Zillow Promised to Buy Your House in Two Days. It Lost $881 Million Finding Out Why That Doesn't Work.
An algorithm built for speed instead of accuracy bought thousands of homes for more than they were worth.
Zillow spent 2019 through 2021 running an experiment that sounded, to a stressed-out home seller, almost too good to be real: skip the listing, skip the showings, skip the months of uncertainty. Tell Zillow's algorithm about your house, and within two days, Zillow itself would make you a cash offer and buy it outright. In November 2021, Zillow shut the entire program down, having lost $881 million finding out exactly why that promise doesn't scale.
An Algorithm Betting Against Its Own Speed
Zillow Offers — launched in December 2019 as an expansion of its earlier Instant Offers pilot — depended on a pricing algorithm that had to move fast enough to compete with the two-day promise, which meant it had to price homes with far less scrutiny than a human buyer would apply. Speed and accuracy were in direct tension, and the program was built to favor speed. For a while, in a rising market, that tension didn't show up in the numbers. Then the market's direction stopped cooperating.
Overpaying at Scale, Then Selling at a Loss
By the time Zillow pulled the plug, its home-flipping segment had lost roughly $420 million in just the third quarter of 2021 alone, on top of the $881 million the segment lost across the full year. The company had, in effect, bought thousands of homes at prices its own algorithm now admits were too high, then had to resell many of them for less than it paid — the exact opposite of what a house-flipping business is supposed to do, executed at national scale before anyone caught it.
Two Thousand Jobs, One Bet on Forecasting
Zillow laid off approximately 25% of its workforce, roughly 2,000 employees, as part of unwinding the program — a direct cost of a pricing model that turned out to be much better at generating fast offers than at forecasting where home prices were actually headed. The company's own public statements pointed to "unpredictability in forecasting home prices" as the reason for shutting the business down entirely, rather than trying to fix the algorithm and continue.
A Nice Thing That Only Worked If Nobody Looked Closely
The instant-offer pitch wasn't a scam and nobody was defrauded — sellers who used it got exactly what they were promised, fast and in cash. The part that couldn't survive contact with reality was the business built underneath it: a company betting it could out-price the entire national housing market by moving faster than everyone else, with no room in the model for being occasionally, expensively wrong. It only takes one bad forecasting year to turn "instant offer" into "instant $881 million problem."
Filed from public reporting:
