The encyclopedia · Product & Design · Product decision · 2018–2021
Zillow's algorithm bought 27,000 houses — and lost $881M doing it
Zillow's algorithm bought 27,000 homes and overpaid. The division shut down in November 2021 with an $881M charge and 2,000 jobs cut.
Zillow · Opendoor · 2021-11
What happened
Zillow, the online real estate marketplace, launched Zillow Offers in 2018 — an 'iBuying' service that used its Zestimate algorithm to make instant cash offers on homes. The plan: buy homes algorithmically, do light renovations, and resell at a profit. At its peak, Zillow Offers operated in 25 markets and had purchased over 27,000 homes.
The algorithm systematically overpaid. Zillow's models could not accurately predict renovation costs, local market conditions, or the time a home would sit on the market. In Q3 2021, Zillow reported that its home inventory was worth less than it had paid. The company took a $304 million writedown on homes it already owned and projected another $265 million in losses on homes under contract.
On 2 November 2021, Zillow announced it was shutting down Zillow Offers entirely. The company cut 25% of its workforce (approximately 2,000 employees) and took an $881 million total charge. CEO Rich Barton admitted the algorithm could not predict home prices with sufficient accuracy. The division that was supposed to transform real estate was liquidated in a fire sale of 7,000 remaining homes.
Why it happened
- The Zestimate algorithm was designed for browsing, not buying — it could estimate a home's value within a range, but the range was too wide for profitable iBuying at scale.
- Renovation costs, holding costs, and market timing are local and physical; an algorithm trained on listing data could not predict them accurately.
- Buying 27,000 homes created a massive inventory risk — when the algorithm overpaid, the losses scaled with the volume.
- The iBuying model required thin margins and high volume; the algorithm's error rate made thin margins impossible.
The lesson
An algorithm good enough for browsing is not good enough for buying. The Zestimate's $10,000 error range was the entire margin on a $300,000 home.
Aftermath
Zillow returned to its core marketplace business and recovered. Opendoor, the largest pure-play iBuyer, survived but at a reduced scale. The iBuying model proved viable only in specific markets with tight margins. Zillow Offers is cited as the definitive example of applying a digital algorithm to a physical, local, high-variance market.
Sources
- Zillow Offers — Wikipedia (launched 2018; 25 markets; 27,000+ homes purchased; $304M Q3 2021 writedown; $881M total charge; shut down 2 November 2021; 25% workforce cut ~2,000 employees; 7,000 homes fire-sold)
- Stanford GSB — Flip Flop: Why Zillow's Algorithmic Home Buying Venture Imploded (shutdown announced early November 2021; 25% of employees laid off; $569M in write-downs, ~$30,000 per home; CEO Rich Barton: 'unpredictability in forecasting home prices far exceeded what we anticipated'; algorithm missed qualitative factors; Lemons Problem; expanded beyond cookie-cutter homes)
spotted an error? The club wants to know.
More like this
Frigidaire's gas ranges delayed ignition long enough to build up gas
191,000 Aroeve air purifiers recalled after 37 overheating reports
Panasonic recalled 11,480 toaster ovens over an exposed power cord
Somewhere, someone solved the problem this company failed at. 2nd Opinion →
Comments · 0
Sign in to join the comments.