Live Shopping is a TV home shopping experience reimagined for the Internet.
Hosts go live on the platform, demonstrating their products, taking questions from viewers in the comments, and offering limited-time deals. Viewers can make purchases with one tap without leaving the stream. This format reduces the distance between discovery and transaction to just a few seconds.
In China, live shopping will generate about $900 billion in sales in 2025, close to the size of the entire U.S. e-commerce market, consumer intelligence firm NIQ estimated in a press release on Friday (July 17). In contrast, 68% of North American consumers said they have never purchased anything through social media.
The format, which generates hundreds of billions of dollars in Chinese e-commerce, has yet to gain a serious foothold in the U.S.
The gap is not a matter of content. It’s about discovery. Thousands of live streams run simultaneously. Inventory changes constantly. The auction will end in a few seconds. Viewers who join late will miss out on products that would have converted. Matching the right buyers to the right streams at the right time requires personalization at a speed that static recommendation systems cannot deliver. That’s the problem Whatnot, a live shopping marketplace worth more than $11 billion, is trying to solve.
Whatnot processes over 500,000 hours of live video and millions of user interactions every week. The company on Wednesday (July 15) acquired Shaped, a machine learning startup that builds real-time recommendation and search infrastructure, TechCrunch reported Wednesday. Financial terms were not disclosed.
Shaped founder and CEO Tullie Murrell, who worked on recommendation system development at Meta before founding Shaped in 2021, will join Whatnot along with nearly a dozen engineers and researchers to lead a new applied AI research group, the report said.
Live shopping requires AI to match buyers to streams in seconds
The challenges Shaped must address are unique to live commerce. Unlike traditional e-commerce platforms, where product catalogs are relatively stable, Whatnot’s inventory is constantly changing, according to the report. Buyer interests may change during the broadcast.
“By combining Shaped’s technology with Whatnot’s existing systems, we can make recommendations that are faster, more responsive, and more personalized,” said Emmanuel Fuentes, Whatnot’s vice president of data and AI, according to the report.
According to the report, Shaped’s system integrates customer data with AI models and machine learning to deliver personalized search and discovery.
This combination could help Whatnot suggest the most relevant streams to users based on previous purchases and real-time behavioral signals.
According to the report, Whatnot launched over 35 new product categories in 2025 and 45 new product categories in the first half of 2026, with over 1 billion orders placed by sellers.
With each new category, the discovery problem becomes more complex. Shoppers who access your live feed at the wrong time, such as after an item has sold or before they see the item they want, may get little value and may not return.
AI personalization could bridge the gap between content and purchase
The PYMNTS Intelligence report, “A Deep Dive into Agenttic Commerce: Payment Infrastructure and the Path to Seller Trust,” in collaboration with Visa Acceptance Solutions, found that 48% of online shoppers use AI to research recent purchases.
Meanwhile, AI-driven traffic to US retail sites grew 393% year over year in the first quarter of 2026.
Consumer appetite for AI-mediated shopping is growing. Real-time AI can show the right stream to the right buyer at the right time, updating as behavior changes within a session. By acquiring Shaped, Whatnot believes it can move live commerce from a niche format to a mainstream channel by solving discovery problems.
For all of our coverage of PYMNTS AI, subscribe to our daily subscription AI Newsletter.
