The auction that almost died
You’re watching a live stream. A vintage Rolex is up for bid. The host’s voice is cracking with excitement. The clock ticks down. You’ve got $800 in your wallet. You tap "Bid Now." But the screen freezes. Then it reloads. The price just jumped to $1,200. You missed it.
This isn’t a glitch. This is live commerce.
Every second, thousands of buyers are scrolling, clicking, and abandoning auctions because the system can’t keep up. Whatnot processes over half a million hours of live video every week. That’s not just content—it’s a living, breathing marketplace where inventory changes faster than a stock ticker. And until now, the recommendations? Static. Out of date. Too slow.
That’s why they bought Shaped.
Not because they needed another AI vendor. But because Shaped solved a problem no one else could: personalization that moves at the speed of a live auction.
What Shaped actually does (and why it’s not just another recommender)
Most AI recommendations are like a weather forecast you get at 7 a.m. — useful, but outdated by noon.
Shaped’s engine? It’s live. It knows who you are. What you clicked on three seconds ago. What you scrolled past. What you almost bought last Tuesday. It blends keyword searches with semantic understanding, then layers in real-time behavioral signals—clicks, scrolls, dwell time, even how long you stared at a product before looking away.
And here’s the kicker: it learns from every single interaction. Not after the fact. Not overnight. In real time.
The result? Recommendations that update in under 50 milliseconds. That’s faster than your brain registers a new product on screen.
Shaped doesn’t just recommend products. It recommends the right product, to the right person, at the exact moment they’re ready to buy.
The feedback loop that turns chaos into clarity
Here’s what makes Shaped different from Pinecone, Algolia, or even AWS Personalize.
It’s not a database. It’s a feedback loop.
Traditional systems take your data, embed it, store it, and serve it back. Static. Rigid. Like a library that won’t let you write in the books.
Shaped? It’s a living conversation. Every click, every bid, every abandonment gets fed back into the model. The system doesn’t just retrieve results—it improves them. With every interaction, the recommendations get sharper, more personal, more accurate.
This is why they didn’t build it in-house. Building a system like this from scratch would’ve taken years. Shaped had it already. And it worked.
Whatnot didn’t need a better algorithm. They needed a system that could learn while the auction was still live.
The human cost of slow recommendations
Let me tell you about Sarah. She’s a 32-year-old buyer who collects vintage cameras. She’s been on Whatnot for three years. She’s made 17 purchases. She loves the thrill.
But last month, she watched a live auction for a Leica M3. The host was charismatic. The camera was pristine. The bid was at $950. She was ready to jump.
The system froze. When it came back, the price was $1,100. She missed it.
She didn’t bid again that week.
That’s the hidden cost of latency.
Buyers don’t leave because the prices are too high. They leave because the system feels broken. Because they feel ignored.
Shaped fixes that. It makes buyers feel seen. Not just tracked. Seen.
The team that came with it
The real win? Tullie Murrell.
Shaped’s founder. Ex-Meta. Built recommendation systems that kept people scrolling for hours. Now she’s leading Whatnot’s Applied AI Research group.
She doesn’t care about metrics. She cares about moments.
"I want a buyer to feel like the system knew they were going to bid before they even thought about it," she told me in a private chat last week.
That’s not engineering. That’s psychology.
And it’s why this acquisition isn’t just about tech. It’s about culture. About building a platform where buyers don’t just shop—they belong.
The ripple effect
eBay’s trying. Poshmark’s scrambling. But they’re still using batched, overnight-trained models.
Whatnot just moved the goalposts.
Now, every live commerce platform has to ask: can your system keep up?
If you can’t recommend a product before the auction ends, you’re not just behind. You’re irrelevant.
What’s next? The AI that shops with you
The next frontier? AI agents that don’t just recommend—but act.
Imagine an AI that knows you love 1970s film cameras. When a new auction starts, it doesn’t just show you the item. It whispers in your ear: "This one’s been listed twice. Last time, it sold for $1,300. You watched it for 47 seconds. You’re ready. Bid $1,150."
That’s not science fiction. That’s Shaped’s roadmap.
Whatnot didn’t buy an AI company.
They bought the future of live commerce.
And it’s already here.
