Signal & Noise
Menu
Real EstateAI / MLNoise

The Zestimate could size up a market. It couldn't safely bet the company's balance sheet on one.

Zillow ยท 2018โ€“2021

What they did

Zillow Offers used an ML pricing model to make instant cash offers on homes, renovate them, and resell for a margin ("iBuying") โ€” a business model that depended on predicting a specific home's near-term resale value within a narrow error band, not just estimating a market trend.

What happened

During the 2021 pandemic housing boom, prices moved faster than the model could track, and it overpaid systematically in hot markets like Phoenix and Atlanta. In November 2021 Zillow shut the division entirely, took a loss of over $500 million on homes it couldn't resell at a profit, and cut 25% of its workforce.

The so-what

A model that's a genuinely useful directional estimate (the Zestimate, still live today) can be actively dangerous the moment a business puts real capital behind its point estimate being right within a few percent, at scale, in a volatile market.

Sources

Related case studies

โ† Back to all case studies