Zillow just announced it is shutting down its iBuyer program, Zillow Offers, because its pricing algorithm is unable to accurately predict the future value of housing. Its algorithm is actually quite well known: the Zestimate.
Zillow has done an amazing job convincing the public that the Zestimate accurately reflects the market value of their home. For years a lot of my client conversations include, at some point the client stating, that their property is worth $X because "the Zestimate said so." (Often with conviction.)
At those moments I understand how doctors must feel when we march into their offices armed with our google-researched diagnoses and treatment protocols.
I have nothing against Automated Valuation Models (AVMs). They're good tools to begin putting rough boundaries around a property's value. The biggest mortgage institutions in the country, Fannie Mae and Freddie Mac use AVMs extensively for initial risk assessment on loan applications, and their algorithms are far more robust than any Zestimate, Redfin, or other consumer-facing models.
But in the end, these mortgage giants still require a human with knowledge and expertise to go to the property and provide the analysis and ultimate opinion of value. They know that true market value is derived from things that only knowledgeable people, not machines, can see and interpret.
The Kasparov Principle posits that the best outcomes actually come from humans enhanced by technology, not by either alone.
AVMs are good at evaluating fixed, quantifiable variables: square footage, number of rooms, bedrooms, bathrooms, lot size, and even aggregate market momentum. They fail with the extensive observations and conversations that can't be pulled from a database, yet have a significant impact on what someone might pay for a home.
An AVM might compare 3 bed, 2 bath, 1500 sq. ft. homes, but will never know that one has no true master en-suite bedroom, or that one has a floor plan and siting that makes it very dark, or that one is renovated and modern while the other is, let say original. The lots may encompass the same square footage, but one is an odd triangular shape or a flag lot. Maybe one's neighbor keeps a chicken coup on the fence line, or one was a distressed seller. I could go on, but you get the point: to truly understand the market value it takes local, human investigation, expertise, and understanding. Otherwise, your Zestimate is just a Guesstimate.
Despite their popularity, the Zestimate and other consumer models are known to be highly inaccurate. Pick any address and check its value on different sites: Zillow, Redfin, Realtor.com, In most cases, the estimate ranges will vary somewhere in the 5% - 10% range.
Zestimate Epic Fails
In 2016 Spencer Rascoff, then the CEO of Zillow sold a home in Seattle with a Zestimate of $1.75M. The home sold for only $1.05M, 40% below the Zestimate! Embarrassing fail #1. Then in 2019 Zillow took the big swing, refocusing its entire business model to enter the brokerage business as an iBuyer, with its Zestimate valuation model at the center. Barely two years later we have epic fail #2, and Zillow's stock price dropped ~20% in one day on the news.
Of course, there are other factors that contributed to the death of Zillow Offers, but the inaccuracy of fully automated valuation models lies at its center. Other iBuyers claim to have better models, and maybe it's true. But a rising tide floats all boats. The iBuyer model is a financially precarious one. We'll see what happens to this slice of the industry when the inevitable market cycle rolls in.