Zillow Tried to Buy Houses With an Algorithm and Lost Money Fast
The company shut its home buying operation in November after discovering that estimating a price and committing capital at that price are entirely different businesses.
The Idea
Zillow had spent years building a home valuation estimate that consumers used constantly. The logical extension seemed obvious. If the model can price a house accurately, the company can buy houses directly, make light repairs, and resell them for a spread, offering sellers speed and certainty in exchange for a fee.
This business, known as iBuying, was not unique to Zillow. Several firms pursued it. Zillow's advantage was supposed to be the model and the traffic.
Why the Model Was Not the Business
The failure is a precise lesson in the difference between a prediction and a position. An estimate of value carries no consequence when it is wrong. It is a number on a webpage, and errors in both directions average out across millions of listings.
Buying the house converts that estimate into a position. Now an overestimate means overpaying with real capital, and errors no longer offset because the company only buys the homes it valued too generously. If your bid is above the true value, you win the auction. If it is below, someone else buys it.
Being wrong in a valuation model costs nothing. Being wrong while bidding means you specifically acquire the houses you misjudged, which is adverse selection doing its work.
The Timing Made It Worse
The operation scaled aggressively into a market that was moving fast in both directions. Home prices had risen sharply through 2021, and a model trained substantially on recent appreciation will extrapolate that appreciation forward.
When the pace of appreciation slowed in the second half of the year, the company held inventory purchased at prices assuming continued gains. Reports at the time indicated a large share of its holdings were carried above what it could sell them for. Add renovation costs, property taxes, insurance, and financing while the home sits, and the negative carry accumulates every month.
The Operational Constraint Nobody Modeled
The other binding limit was physical. Reselling thousands of homes requires contractors, inspectors, photographers, and closing capacity in each local market. Those resources do not scale with a software deployment.
Labor and materials were both scarce in 2021, so renovation timelines extended, which extended holding periods, which increased carrying costs. A business with thin intended margins cannot absorb an extra sixty days of holding cost per unit.
What It Says About Capital Intensity
The strategic error was treating a capital intensive, operationally complex business as a software business because software was involved. Zillow's original model was excellent: aggregate listings, sell advertising to agents, earn high margins on a business requiring little capital.
iBuying required buying inventory with borrowed money, managing physical assets, and accepting price risk on every unit. The margin structure and the risk profile were completely different, and the brand and traffic that made the advertising business strong did not transfer.
The Bottom Line
Zillow proved that predicting a price and committing capital at that price are different disciplines. The moment you bid, you stop measuring the market and start being selected against by it.