity rather than a nice-to-have tool. Rivals such as eBay and Poshmark have also been pushing deeper into AI-driven features, which raises the stakes for platforms trying to stand out through better search and personalized discovery.

In marketplaces where buyers are overwhelmed by choice, the company that can surface the right item first may win the sale. That is especially true in live shopping, where the window to act is short and the cost of poor recommendations is immediate.

Whatnot is betting that improved recommendation speed will help it hold attention longer, convert more visits into purchases and support the expansion of its marketplace into many more niche categories. The Shaped acquisition gives the company a way to sharpen those systems while reinforcing its internal AI talent base.

Why timing matters so much in live commerce

Timing is central to the live-shopping model because the product being sold may exist in only one quantity, for a limited duration, or within a fast-moving auction. If a recommendation arrives too late, it can be useless even if it is accurate.

That is why Whatnot has spent years driving down latency. The company is not just trying to make suggestions better; it is trying to make them relevant at the exact moment they can still influence buying decisions. Shaped’s systems are expected to help close that gap further.

What happens next after the acquisition?

Whatnot has not disclosed financial terms for the deal, and it has not outlined a public roadmap for specific Shaped-powered features. But the structure of the acquisition suggests the company intends to fold Shaped’s expertise directly into product development rather than treating it as a standalone business unit.

In the near term, shoppers may not notice a dramatic redesign. The bigger changes are likely to happen underneath the surface, in how Whatnot ranks inventory, predicts interest and adapts recommendations while live shows are underway.

Over time, that could translate into more relevant search results, more accurate product suggestions and a smoother buying experience across a growing number of categories. For sellers, stronger discovery tools could mean better reach and faster sales. For Whatnot, the payoff could be higher conversion, deeper engagement and a stronger moat in a crowded commerce market.

Timeline of Whatnot’s recent growth and AI push

Whatnot’s acquisition of Shaped is the latest step in a rapid buildout of its marketplace and technology stack. The sequence below shows how the company has scaled and why the deal fits its current stage of growth.

1. 2019: Whatnot launches as a livestream shopping platform.
2. Past six years: The company steadily improves recommendation speed, cutting latency from about a day to a few minutes.
3. Last year: Whatnot adds more than 35 new categories, including art, golf and vinyl.
4. First half of 2025: Another 45-plus categories are added as subcategories continue rolling out.
5. Earlier this year: Whatnot raises $225 million in Series F funding at a valuation above $11 billion.
6. This week: Whatnot announces the acquisition of Shaped to improve real-time recommendations and search.

The bottom line

Whatnot’s acquisition of Shaped is a strategic bet that better AI infrastructure will be crucial to winning live commerce. By bringing in a company built around recommendation and search systems, Whatnot is trying to keep pace with its own growth and stay ahead in a sector where speed, relevance and personalization directly affect revenue.

The deal also highlights how competition in ecommerce is shifting. As marketplaces become larger, more dynamic and more AI-driven, the companies that can process behavior in real time may have the clearest path to growth.