Zillow & Zestimate: The Future of Real Estate?
The speakers during Friday’s lecture focused primarily on machine learning and it’s applications to different fields. In this blog post, I’ll be exploring the impact machine learning has had in the real estate industry, specifically Zillow’s Zestimate tool.
Companies like Redfin, Zillow, and Trulia are changing the landscape of an industry that’s been entrenched in it’s old ways for decades. Michael Krigsman focused on the rapid growth of Zillow over the past decade in an article published on July 30th, 2017 titled “Zillow: Machine learning and data disrupt real estate”. Anyone searching for a new home (or even existing homeowners) are well aware of Zillow’s tool, Zestimate. Introduced in 2006, the tool, “uses a variety of data sources and models to create an approximate value for residential properties.”Krigsman argues that tools like Zestimate have primarily helped only the buyer, as it allows for significant transparency into the world of appraisals and price information that’s generally only been available to real estate agents and brokers in the past. Patrick Sisson agrees with Krigsman’s estimate in an article published on July 28th, 2017 titled “Will the home appraisal industry be replaced by technology?” For the agents, he argues, it’s created a headache, as it creates a sense of price expectations amongst consumers.
Stan Humphries, the Chief Analytics Officer, described the motivation for starting Zillow as similar to that of many other tech startups, like Uber and Airbnb. He felt that there was a significant amount of data that was being under (or improperly) utilized. He said, “We had a sense that it was, from a consumer’s perspective, like the Wizard of Oz, where it was all behind this curtain. You weren’t allowed behind the curtain, and [we figured] you’d like the website to show you both the core sale listings and the core rent listings.”
Despite the easy access Zillow provides to consumers in order to make a more educated real-estate decision, they argue that there is still an important place for the real estate agent. Instead of seeking to completely replace real-estate agents with their tools (like what Redfin’s attempting to accomplish), Zillow seeks to “enrich” the conversations between the real estate agent and consumer. By being armed with a lot more data, they feel that consumers are able to approach the transaction with a lot more confidence, thereby allowing them to make an unbiased decision with the help of the realtor.
So how does machine learning fit into all of this? When Zestimate was first introduced in 2006, Zillow had approximately 43 million homes in their database and they used a few terabytes of data to run about 34,000 statistical models to come up with estimated valuations of each home. Compared to what was the status quo at that point, this was considered groundbreaking from both the computational standpoint as well as the number of models used. “Back in 2006, when we launched, we were at about 14% median absolute percent error on 43 million homes…As we went from 43 million to 100 million homes…we almost tripled our accuracy rate from 14 percent down to 5 percent,” says Humphries. The drop in error can be attributed to the significant increase in the amount of data collected, as well as the accuracy and sophistication of the algorithms now used. For comparison’s sake, Zillow now uses seven to eleven million statistical models that evaluate home valuations daily.
Although Zestimate and other similar software will certainly impact the appraisal field within real estate, it’s not going to replace it. Software like Zestimate ultimately depend on data inputs (billions of them) from actual appraisals to accurately predict home prices. Without a base starting point of human-produced data points, Zestimate’s predictions will become far less accurate.
5 comments on “Zillow & Zestimate: The Future of Real Estate?”
Comments are closed.
Jega is also a graduate student in the Department of Aeronautics & Astronautics at Stanford University. He graduated from the University of California, Berkeley in 2015 with a B.S in Mechanical Engineering (Honors) and a minor in Public Policy. Before joining Boeing full-time in August 2015, he interned at NASA, Jaguar Land Rover, and General Motors.
His interests include aircraft design, aircraft interiors, commercial aircraft product strategy, science & technology policy, and intellectual property law. In his free time, Jega enjoys traveling, reading, and coaching high school debate.
Great post, this is a very interesting application and strikes a chord with me as I am looking for an apartment and using these services…
The Zestimate tool is similar to real estate to what the kelley blue book value does for cars. I have found it to be helpful just to get a baseline for what to expect in different neighborhoods. Incidentally kelley blue blue also seems to be incorporating more AI into its services …
https://venturebeat.com/2017/07/12/kelley-blue-book-turns-to-aws-for-ai-to-power-its-new-chatbot/
I hadn’t thought about the wizard of oz “behind the curtain” part of it before. Personally I have definitely had times where I am surprised by the number and want to know more about where it comes from. This would be especially helpful when talking to realtors as simply pointing to a number on a website doesn’t do much to validate an opinion. Though I would like more transparency, your post has certainly given me more appreciation for the tools I have been taking for granted recently.
Hey Jega,
Thanks for the post; interestingly this has come up several times at work.
A bit of background, we took a technology company public called Real Matters (REAL) at a valuation of about ~1B. They essentially operate a platform/network for mortgage appraisals. TAM is about ~16B of which they are one of the largest players in a fragmented market with around 5% market share. They point to performance metrics like the turnaround time of 5.3 days vs. 7-9 days for a traditional appraiser and defect rates of 5.6% vs 15-20% for traditional appraisers of why their network/platform for appraisals is better and more accurate than the traditional model.
On the flip side you have companies like Zillow and Trulia using ML to value properties. When we posed the question to REAL about whether they saw these tools like Zestimate as potential disruptors to their business model and they said that the error rates for valuations from these data driven tools like Zestimate were too high and weren’t suitable for a highly regulated industry like the mortgage appraisal market. They said the issue really came down to the quality of the underlying data and the fact that a lot of data, particularly at the government level was still not digitized and was still siloed. I guess at what point do you think these ML driven valuation tools will be good enough to replace traditional valuation methods? I see that it already has for your average consumer but for sophisticated investors, regulated banks and insurers it’s still a bit for off in terms of quality and accuracy.
I agree with your closing point that these ML driven tools will be complementary and human produced data points are still going to be a big part of the real estate valuation industry going forwards.
Cheers,
Johnny
Hi Jega,
Thanks for an interesting post! You’re right about pointing out that AI in real estate appraisal is not going to replace human inputs and interactions. I took a real estate finance class at Stanford last quarter and learned about different ways real estate agents use to appraise a property. Most of these are highly automatable: location, school district or not, size, carport, utilities, etc. However, in the actual sales process, many factor involving actual interactions with these agents would change the real estate appraisal outcomes significantly. Depending on who the buyer is, in China, for example, people care a lot about Feng Shui (a set of beliefs about a property’s location in relation to rivers, mountains, or other buildings) and the past history of the property (whether there was any death in the property), so the pricing could be influenced simply by the number of the apartment number or even whether who the previous owner of the property is.
However, websites like Zillow and Opendoor have made the process much more transparent. Without these easily manipulatable inputs, prices can be determined in a more straight-forward manner. Opendoor, especially, is also revolutionizing the industry and using advanced machine learning techniques to simplify the home selling and buying process.
Hi Jega,
Thank you for your post and information.
AI will be a tool to replace the agent in a transaction in different industries in the future. I think the most powerful function of AI in real estate is the big data for Intelligent Searching and Matching. AI can use data about the properties of the house, neighborhood or even transportation information to better fit buyers with sellers.
With AI in real estate industry, it will create a simpler searching process and create a better outcome.
MS&E 238-A student
Fung Tsz Sum
Kyle’s analogy comparing the benefit of Zillow to Kelly Blue Book is an excellent one. The Zestimate service has really changed the game when it comes to setting price expectations. In the early years there was some merit to the push back from realtors and brokers about the error in Zestimates for properties, but with the precipitous decline in the error rate and the ever expanding pool of data and transactions to pull from, this tool is only going to continue to empower consumers. Although the source you used above claims that it primarily benefits buyers, I think a benefit to sellers that was ignored was that it allows the seller to compare the listing advice they are receiving from their agent with a rough ballpark figure. This allows them to ensure that the agent isn’t excessively focused on making a sale and pressuring the seller to lower the price. It works as a check and balance for both the buyer and the seller in this way.