The Need for Speed: How Whatnot is Leveraging AI to Revolutionize Live Commerce

In the high-stakes, hyper-fast world of live-streamed retail, the difference between a successful sale and a missed opportunity is often measured in seconds. For Whatnot, the leading marketplace for collectibles and rare items, this reality has necessitated a radical shift in how it approaches product discovery and recommendation systems. By prioritizing extreme low-latency data ingestion, the company is attempting to solve one of the most complex puzzles in modern e-commerce: matching high-intent buyers with fleeting, real-time inventory.

This strategic imperative reached a milestone last week with the announcement that Whatnot has acquired Shaped, an innovative machine learning startup that specializes in real-time retrieval engines for search, feeds, and AI agents. The acquisition signals a deeper commitment to ensuring that Whatnot’s recommendation engine is not just accurate, but instantaneous.

The Chronology of an Evolution

Whatnot’s journey toward an AI-integrated marketplace did not happen overnight. Since its inception, the company has operated on a model that prizes human connection and community-driven commerce. However, as the platform scaled, the sheer volume of live shows and shifting inventory necessitated a more sophisticated backend.

• Early Days: The platform relied on manual discovery and broad categorization, focusing primarily on building the community of sellers and buyers.
• The Scaling Phase: As the volume of concurrent livestreams increased, the product team—led by Chief Product Officer Tom Verrilli—began focusing heavily on the "discovery" problem. They recognized that if a user cannot find a show relevant to their niche interests within seconds, they will leave the platform.
• The Latency Milestone: Over the past few years, the team has successfully reduced recommendation latency to just minutes. This allows the system to ingest data—such as a product selling out during a live show—and instantly adjust the algorithm to stop recommending that item while prioritizing other available inventory.
• The Shaped Acquisition: In 2024, the acquisition of Shaped marked a shift toward institutionalizing these real-time capabilities. Shaped, which has previously worked with heavyweights like Vox Media and QVC to personalize feeds, brings a specialized technical stack that will be integrated into Whatnot’s existing infrastructure.

Understanding the "Live" Complexity

To understand why Whatnot is investing so heavily in machine learning, one must first understand why "live" is a fundamentally different beast than traditional e-commerce.

In a traditional, static retail environment—such as Amazon or a standard Shopify storefront—the inventory is largely fixed. A customer can browse products that have been sitting in a warehouse for weeks. The recommendation engine has the luxury of time; it can analyze historical purchase data and suggest items that are statistically likely to be bought.

"Genuinely, the thing that sets live commerce apart and makes it a uniquely hard recommendations problem is the speed," Verrilli explains. "Inventory changes second by second. We actually don’t know when a show is going to end. You can’t necessarily turn through things like you’re recommending movies."

When a consumer sits down to watch a movie on a streaming service, the algorithm knows the user has a long-form engagement period. In contrast, a Whatnot user might hop into a stream, watch for three minutes, bid on a rare trading card, and leave. If the recommendation system suggests a product that sold out five minutes ago, the user experience breaks.

Supporting Data: Why Latency is the North Star

The drive for millisecond-level updates is not just a vanity metric for the engineering team; it is a direct contributor to the platform’s bottom line. Sellers are drawn to Whatnot precisely because it allows for rapid inventory turnover.

"One of the things that makes sellers choose Whatnot more than anything else is that you can make more money on Whatnot than anywhere else," Verrilli says. "That’s because you can be running auctions quickly. You can also just respond to whatever your customers want in real time, as opposed to the traditional model where I buy a million dollars of inventory, upload it to a website, and wait and see what will sell."

The data supports this "high-intent" model. The average user on Whatnot spends approximately 95 minutes per day on the app. This high level of engagement is predicated on the feeling that the user is part of a live, evolving event. By keeping the recommendation engine in sync with the current state of a live show, Whatnot ensures that the "intent" of the buyer is met with the "current reality" of the seller’s inventory.

The Role of AI: A Tool, Not a Replacement

In an era where "AI" is often used as a catch-all marketing buzzword, Whatnot is taking a nuanced, utilitarian approach. Under the guidance of VP of Data and AI Emmanuel Fuentes, the company has integrated machine learning across various departments, but it draws a firm line at replacing the human element of the business.

"We don’t believe that the best use of AI is trying to replace the seller who is actually merchandising the goods," Verrilli notes. "We think fundamentally that human connection—of having a live seller selling—is the thing that generates trust and expertise."

Instead, the company focuses on using AI to reduce the "drudgery" of being a seller. These tools include:

• Automated Listing: Using computer vision to convert images of inventory directly into structured product listings.
• Operational Support: Baking AI into scanners to help sellers track inventory during the fast-paced chaos of a live auction.
• Logistics: Using machine learning to optimize shipping labels and provider selection, ensuring items get to buyers faster.

This philosophy extends to the recent discourse regarding AI-generated show thumbnails. While some users have expressed frustration on social media about the influx of AI-generated content on the app, Verrilli remains pragmatic. "Sellers are able to create their own thumbnails to best merchandise their show… If they want to use AI tools to do that, I’m OK with that. We have no intent to try and manage that out of the system."

Implications for the Future

The acquisition of Shaped and the continued focus on real-time latency suggest that Whatnot is positioning itself to be more than just a marketplace; it is becoming a highly efficient, AI-augmented infrastructure for the creator economy.

1. The Rise of the "Intelligent" Marketplace

As the platform continues to refine its recommendation engine, the barrier to entry for new sellers will likely decrease. By automating the backend—inventory management, shipping, and discovery—Whatnot allows the seller to focus exclusively on their performance and community engagement.

2. A Shift in Buyer Expectations

As users become accustomed to the "real-time" nature of Whatnot’s recommendations, they will likely come to expect the same level of speed across all digital experiences. This puts pressure on competitors to either innovate their own recommendation stacks or risk appearing "stale" in comparison.

3. The "No AI Priorities" Stance

Perhaps the most significant takeaway from Verrilli’s approach is the rejection of "AI priorities" as a standalone corporate goal. By framing AI as a means to achieve "seller and buyer priorities," Whatnot avoids the common pitfall of implementing technology for technology’s sake.

"I don’t have AI priorities—I have seller and buyer priorities," Verrilli states. "For sellers, we want to continue to take the work out of being a seller. And for buyers, we’ll continue to invest in discovery so that people can find great sellers."

Conclusion

Whatnot’s acquisition of Shaped is a testament to the fact that in the world of live commerce, the technology stack is just as important as the product itself. By treating recommendations as a dynamic, real-time data problem rather than a static catalog search, Whatnot is securing its position as a dominant force in the creator economy.

As they move into the remainder of the year, the company’s path is clear: continue to use technology to strip away the friction of commerce, allowing the human connection between buyer and seller to take center stage. For the users of Whatnot, this means more relevant, timely, and engaging experiences, proving that while AI may be the engine, the human seller remains the heart of the platform.