Why Real Estate Agents Who Don’t Understand AVMs Keep Losing Listing Appointments to an Algorithm
I want to tell you about a moment that happens thousands of times a day in listing appointments across the country. An agent sits down with a seller, presents a carefully prepared CMA recommending a list price of $485,000, and the seller pulls out their phone and says: ‘But Zillow says it’s worth $512,000.’
And then — silence. Or worse, stammering. Or worse still, the agent starts walking back their own analysis to stay in the room.
That moment does not have to go that way. But it will keep going that way for any agent who does not understand, in specific and credible terms, exactly what automated valuation models are, how they generate their estimates, and where they consistently fail.
I am Emily Terrell, real estate coach and speaker with Tom Ferry, and helping agents build systems that work in real-world conversations is exactly what I do. The AVM conversation is not optional anymore. Every single one of your sellers has already looked at Zillow. Your job is to know more about that number than they do.
The Architecture Behind the Estimate
To win the AVM conversation, you need to understand what actually goes into building one of these estimates. Let’s break it down into the actual how.
Automated valuation models are built on three primary methodologies, and most major platforms use some combination of all three.
Hedonic Pricing Models
This is the foundational approach. A hedonic model treats a property as a bundle of individual attributes — square footage, bedroom count, bathroom count, lot size, garage spaces, year built, school district — and assigns a dollar value to each attribute based on how those attributes have correlated with sale prices in the market. The model then adds up the attribute values to produce an estimate.
Hedonic models are good at capturing structural differences between properties but are weak on condition, quality of finish, and anything that is not in the public data record.
Repeat Sales Analysis
Some AVMs incorporate repeat sales data, tracking how the value of specific properties has changed between sale events over time. The Case-Shiller Home Price Index uses this methodology at a macro level. For individual property valuations, repeat sales analysis can help anchor the model to a property’s actual price history, but it requires a property to have sold at least twice, and it is sensitive to the time gap between sales.
Machine Learning and Neural Networks
The most sophisticated modern AVMs layer machine learning on top of traditional statistical methods. These models ingest far larger datasets — including listing photos, walkability scores, proximity to amenities, local economic indicators, and sometimes social media sentiment data — and use neural networks to identify non-linear patterns in the data. Zillow has invested heavily in this approach, including the use of computer vision to analyze listing photos.
Machine learning models can capture relationships that traditional statistical models miss, but they are as dependent on data quality and quantity as any other approach. Garbage in, garbage out — just with more sophisticated math.
The sophistication of the algorithm does not eliminate the fundamental problem: the AVM is working with public data that is incomplete, lagging, and incapable of accounting for what is physically in front of you.
What AVMs Can See vs. What They Cannot
This is the framework I use with coaching clients when we are building their AVM conversation script. Understanding this split is what transforms the AVM from a client objection into a professional teaching moment.
| What the AVM Can See | What the AVM Cannot See |
| Recorded sale price and date of comparable sales | Seller concessions or credits not captured in the recorded price |
| Public record square footage and room count | Unpermitted additions, renovations, or conversions |
| Assessed property tax information | Interior condition, finish quality, deferred maintenance |
| Geographic proximity to recorded comps | Micro-location desirability (views, street noise, privacy) |
| Historical price trend for the area | Real-time market shift in the last 30 to 90 days |
| School district and basic neighborhood data | HOA restrictions, pending assessments, or special conditions |
| Listed amenities from public MLS fields | Staging quality, curb appeal, or buyer emotional response |
Every row in that table is a conversation you can have with a seller. Every one of those gaps is a reason why your professional judgment adds value that the algorithm simply cannot replicate.
The Confidence Interval Problem Nobody Talks About
Here is something that most consumer-facing AVMs do not make easy to find: confidence intervals. When Zillow produces a Zestimate, it is not just producing one number — it is producing a range. That range is the model’s acknowledgment that it is not certain. The headline number is the midpoint of that range.
For properties with good comp support in active markets, that range might be relatively tight. For properties with thin comps, unique characteristics, or in markets with recent volatility, that range can be very wide. A $500,000 Zestimate with a published range of $440,000 to $560,000 is telling you that the algorithm itself is not confident in that headline number.
Pulling up the confidence interval for your client’s property on Zillow — and then explaining what it means — is one of the fastest ways to shift the conversation from ‘the algorithm says X’ to ‘here is what the algorithm actually knows about your property, and here is where my analysis fills in the gaps.’
How Different AVM Providers Diverge on the Same Property
One of the most effective demonstrations you can do in a listing appointment is pull the AVM estimate for the same property from three different sources simultaneously: Zillow, Redfin, and a third platform if available. In most markets, those three numbers will be meaningfully different from each other. Sometimes the spread between them is $20,000 or $30,000 or more.
That divergence is not an anomaly. It is the expected outcome when different algorithms with different data inputs, different comp selection methods, and different calibration approaches are applied to the same property. The question then becomes: if three different algorithms produce three different numbers, which one is right? And the answer — which you then deliver — is that none of them is as right as a professional comparative market analysis built by someone with actual knowledge of this property and this market.
Redfin’s Approach Versus Zillow’s
Redfin’s estimate tends to rely more heavily on MLS-sourced data, which it has direct access to through its brokerage operations. This can make its estimates more responsive to recent market activity in active MLS markets. Zillow uses a broader data set and a more complex model, which can improve accuracy in some contexts but also introduces more noise in others. Neither is consistently more accurate — they are differently calibrated tools.
Lender AVMs vs. Consumer AVMs
Lenders do not use Zillow. They use institutional AVM providers — CoreLogic, First American, Veros, and similar — that are calibrated specifically for underwriting risk, not for consumer market guidance. Lender AVMs tend to be more conservative and may weight certain data differently than consumer platforms. An agent who understands this distinction can explain to a seller why the appraisal does not match their Zestimate — before the appraisal comes in and creates a problem.
Building Your AVM Conversation Into a Repeatable System
This is where the strategy meets the system. Knowing all of this is valuable. But knowing it and being able to deploy it consistently in a high-stakes listing conversation are two different things.
Here is the framework I teach:
- Acknowledge the AVM directly. Do not wait for the client to bring it up. Pull it up yourself early in the listing conversation. This signals confidence, not defensiveness.
- Show them the accuracy data. Pull Zillow’s published error rates. Use the specific number for their market if available.
- Run the comparison. Show the comp set the AVM is using. Walk them through the quality of those comps relative to their property.
- Identify the gap. Point to two or three specific factors about their property that the AVM cannot account for: the renovated kitchen, the larger-than-average lot, the view, the condition premium or discount.
- Present your CMA as the professional layer. Your analysis is not replacing the AVM — it is completing the picture the AVM cannot finish on its own.
That five-step conversation, delivered consistently, converts listing appointments at a much higher rate because it positions you as the authority on valuation methodology — not just an agent with an opinion about price.
Agents who understand AVMs do not argue with the algorithm. They teach the client why professional judgment is the layer the algorithm was never designed to replace.
Where AI Tools Fit Into the AVM Conversation
As the top AI coach for real estate agents in the Tom Ferry organization, I get asked constantly about how AI tools change the AVM dynamic. Here is the honest answer: AI tools like ChatGPT, Grok, and Claude are not valuation tools. They are research and communication tools.
What that means practically: you can use AI to build a client-facing explanation of how AVMs work, calibrated to your market and your client’s specific situation. You can use AI to draft the AVM section of your listing presentation. You can use AI to create a one-page comparison document that shows Zillow, Redfin, and your CMA side by side with an explanation of the methodology differences.
What you cannot use AI to do is produce a more accurate valuation than a properly built CMA anchored in real local market data. AI tools are force multipliers for how you communicate — not replacements for what you know.
FAQ: How Automated Valuation Models Affect Real Estate Transactions
Why is the Zestimate different from what my home actually sold for?
The Zestimate is generated from publicly available data with no physical inspection of the property. Factors including condition, finish quality, renovations without permits, micro-location advantages or disadvantages, and real-time market conditions are often not fully captured. The gap between the Zestimate and the final sale price reflects these data limitations.
How do I use AVM data in my listing presentation without sounding like I am attacking Zillow?
Lead with curiosity and education rather than opposition. Present the AVM data yourself before the client does, explain how the model works, and use the data’s own limitations to contextualize why your CMA methodology produces a more complete picture. You are not dismissing Zillow — you are demonstrating a deeper level of professional expertise.
What is the biggest reason AVM estimates are wrong in my market?
The most common driver of local AVM inaccuracy is comp quality — specifically, the relevance of the sales the model is using. Thin markets, rapid price movement, or properties with characteristics that are hard to find true comparables for all amplify AVM error rates. A professional CMA that selects and manually adjusts the most relevant comparable sales will consistently outperform the algorithm on these property types.
Can an AVM account for the difference between a renovated and un-renovated property of the same size?
Only partially. AVMs can apply adjustments for certain features if that data is in the public record, but condition and finish quality are almost never captured accurately in public assessor data. The algorithm may know that a kitchen was remodeled if a permit was pulled, but it cannot assess the quality of that remodel or the premium buyers in your specific market are willing to pay for it.
Do sellers trust AVMs more than they trust agents?
Many sellers arrive at the listing appointment with a strong prior belief in their AVM-generated value, particularly if that number is higher than the agent’s recommendation. The way to shift this is not by attacking the tool but by demonstrating deeper expertise about how the tool works. Agents who can explain AVM methodology in specific terms consistently report stronger listing conversions.
OTHER RESOURCES
External Authority Resources
Zillow Research — Zestimate Methodology — https://www.zillow.com/research/
CoreLogic Insights — Real Estate Data and Analytics — https://www.corelogic.com/intelligence/
Emily Terrell Resources
Coach Emily Terrell — Coaching for Real Estate Agents — https://www.coachemilyterrell.com
Coach Emily Terrell — Blog — https://www.coachemilyterrell.com/blog