How did Zillow get AI so wrong

This is the story of how Rich Barton, the billionaire founder behind Expedia and Glassdoor, came back to Zillow as CEO and made a huge bet on AI. Zillow's 2021 shopping spree is over.

Turns out a lot of those flips turned out to be flops. The biggest fail in business for the year.

This gamble that an algorithm could buy and sell homes better than any human. But less than a year later, the strategy collapsed. Zillow has decided to stop buying homes and flipping them for resale. $880 million in losses, 2,000 layoffs, and lawsuits from investors. Barton says it wasn't a total failure, but in this video, I'll walk you through what actually happened. It's a clear warning about what can go wrong when leaders put too much faith in AI and cut out human judgment.

Zillow offers this kind of a magic thing that just short circuits that and solves that for people.

All right, before we get into Zillow, you need to know who Rich Barton is, because this guy is not your typical CEO.

My name is Rich Barton.

He's one of the rare founders who's built multiple unicorns. He launched Expedia in the 90s, then co founded Glassdoor, which sold for over a billion dollars in 2018. He takes big swings, even when the risks are obvious.

I had this bhag, this big, hairy, audacious goal that was become the largest seller of travel in the world.

And in 2019, he returned to Zillow to take his biggest swing ever. On his first earnings call, he told investors Zillow would eventually sell 5,000 homes a month. That's enough to rival the biggest home builders and in the US up until this point, Zillow was basically a home listing site. It was simple, profitable and light on overhead. Barton wanted to pivot that 19 year old business into a capital intensive operation. The goal was to transform Zillow into a real estate transaction machine. A move centered around Zillow offers the company's iBuying division. Barton wasn't inventing this out of thin air. IBuying was already a thing. Opendoor had been doing it for years. And Barton saw that company as a serious threat. If they could bypass platforms used to list and sell homes, Zillow could get cut out entirely. But Barton believes Zillow had an edge, speed, scale and the Zestimate, which is their home valuation tool.

We found the Zestimate, which was our killer, our killer feature.

So here's how Zillow offers actually worked because honestly, it sounded Great. If you wanted to sell your home, you could skip the showings, paperwork and negotiations. Instead, Zillow's algorithm would generate an instant offer using housing data. If you accepted, they'd buy your house in a few days. And for a bit, it looked like it was working. Homeowners loved the convenience. The housing market was on fire, and Barton even told investors this could become a $20 billion a year business. But here's the part of this story that people miss. At first, Zillow's iBuying approach was actually pretty cautious. The AI model would first suggest a price, then human analysts reviewed every deal. Next, inspectors would visit homes, estimate renovation costs, and flag any issues. So the system had plenty of guardrails. But Barton started pulling them off because the pressure to grow was boiling over. By 2021, Zillow was falling behind their home buying targets. That's when Barton launched something called Project Ketchup to catch up to the competition. Employees even got ketchup themed merch, including actual bottles of Heinz Cringe. Branding aside, the strategy was clear. Zillow would move faster, spend more, and cut out human verification. Barton started over, relying on the model. Analysts were told to stop questioning it, and Zillow would actually bump the model's price, sometimes by tens of thousands, just to win bids. Internally, they called this offer calibration. To go even further, they introduced auto evaluations, effectively letting sellers inspect their own homes. What happens next is what actually killed Barton's iBuying scheme. Just as Zillow ramped up its buying, the housing market started to change. OpenDoor, Redfin and OfferPad pulled back on home purchases. Zillow didn't. And with safety nets gone and fewer humans in the loot, the AI model was flying blind, with no one there to stop it. Analysts raised concerns because they could see Zillow was overpaying. But they were told it was all part of the plan. In Q2 of 2021, Zillow bought 3,000 homes, doubling the previous quarter's purchases. Barton even bragged to investors that the biggest complaint came from customers. It was that the offer seemed too good to be true. By September, two thirds of the homes Zillow had purchased were worth less than what they paid. One month later, Barton finally hit pause on new purchases. Zillow is hitting pause. They're pressing pause. Zillow is pausing.

Oh, my gosh.

Publicly, he blamed labor shortages and supply chain delays. But inside the company, it was clear the strategy had failed. Zillow started offloading homes at discounts, almost like a liquidation sale. Internally, panic set in. In November, Barton shut down Zillow offers completely, and the stock tanked, wiping out over $9 billion overnight. And with the strategy dead and losses mounting, there was only one thing left to do. Layoffs.

When that home flipping branch of Zillow goes out of business, about 2,000 people will lose their jobs.

Now, Barton has built some great companies, but this. This is not how you want to fire 2,000 people. Employees were summoned to a video call, and according to former workers, Barton didn't speak. Instead, the VP of Zillow offers read a quote from Dr. Seuss, don't cry because it's over. Smile because it happened. Some were asked to stay on just long enough to help sell off the remaining homes. Zillow tried to soften the blow with bonuses and severance, but senior leaders still left, including the company's chief economist. He insisted iBuying wasn't a total waste. Instead, he promised to repurpose the tech into a super app for housing with 45% margins. But investors ended up suing Zillow, claiming the company misled them about the accuracy of its AI model. When questioned by reporters, Barton quoted the big Lebowski, new shit came to light. As of 2025, the lawsuit is still unresolved. Barton has since left the company, and competitors like Opendoor, OfferPad, and Redfin, who are all using similar AI models, they're still flipping homes. Here's the thing about this story that blows my mind. Even if you remove AI from it, Zillow would have probably still got it all wrong. This is not really about AI. It's about chasing growth at all costs. And that's the problem. So what's the real takeaway here? Even when AI works, it still needs humans to have good judgment on how to use it properly.