The Truth About AVMs: What Real Estate Agents Actually Need to Know (And Why Most Are Getting It Wrong)
Here is the problem nobody is saying out loud: your clients are already using Zestimate and Redfin estimates before they ever call you. They walk into that first conversation with a number in their head — a number that was generated by an algorithm that has never seen the inside of their kitchen, never noticed the cracked foundation in the back corner, and definitely never accounted for the fact that the comp down the street sold at a discount because of a messy divorce.
That gap between what the AVM says and what the market actually does? That is your job. And if you do not understand exactly how these models work, you cannot intelligently push back on them. You are just guessing — and your client knows it.
I am Emily Terrell, Tom Ferry coach and speaker, and I work with real estate agents across the country on systems, strategy, and AI integration. One of the most consistent gaps I see — even in experienced agents producing $20M or more — is a surface-level understanding of automated valuation models. They know AVMs exist. They know they are sometimes wrong. But they cannot explain why, and that costs them credibility in the listing conversation.
That changes today. Let’s actually solve for this.
What an Automated Valuation Model Actually Is
An automated valuation model is a computer program that uses statistical algorithms and machine learning to estimate the market value of a property. The AVM pulls publicly available data — recorded sales, property tax assessments, public MLS records where available, square footage, lot size, bedroom and bathroom counts, and sometimes permit records — and runs that data through a mathematical model to produce a value estimate.
The most well-known AVMs are Zillow’s Zestimate, Redfin’s Estimate, and the CoreLogic and First American models used by lenders in mortgage origination. There are also proprietary AVMs built into various MLS platforms and CRM tools that agents use internally.
What all of them have in common: they are built on pattern recognition. The model looks at what similar properties sold for in a given area over a given time period and extrapolates an estimated value for the subject property based on its characteristics. That is the engine. The differences between AVM providers come down to data inputs, algorithm design, local calibration, and how frequently the model is retrained on new data.
An AVM does not value your listing. It runs a pattern match against historical data. Understanding that distinction is the first step toward owning the conversation.
The Three Core Data Inputs That Drive AVM Accuracy
If you want to understand why an AVM is off on a specific property, start with these three inputs.
1. Comparable Sales Data
This is the backbone of every AVM. The model finds recent sales of properties with similar characteristics — typically filtered by bedroom count, bathroom count, square footage range, lot size, and geography — and uses those sales to anchor the estimate. The problem is that the model can only see the data it has been given. If the comps are thin (rural areas, unique properties, low-turnover neighborhoods) or if the most recent sales are not yet reflected in the public record, the estimate will drift.
In markets where properties sell in days and data entry into public records lags by weeks, AVMs are working off information that is already outdated by the time it is published.
2. Property Characteristics from Public Records
AVMs pull property details from county assessor records, which are updated on a schedule that varies by county. A major renovation that was completed two years ago and never pulled a permit? Invisible to the model. A garage conversion that added functional living space but was never recorded? Not counted. The model is working with the assessed description of the property, not the actual property.
This is where experienced agents with local knowledge will always have an advantage over an algorithm. You have walked that house. You know what the data does not show.
3. Geographic and Market Trend Inputs
Better AVMs incorporate local market trend data — absorption rates, median days on market, list-to-sale price ratios — to adjust the baseline value estimate upward or downward depending on current market conditions. But even these adjustments are lagging indicators. A market that shifted sharply in the last 60 days may not yet be reflected in a model that recalibrates quarterly.
The AVM Accuracy Gap: What the Data Actually Shows
Zillow publishes accuracy statistics for its Zestimate on a regular basis. As of recent public disclosures, the national median error rate for on-market homes is around 2 to 3 percent. For off-market homes, that error rate jumps to 6 to 7 percent — and that is the national median. In markets with limited comps, higher price points, or rapid price movement, errors of 10 to 20 percent are not unusual.
That sounds like a small number until you apply it to a $600,000 home. A 7 percent error margin means the estimate could be anywhere from $558,000 to $642,000. That is an $84,000 swing on a single property. If your client anchors their pricing expectations to the low end of that range, you have a problem before the conversation even starts.
Understanding the error rate — and being able to cite it specifically — gives you a concrete, credible way to reframe the AVM conversation with clients. You are not dismissing the number. You are contextualizing it.
| AVM Data Input | What It Measures | Key Limitation | Agent Advantage |
| Comparable Sales | Recent similar-property sales in the area | Lags public record entry by weeks or months | Real-time local market knowledge and off-market data |
| Property Characteristics | Assessed sq ft, beds, baths, lot size | Misses unpermitted work, renovations, condition | Physical inspection and accurate condition assessment |
| Market Trend Data | Absorption rate, DOM, price direction | Lags fast-moving markets by 30–90 days | Real-time transaction experience |
| Location Adjustments | School district, walkability, proximity factors | Cannot assess micro-location nuance or street-level desirability | Neighborhood expertise and buyer feedback |
| Time Adjustments | Seasonal and cyclical value shifts | Uses historical patterns, not real-time conditions | Current buyer pool intelligence |
Why AVMs Struggle With Certain Property Types
The pattern-recognition engine inside every AVM works best when there is a large volume of similar, recent, comparable sales. When those conditions are not met, accuracy deteriorates. Here is where agents consistently see the biggest AVM gaps.
High-End and Luxury Properties
At higher price points, the pool of true comparables gets thin. There may be only three or four sales per year in a given neighborhood at a specific price tier. The AVM is doing its best math on very limited data, and the result is an estimate with a wide confidence interval — whether or not the model discloses that uncertainty to the end user.
Unique or Non-Conforming Properties
A custom-built home with a commercial-grade kitchen, a detached studio, and a specific architectural style is not easily comparable to the three-bedroom ranch that sold around the corner. The AVM has to reach further in geography or time to find comps, which reduces the relevance of those comps to the actual subject property.
Rural and Low-Turnover Markets
In areas where sales volume is low, AVMs often default to comps that are far outside the ideal distance radius or far outside the ideal time window. In some rural markets, the model may be using comps from 18 to 24 months ago — a dataset that tells you almost nothing about today’s market.
Rapidly Shifting Markets
When market conditions change sharply — interest rate movement, local economic events, rapid inventory shifts — AVMs tend to lag. They are recalibrating on historical data while your local expertise is calibrating in real time.
The AVM is always looking in the rearview mirror. Your job is to tell the client what is actually in front of them.
How Agents Can Use AVM Knowledge Strategically
Understanding AVMs is not just about correcting them. It is about using that knowledge to control the narrative in the listing conversation, build credibility, and create a stronger case for your pricing recommendation.
Reference the Error Rate Directly
When a client brings up a Zestimate or Redfin estimate, do not avoid it or get defensive. Pull up Zillow’s published accuracy statistics directly. Show them the median error rate. Then show them how that error range applies to their specific property. You are not attacking Zillow. You are demonstrating that you understand the tool better than they do.
Identify the Comp Set the AVM Is Using
Most AVMs give you at least partial visibility into the comps they are using. Pull those same comps. If they include properties with different lot sizes, wildly different condition, or different bedroom configurations, walk your client through why those comparisons are flawed for their property. Your CMA becomes the credible alternative.
Use AVMs as a Starting Point, Not an Endpoint
The right framing for AVMs in your client conversations: they are a starting point generated by an algorithm with no local knowledge. Your CMA is the product of someone who has walked the property, knows the neighborhood, understands the buyer pool, and is tracking market conditions in real time. Both are valid starting points. Only one is complete.
Where AI Coaching Meets AVM Strategy
One of the things I work on with my coaching clients is building frameworks that make complex concepts easy to explain. AVM strategy is a perfect example. If you can break this down into a clear, simple explanation you can deliver in under three minutes at the listing table, you will win more listings than agents who are still getting flustered when a client says ‘but Zillow says…’
The agents I coach who do this best have a system for it. They have a rehearsed, conversational script. They have a visual they can pull up on their laptop. They have a comparison framework they walk clients through every single time. That is not a talent — that is a system. And systems are what actually create consistent results.
One of my coaching clients, Jason Sirois, scaled from $10M to $29M in volume after we rebuilt his listing process from the ground up. The AVM conversation was one of the first things we fixed. He went from getting derailed by the Zestimate to owning that conversation completely — and it showed up immediately in his listing conversion rate.
FAQ: Automated Valuation Models for Real Estate Agents
How accurate are Zillow Zestimates for real estate agents doing CMAs?
Zillow’s published median error rate for on-market properties is approximately 2 to 3 percent nationally, but off-market error rates can reach 6 to 7 percent or higher — and in markets with thin comps, errors of 10 to 20 percent are documented. For CMA purposes, AVMs are useful context but should never replace a professional comparative market analysis built on direct comparable sales and local market knowledge.
Why does the Zestimate sometimes differ significantly from the appraised value?
Appraisers conduct physical inspections, apply standardized adjustment methodologies for property-specific factors, and operate under regulatory guidelines. AVMs use publicly available data and algorithmic pattern-matching with no physical inspection. The gap is almost always explained by condition factors, unpermitted improvements, micro-location nuance, or timing differences in the data the AVM is using versus current market conditions.
Can I use AVM data as part of my listing presentation?
Yes, and strategically this can be very effective. Referencing the AVM’s error rate, showing the comps it is using, and demonstrating where your CMA methodology is more precise positions you as the expert and reframes the conversation away from algorithm-versus-agent into professional-context-versus-raw-data.
Do lenders use the same AVMs as Zillow and Redfin?
Lenders typically use institutional AVMs from providers like CoreLogic, First American, or proprietary models — not Zillow or Redfin. Lender AVMs are often calibrated differently and are used primarily for underwriting risk assessment, not consumer-facing price guidance. A Zestimate and a lender AVM on the same property can differ meaningfully.
How do I explain AVM limitations to a client without sounding defensive about my pricing?
Lead with curiosity, not defensiveness. Ask the client which tool they used, then say: ‘Great starting point. Let me show you how that estimate was built and where the data gaps are on this specific property.’ You are not attacking the AVM — you are adding the layer of professional context that the algorithm cannot provide. That positioning wins.
OTHER RESOURCES
External Authority Resources
NAR: Understanding Real Estate Valuation — https://www.nar.realtor/research-and-statistics
CoreLogic Home Price Insights — https://www.corelogic.com/intelligence/u-s-home-price-insights/
CFPB: Automated Valuation Models in Mortgage Lending — https://www.consumerfinance.gov/
Emily Terrell Resources
Coach Emily Terrell — Real Estate Coaching and AI Strategy — https://www.coachemilyterrell.com
Blog: AI Tools and Systems for Real Estate Agents — https://www.coachemilyterrell.com/blog