Zestimates, AVMs & Online Home Values: How Does a Website Know What a House Is Worth?

Aug 19

6 min read

Type a home address into a real estate website and, within seconds, you may see an estimated value for the property.

It might be a Zestimate. Another website may show a different estimated value. A real estate report might include an AVM or another type of valuation model. And sometimes those numbers are surprisingly close to one another.

Other times, they aren't.

So how does a website put a value on a house when no one from that website has actually walked through the front door?

The answer starts with something called an Automated Valuation Model, or AVM.

What Is an AVM?

An AVM is a computer-based model that uses available property and market data to estimate the value of real estate.

Depending on the model, that information may include public property records, previous sales, property details, market activity, location information and other available data. Some models also have access to listing information.

The model analyzes that information using its own method and produces an estimated property value.

That distinction matters.

The website isn't simply looking up a home's value in a database. The value is being estimated from the information available to the model.

And AVMs aren't simply internet tools that give homeowners an interesting number to look up. Automated valuation technology is also used within the broader real estate and mortgage industries for a variety of valuation-related purposes.

A Zestimate Is an AVM—But Not Every AVM Is a Zestimate

Zillow's Zestimate is probably the automated home-value estimate most people recognize, but it is one example of a much larger category of valuation technology.

Different AVMs may use different sources of information, different mathematical models and different ways of analyzing that information.

They may also look at different geographic areas when studying market data. One model may pull information from a fairly broad area, while another may stay closer to the property when enough useful data is available. A wider search area can give a model more sales to analyze, but those properties may not always be as similar to the home being valued.

That helps explain something that can be confusing: the same property can have different estimated values depending on where you look.

One model may have information another doesn't. The models may weigh property details differently. Their information may have been updated at different times. They may also use different methods to account for market activity and properties with limited nearby sales.

In other words, there isn't one universal AVM quietly powering every home-value website.

Even within the real estate industry, there are variations.

REALTORS Property Resource®, commonly known as RPR®, for example, distinguishes between its AVM and its RVM® (REALTORS Valuation Model®). RPR explains that its AVM uses publicly recorded sold data, while its RVM can incorporate on-market and off-market MLS listing data along with publicly recorded sold information when enough data and other criteria are met.

Both are model-generated estimates, but the information available to the models isn't necessarily the same.

The Data Behind the Number Matters

An AVM can analyze an enormous amount of information much faster than a person could reasonably process it. That's part of its value.

But a model still has to work with the information available to it—and that information isn't always perfect.

Property tax and public records are common sources of information for valuation models. Those records can contain useful facts about a property, but they may also be incomplete or out of date. An addition may not appear as expected. Square footage may differ from another source. Property details may have changed since the record was last updated.

That matters because the model cannot walk through the front door and see what's really there.

There can also be challenges when a property doesn't fit neatly with the other homes around it.

Many automated home-value models are built primarily around residential property data. A home with acreage, several outbuildings or unusual improvements can be harder for a model to evaluate when those features aren't fully reflected in the data it uses.

Age can create another challenge. A newer home and a home built decades ago may look similar on paper in some respects—perhaps similar square footage, bedroom count or location—but that doesn't make the properties the same. Construction, updates, condition and features may be very different. How well a particular AVM recognizes those differences depend on the information and methods used by that model.

Now imagine two very different properties.

One is located in a neighborhood where many relatively similar homes have sold recently and where detailed property information is readily available.

The other is a rural East Tennessee property with substantial acreage, an older home, several outbuildings, improvements made over many years and very few truly similar recent sales nearby.

Those are very different valuation puzzles.

The second property doesn't make an AVM useless. But it may give the model a more difficult problem to solve.

Modern valuation models have increasingly sophisticated ways of working with large amounts of information, but the basic principle remains important:

The estimate depends, in part, on what the model knows about the property and its market.

Look Beyond the Headline Number

It's easy to focus on the biggest number on the screen.

But sometimes the information surrounding an AVM estimate tells an equally important part of the story.

Depending on the source, an automated valuation may be accompanied by an estimated range, confidence measure or other information designed to provide context for the estimate.

A wide estimated range or lower confidence level doesn't necessarily mean the model has failed. It may simply be telling us that there is more uncertainty surrounding that particular estimate.

That is a very different message from declaring, "This house is worth exactly $X."

When an estimated value is available, it can be useful to look beyond the number itself and understand what else the model is telling you.

Where a CMA Can Add More Context

An AVM and a comparative market analysis, or CMA, aren't trying to do exactly the same job.

A real estate agent preparing a CMA can look more closely at the property and choose comparable sales based on how they relate to that particular home. The analysis can consider differences that may be difficult to understand from public data alone, such as condition, updates, location within an area and other property features.

The agent can also look at the AVM as another piece of information rather than automatically accepting or rejecting it.

That combination can help narrow the likely market range and develop a pricing strategy based on the property's characteristics, current competition and the seller's goals.

The AVM hasn't lost its value. It has become one piece of a larger picture.

And What About an Appraisal?

An appraisal is different from both an AVM and a CMA.

An appraisal is an opinion of value developed by a licensed or certified appraiser for a specific purpose. The appraiser researches the property and relevant market information and follows professional appraisal standards in developing that opinion.

Appraisals are often part of financed real estate transactions, but they can be prepared for other purposes as well.

For our discussion, the important point is simple: an appraisal, a CMA and an AVM are different valuation tools. They may use some of the same market information, but they are not interchangeable.

An Estimate Is a Tool, Not the Entire Toolbox

AVMs can provide useful information. They can analyze large amounts of market and property data quickly, identify patterns and provide a consistent data-driven estimate.

They can also be one piece of a larger valuation picture.

That's why the most useful question may not be:

“Which number should I believe?”

A better place to start may be:

“Where did this number come from, and what information went into creating it?”

Or perhaps an even simpler question:

“What does this model actually know about this property?”

Once you understand that, an online home value becomes much more useful—not because it provides a magic answer, but because you understand what the estimate actually represents.

Fireside Reflection

Technology has made an extraordinary amount of real estate information available with a few taps on a screen. That's a good thing. But having more information doesn't always mean we understand the information we're looking at.

Sometimes the most valuable thing we can do with a number isn't immediately accept it or dismiss it. It's simply ask how it got there.