What’s the Difference Between a Human and Machine’s Home Valuation
A common question that comes up when interacting with family, friends, and consumers who are coming to Zillow to check on the value of their home is: why is my Zestimate different from the valuation produced by an appraiser? It’s a really good question and like most good questions the answer is it depends. It depends upon what kind of appraisal was conducted. It depends on whether you claimed your home on Zillow and checked to make sure we have the right information. Perhaps it’s easier if we back up and review what appraisers do and how that compares to the Zestimate.
Zestimate vs. Appraisers: Estimation process
Both human appraisers and automated valuation systems (AVMs) have two major tasks to complete when developing a home valuation. The first is to gather information about the home. Collecting correct data is a critical step for both the Zestimate and appraiser. If an estimate is based on bad data, it will result in a less accurate estimate, whether it was created by a human or machine. Both machines and humans start at the same place, looking at public records to collect basic facts. To create the Zestimate, we augment the home facts supplied by public records with data from brokers and multiple listing services, including images of the exterior and interior of the home, if available. Meanwhile, human appraisers will make a choice about whether to visit the home in person. There are a few different types of appraisal: desktop appraisals, hybrid appraisals, and full appraisals. If an appraiser does travel to the home, the visit can range from a cursory “drive by” inspection of the property features visible from the street to an in-depth visit in which the appraiser walks the property and has access to the inside of the home.
Once the data has been collected, we can move on to the process of developing a valuation for the home. For both approaches, the process starts with looking up nearby homes that recently sold or are currently listed for sale. Typically an appraiser picks three or so homes from this list that he deems roughly similar to the home being evaluated/valued. The appraiser makes this judgement based upon his experience and intuition about what matters to homebuyers in the area. This could include things such as living area, nearby parks, the size of the garage, or views (if any). Once he has the list of comparable homes (“comps”), the appraiser makes a second set of judgments to adjust the sale prices of these homes to account for any important differences. For example, if one of the comparable homes is larger, the appraiser will make an educated guess at what that extra space is worth and subtract that amount off that home’s sale price to bring it more in-line with the target home. The appraiser averages the adjusted sale values of each of the comparable homes to arrive at the valuation for the target home. To sum it up, human appraisers compare similar homes to the target home, and use their local expertise to adjust their sale prices based on differences between the comps and the target home to arrive at the appraised value.
As with a human appraisal, the raw ingredients for a Zestimate are comparable sales as well. But instead of picking a few comps and trying to reconcile the differences between the target home and comparable homes, the Zestimate algorithm leverages its computing power advantage. Zestimate valuations are an average across hundreds of comparable sales. Like an appraiser, the Zestimate more heavily weighs homes that it judges as more similar to the target home. The weights assigned to each comparable sale are based on which home features minimize the difference between the algorithm’s value estimates of recently sold homes and the actual sale prices. The Zestimate is trying to build an optimal model of which facts about a home are important to buyers and thus can predict sale prices.
The recently launched Zestimate Deep Dive feature lets users see which homes the Zestimate algorithm thought were most important comps for estimating the value of a given home. We only show the top ten or so homes, but there are usually hundreds of comparable homes the Zestimate uses to make its valuation. The strength of the Zestimate is that the differences between the subject home and the comparables will cancel out when averaged across many homes.
But why are they different?
Valuation methods that are this different will never arrive at exactly the same valuation for a home. In fact, it’s not uncommon for two human appraisers to be five percent different in their valuations of a home. Similarly, two different approaches to machine valuation will also not arrive at the exact same valuation. But, if two different valuation approaches are in close agreement, chances are the price the home will sell for is pretty close to that too.
We created the “Inside the Zestimate” feature to help people understand the range of reasonable valuations for a home by showing the estimates produced by a number of different algorithmic approaches. Again, some differences are expected. For example, a home that sold for $250,000 could be valued by an appraiser at $257,000 and by the Zestimate at $243,000. These valuations are less than 5 percent apart and each is within 3 percent of the true value. On a home worth $750,000, the same would hold true with a Zestimate of $771,000 and an appraisal of $729,000.
We advise potential home sellers to work with an agent with a lot of local knowledge because algorithms are helpful for coming up with a range of fair valuations for their home, but an agent is an expert on what kind of pricing and marketing strategies can maximize the final sale price ofa home. On more expensive homes, this can mean tens of thousands of dollars. No two homes are exactly alike, so valuations conducted by different people or by different algorithmic approaches will never match exactly, but both are incredibly helpful in figuring out the approximate value of a home.