Full width home advertisement

Welcome Home


Post Page Advertisement [Top]

Zillow's home-buying fiasco demonstrates how difficult it is to use artificial intelligence to value real estate



According to a February announcement, Zillow is so confident in its ability to estimate the value of homes using artificial intelligence that it is offering a new option: for certain homes, the company's so-called "Zestimate" will also serve as an initial cash offer to purchase the property.

 

Described at the time by a company executive as a "exciting advancement," the move was intended to streamline the process for homeowners who were considering selling to Zillow as part of the company's home-flipping business. In its marketing, Zillow promoted this option as a way to make selling a home more convenient while also reducing interactions with other people during the pandemic. However, only eight months after starting the business, Zillow Offers, the company has decided to shut it down completely.

 

In the wake of the decision, which was announced last week, Zillow has suffered a stunning setback in its business. Real estate listing company Coldwell Banker recorded a $304 million inventory write-down in the third quarter, blaming the loss on recently acquired homes that were purchased at prices that were higher than the company believed they could be sold for. The stock price of the company has plummeted, and the company now plans to cut 2,000 jobs, or 25 percent of its total workforce.

 

The fallout from this business venture, on the other hand, demonstrates more than just the difficulties associated with buying and selling houses for a profit. As a result, it demonstrates just how difficult it is to use artificial intelligence to assist in making costly real-world decisions, particularly in an ever-changing market that can be difficult to forecast months or even weeks in advance, and with prices that can be influenced equally by gut instinct and hard data points. "Unpredictability in forecasting home prices," said Rich Barton, CEO and cofounder of Zillow, as the reason for the company's decision to discontinue Zillow Offers. He added that the company's expectations for the service had been "far exceeded."

 

Other real estate companies, such as Zillow, operate under the "iBuyer" model, which involves purchasing homes directly from sellers and then re-listing them after making minor repairs. For Zillow, one of the first steps in deciding whether or not to purchase a home is to obtain a "Zestimate" — a machine-learning-assisted estimate of the home's market value calculated using reams of data about the property gathered from sources such as tax and property records, homeowner-submitted details such as the addition of a bathroom or bedroom, and photographs of the house — before visiting the property. Redfin and other competing platforms, for example, have their own estimates based on comparable data.

 

Homeowners who are interested in selling their property to Zillow can learn more about how the company calculates an estimated sale price by visiting its Zillow Offers webpage. "An estimated sale price is calculated using the Zestimate, the facts you provide, and comparable homes nearby," Zillow explained to homeowners who are interested in selling their property to the company. This page has been updated to reflect the fact that the company is "winding down" the service and is no longer accepting new offers on properties. As a result of that estimate, Zillow visits the property in person to assess its condition, determine the cost of repairs necessary to resell it, and then makes a final offer. When Zillow Offers first launched, the company acquired a large number of homes, but it has sold far fewer than it has acquired: according to the company's most recent quarterly results, the company acquired 27,000 homes between April 2018 and September 2021, but sold nearly 17,000 of them.

 

A request for an interview with Krishna Rao, Zillow's vice president of analytics, was turned down by the company's executives. A statement from Zillow stated that the company used Zestimates for Zillow Offers "in the same way that we encourage the general public to use it: as a starting point."

 

According to Shelton, "the challenge we faced with Zillow Offers was accurately forecasting the future price of inventory three to six months out in a market where home values were changing at a faster rate than we had ever seen before."

 

As a matter of fact, since Zillow entered the home-flipping business in 2018, real estate markets have experienced extreme fluctuations. Due to the pandemic, the housing market experienced a brief halt. This was followed by a supply and demand imbalance that resulted in an unprecedented rise in home prices. According to Zillow, this may have complicated the company's decision to include the Zestimate — which is not an appraisal but rather a "computer-generated estimate of the home's current value, based on available data" — as part of the Zillow Offers process in more than 20 cities, despite the fact that it is not an appraisal.

 

Using artificial intelligence to determine an appropriate price for a home, artificial intelligence can analyze far more data, much more quickly, than a single human can, taking into consideration factors such as comparable home sales in the area, the number of people actively searching in a particular neighborhood, and so on. In spite of this, "you can have a real estate agent look at a house and pick out one critical factor of the valuation in a split second that simply does not exist as ones and zeros in any database," according to Mike DelPrete, a real estate technology strategist who is also a scholar-in-residence at University of Colorado Boulder.

 

Because it was introduced with the launch of Zillow's website in 2006, the Zestimate has become an integral part of the company's brand. In addition to being prominently displayed on millions of Zillow's home listings, the term has been trademarked by the company and appears 61 times in the company's 2011 initial public offering documentation.

 

Using over 500,000 unique valuation models built on 3.2 terabytes of data, the company generates current Zestimates for over 70 million US homes three times per week, according to a 2011 securities filing. After more than a decade in business, the company now publishes Zestimates for more than 100 million homes throughout the United States.

 

On Zillow's website or mobile app, when looking for a home, the Zestimate is prominently displayed alongside each listing, regardless of whether the home is currently on the market. If the house is currently for sale, a red dot will appear next to the words "House for sale," and the Zestimate will appear on the same line, if it is available for that particular home, if it is available for that particular home.

 

However, while the company emphasizes that the Zestimate is not a real-estate appraisal, the accuracy of the feature has been called into question over time. The subject of a lawsuit brought by homeowners, for example, was the subject of one in 2017. (That particular lawsuit was ultimately dismissed.)

 

It has taken years for Zillow to perfect the Zestimate algorithm, and the company even held a multi-year data science competition to improve the algorithm's accuracy. The $1 million prize was awarded to a three-person team by the company in the beginning of 2019.

 

Shelton explained that Zillow's Zestimate currently has a median error rate of 1.9 percent for homes currently on the market, which means that half of the homes currently on the market have a Zestimate that is within 1.9 percent of the actual selling price, on average. According to Shelton, the percentage of error is much higher for off-market properties, at 6.9 percent. Although the difference between a Zestimate of $500,000 and the actual value is as small as 1.9 percent, the difference is nearly $10,000 when multiplied by the thousands of homes in various cities throughout the United States.

 

An accurate model on a website is one thing; creating an accurate model in real life is quite another. Then there's the challenge of trying to apply that model in the real world to make extremely expensive bets — and doing so on a large scale, according to Nima Shahbazi, a member of the winning Zestimate algorithm team and the CEO of Mindle.AI, which assists businesses in using artificial intelligence to make predictions. The Zestimate would be unable to predict issues such as a missed crack in the foundation in any Zillow-purchased homes, for example, if any Zillow-purchased homes had hidden issues such as a missed crack in the foundation, he explained.

 

It is possible for a number of components in the process of developing a very good model and putting it into production to fail, according to him.

 

It was Zillow that was utilizing the Zestimate to assist it in making purchasing decisions for homes on which it was hoping to make a profit in the long run. In contrast, Nikhil Malik, an assistant professor of marketing at the University of Southern California, believes that algorithms are generally good at making fine-grained, short-term predictions, such as forecasting stock prices a split second ahead of time. Malik, a researcher who specializes in algorithmic pricing and has conducted extensive research on the Zestimate in particular, believes that there is simply not enough data for an algorithm to learn about longer busts and booms.

 

DelPrete pointed out that when determining the value of a home, there are numerous intangible factors to consider, such as the value of living in the neighborhood where you grew up or across the street from your parents. Individuals' preferences can differ significantly, making it even more difficult to automate the home valuation process.

 

While the Zestimate is a useful tool for what it is, DelPrete cautioned that it is a mistake to believe it can accurately predict current or future house prices. According to him, it's "almost a toy" that's meant to pique your interest when you're searching for your or your neighbor's home online.

 

In order to participate in iBuying and make thousands of offers daily, he explained, "you must be extremely adept at valuing homes, not only today, but three to six months from now." That, my friends, is both an art and a science in its own right."

 

No comments:

Post a Comment

Bottom Ad [Post Page]