The Five Numbers I Look At Before I Tell You What We Should List Your House For

The Five Numbers I Look At Before I Tell You What We Should List Your House For

How do you price a house? By building five separate estimates: comps adjusted like an appraiser, the MLS tax-roll AVM, an AI reasoning from those same comps, the public Zestimate, and a read on recent market motion. The disagreement between them is the useful part.

If you ask most people how a house gets priced, they will say "you run the comps." That is where it starts. It is not where it ends and anyone who tells you it is has never had to defend a number to a seller, a buyer's agent and an appraiser in the same thirty days.

Here is what actually happens on my end before I sit down at your kitchen table with a number. I build five estimates from five different directions. Each one is blind in a different place. When they agree, I have a narrow target. When they scatter, that scatter tells me something specific about your house and I go find out what it is.

This is the long version and it is deliberately transparent. I would rather you understand where the number comes from than take it on faith.

Why one number is the wrong answer

Value is not a fact about your house. It is a prediction about a future buyer and predictions come with error bars whether anyone prints them or not.

Every method below is an attempt to answer the same question, which is what a ready and willing buyer is likely to pay for this house, in this condition, in the next sixty days. The methods use different data and different logic, so they land in different places. That is not a flaw in the process. That is the process working.

What I am building is not a price. It is a defensible range, plus a view on where inside that range your house sits and why.

Input one: the comps, adjusted the way an appraiser does it

This is the foundation and it is the one most people think they understand.

The part that gets skipped is the adjusting. A raw list of four nearby sales is not a comp analysis. An appraiser takes each comparable sale and mathematically converts it into your house and the direction of that math trips up almost everyone.

Adjustments are applied to the comparable, not to your house. If the comp has something yours does not, you subtract from the comp. If the comp is missing something yours has, you add to the comp. The comp with the pool gets marked down to make it look like your pool-less house. It feels backward the first time you see it and it is the entire grammar of the method.

The categories an appraiser works through are property features, location, market conditions and terms of the transaction. That last one matters more than sellers expect. A sale where the seller paid five thousand in buyer closing costs did not really sell for its recorded price and the adjustment has to account for it.

Where do the dollar amounts come from? The most common method is paired sales analysis, which isolates one feature by comparing two otherwise similar sales to see what the market actually paid for it. Not what it cost to build. What a buyer paid. Those are different numbers and the gap between them is where most seller disappointment lives.

What this gives me: a defensible range, an average, and a median, built from real closed transactions in your market.

What it cannot do: tell me about a buyer who has not shown up yet. Every closed comp is a decision somebody made weeks or months ago under conditions that may no longer exist.

Input two: the NTREIS AVM in our tax roll data and what is actually inside it

I look at the automated valuation that rides along with the tax roll data in our MLS system.

The tax and public-record product NTREIS provides to agents is Realist, which is a CoreLogic product (CoreLogic now operates as Cotality). The valuation inside it is RealAVM and it is refreshed weekly. So when I quote you a number off the tax roll screen, that is what you are looking at.

Here is what its own documentation says it does.

It is not one model. RealAVM uses a patented approach that applies multiple methodologies to each property and then reconciles them into a single value. Commercial AVMs generally blend a hedonic model, which prices a house as the sum of its measurable attributes, with repeat-transaction indices, which track how prices for the same properties have moved over time. Each component model produces its own answer and gets weighted.

It emulates an appraiser, at a scale no appraiser can match. The appraiser-emulation component typically analyzes roughly 90 comparable sales for a given property, and in some cases as many as 300. A human appraiser uses three to six. That breadth is a genuine advantage and it is worth respecting.

It publishes its own uncertainty, which almost nobody reads. Every RealAVM value comes with a Confidence Score from 60 to 100 and a Forecast Standard Deviation. CoreLogic's own published example is the clearest thing in the whole document: a value of $100,000 with a Confidence Score of 80 and a Forecast Standard Deviation of 82 means an 80 percent probability that the true value is $100,000 plus or minus 18 percent.

Read that again, because it reframes the entire tool. At a solid confidence score, the model is telling you the answer sits somewhere in a band 36 percent wide, and there is still a one in five chance it falls outside even that. On a $467,000 house, plus or minus 18 percent is a $168,000 window. The AVM is not claiming precision. We are the ones who read precision into it because it prints a specific dollar figure.

The data behind it is enormous. CoreLogic maintains records covering roughly 99.9 percent of United States property records, more than 500 million historical transactions, over 3,100 counties and 145 million parcels. In Texas, which is a non-disclosure state, the MLS-sourced portion of that matters enormously, because sale prices are not public record here.

And here is what it cannot see. The documentation is candid that the less data a model has, the less reliable the result. AVMs in general assume average condition because no model has been inside your house. They lag actual market turns by roughly three to six months because they are trained on closed data. They are weakest on new construction, unusual properties and areas with thin sales volume.

So an AVM cannot tell the difference between your neighbor's original 2003 kitchen and your 2023 kitchen. It cannot smell the dog. It does not know the master shower has been leaking into the subfloor. It assumes you are average and the entire business of preparing a house to sell is the business of not being average.

What this gives me: a large-sample, unemotional second opinion with an explicit uncertainty range.

What it cannot do: see condition, updates, or anything that happened in the last few weeks.

Input three: I hand the comps to an AI and ask it to reason like an appraiser

I want to be straightforward about this one because it is the input people are most curious about and most suspicious of.

I take my selected comparable sales, feed them to an AI and ask it to work the problem as an appraiser would. Identify the adjustments. Assign the dollar amounts. Reconcile to a value. Show the reasoning.

Why I do it: it is a check on me. The single biggest risk in a comp analysis is not arithmetic. It is selection. I choose which four sales to use and by the time I choose them I usually have an instinct about what the house is worth. It is very easy to pick the comps that agree with that instinct without ever noticing you did it. That is called anchoring and every human in this business does it.

Handing the same comps to something that has no relationship with you, no listing to win and no opinion formed in your driveway produces a number I did not produce. When that number matches mine, my confidence goes up. When it does not, I go back and find out which of us is wrong and it is not always the machine.

It is also good at the tedious part. It will surface an adjustment I glossed over, note that two of my comps had seller-paid closing costs or point out that I leaned on a sale that is five months old in a market that has moved.

Now the honest limitations because they are real.

It has never been in the house. Neither has the AVM but at least the AVM is not writing paragraphs of confident prose about it.

It can only see the comps I give it so if my selection is bad, its answer is bad. Garbage in, articulate garbage out.

And it does not replace anything. It is one more opinion in the room not the deciding vote. No client's private information goes into it and it never sees anything beyond property characteristics and public sale data.

I am not going to pretend I do not use this. You are going to get a better price from an agent who uses every available check on her own blind spots than from one who does not.

What this gives me: an independent reasoning pass over the same comps that catches my anchoring.

What it cannot do: anything I have not shown it and it will never tell me it is unsure.

Input four: the Zillow number which is not about accuracy

Then I take a screenshot of the Zestimate. Not because I think it is right but because your buyer is going to look at it.

Zillow publishes its own accuracy figures and they deserve credit for that. Their stated median error is about 1.9 percent for homes currently listed and about 7.0 percent for off-market homes. That off-market figure is the one that matters for you because your house is off-market on the day I price it. Seven percent on a mid-$400s house is roughly $32,000 in either direction, at the median. Half of all homes miss by more than that.

The Zestimate draws on public records, tax assessments, recent sales and homeowner-submitted information. Owners can claim a home and correct its facts which does update the model's inputs. What no one can do is push it in a direction the market is not going.

So I do not treat this number as a valuation. I treat it as the number in your buyer's head. It is public opinion about your house, it is free, and it is going to be quoted at you.

If the Zestimate is well above where the evidence lands, I want to know that before we go live because I am going to be answering for it. If it is well below, that is worth understanding too and often it traces to a missing bedroom in the tax record or square footage that never got updated after an addition. Sometimes it is fixable.

What this gives me: a read on the number the public is anchored to.

What it cannot do: serve as evidence of anything, including in front of an appraiser.

Input five: my read on a market that is currently in motion

This is the input nobody can automate and it is the one I am actually paid for.

Four things go into it.

What the last few weeks have done. Not the last six months. The last few weeks. Showing counts, how fast the good listings are going, what is sitting, whether buyers are asking for closing cost help again, whether the calls sound eager or cautious. Around here, late August and late January are different markets, and the comps do not know that yet.

Which of your upgrades a buyer will actually pay for. This is where I earn the conversation because it is frequently unwelcome. The kitchen usually returns real money. The pool is a coin flip and depends entirely on the buyer. The $30,000 you spent on the outdoor kitchen returns some of itself and never all of it. Personalized finishes narrow your buyer pool which is a price consideration whether or not it feels like one.

What you are going to fix and what you have decided not to fix. Both are pricing decisions. If the roof is at the end of its life and you are not replacing it, we are not pricing as though it has twenty years left because the buyer's inspector will find it in week two and we will be having the conversation anyway, only then it will be a negotiation instead of a strategy. Deciding not to fix something is completely legitimate. It just has to be in the number.

And what the appraiser is likely to do. Because if we push past what the comps will support, the deal does not die at offer. It dies at appraisal, weeks later, after you have already packed. I wrote about that whole scenario in detail and I would rather design around it than survive it.

What this gives me: the only input that knows what today looks like.

What it cannot do: be proven in advance. This is judgment and I will always tell you when I am using it.

What this looks like with numbers on the table

Let me make it concrete. This is a hypothetical, built at a realistic Northeast Dallas price point. No client's file.

The subject: 2,450 square feet, four bedrooms, two and a half baths, built 2003, two-car garage, interior lot, kitchen updated in 2023, no pool.

Four comparable sales, adjusted:

Notice Comp 3. It sold for the most money and it adjusted to the least. That is the whole reason you adjust. Raw sale prices in that column run from $441,000 to $495,000, a spread of $54,000. Adjusted, they run from $456,255 to $473,130, a spread of under $17,000. The adjustments are what turn noise into a signal.

Comp range: $456,255 to $473,130 Average: $462,034 Median: $459,375

Now the other four inputs:

Five methods. A low of $458,000 and a high of $481,000. A spread of $23,000, which is 5 percent of the value of the house.

And the AVM, remember, is quietly telling you its own honest band on that $467,000 is something like $383,000 to $551,000.

What I do with the disagreement

The disagreement is the assignment. Here is how I read that table.

The three evidence-based methods clustered. Comps median, comps average and the AI pass all landed between $458,000 and $462,034. Three different routes to nearly the same place is a real signal and it says the honest market value of this house sits in the high $450s to low $460s.

The AVM ran slightly high and there is probably a reason. $467,000 against a comp median of $459,375 is about $7,600 of daylight. The likely explanation is the 2023 kitchen, which the model cannot see but which is embedded in the sale prices of updated comps it swept in. I would not chase it, but I note it, because it points the same direction my own eyes do.

The Zestimate is the outlier and I treat it as a marketing problem. $481,000 sits nearly $22,000 above the comp median. It is not evidence. But every buyer's agent will see it and every seller will have already seen it, which means if we list at $465,000 the public number says we are a bargain and if we list at $485,000 the public number is the only thing supporting us. Those are two completely different conversations and I would rather have the first one.

So I would list in the $465,000 range, with the evidence supporting the mid $460s, the AVM agreeing, the Zestimate above us doing quiet work in our favor and room to hold firm in negotiation without falling below what an appraiser can support. That last clause is the one that protects the closing.

If instead the five numbers had scattered across $80,000, I would stop and go find out why. In my experience the answer is nearly always one of four things: the house has a condition problem the models cannot see, the neighborhood is thin on recent sales, there was a very recent turn in the market that the comps have not absorbed or the property is unusual enough that the models are outside their competence. Each of those has a different pricing response.

Comps are history. The market is in motion.

Here is the sentence I want you to leave with.

Every comparable sale is a decision somebody made in the past, under conditions that no longer exist. A sale that closed in June went under contract in May, which means the price was agreed by two people reading a market that was four months ago. The rate environment has changed since. The inventory has changed since. The weather has changed since and in this market that is not a joke.

Comps are the best available evidence and they are also, unavoidably, a record of what is already over. They tell you where the market has been. They do not tell you where it is standing right now and they certainly do not tell you where it will be the week your buyer writes the offer.

Which is why the number is not the end of the process. It is the beginning of one.

We price it, we watch what the first ten days tell us and we adapt. Showing traffic with no offers means something different than no showing traffic at all. One says the price is close and something else is wrong. The other says the price is wrong. Those get different responses and the response has to come in week two, not week seven, because a listing's best pricing information arrives early and its leverage decays the entire time it sits.

Pricing is not an act. It is a position you keep testing.

What this means if you are getting ready to list

A few practical things.

Ask any agent to show you the adjustments, not the comps. Anyone can print four nearby sales. The adjustments are the work and the willingness to walk you through them tells you what you need to know.

Look up your own Zestimate before you list, and check the home facts behind it. If the bedroom count or square footage is wrong in the record, correcting it is free and it is one of the only levers you have on that number.

Understand that the range is the honest answer. If somebody hands you a single number to the dollar with no discussion of uncertainty, they are selling you confidence, not analysis. The AVM built on 145 million parcels publishes error bars. Your agent should be at least as humble as the algorithm.

And be careful with the listing appointment where the number is the highest. The agent who tells you the biggest number is not the agent who gets you the most money. They are the agent most willing to tell you what you want to hear in the room, which is a preview of how the next ninety days are going to go.

Frequently asked questions

What is the AVM in the NTREIS tax data and how accurate is it? The tax and public-record product in NTREIS is Realist, a CoreLogic product and its valuation is RealAVM, refreshed weekly. It applies multiple modeling methodologies and reconciles them, including an appraiser-emulation component that analyzes roughly 90 comparable sales and sometimes as many as 300. Each value carries a Confidence Score and a Forecast Standard Deviation. CoreLogic's published example is a value with an 80 Confidence Score and a Forecast Standard Deviation of 82, meaning an 80 percent probability the true value is within plus or minus 18 percent. It is a useful second opinion, and it cannot see condition or updates.

How accurate is the Zestimate on a home that is not listed? Zillow publishes a median error of about 1.9 percent for on-market homes and about 7.0 percent for off-market homes. On a $450,000 house, seven percent is roughly $32,000, and by definition half of all homes miss by more than the median. It is best treated as a read on public perception rather than as evidence of value.

Why do appraisers adjust the comparable sale instead of the subject property? Because the goal is to convert each comparable into the subject. If the comp has a feature the subject lacks, the comp's price is reduced. If the comp lacks a feature the subject has, the comp's price is increased. The dollar amounts typically come from paired sales analysis, which isolates what buyers actually paid for a given feature rather than what it cost to install.

If you found this useful, read these next

• The Conversation I Have With Every Seller Before We Ever Talk About Price is what happens before any of the above, and it is the reason the number ends up defensible.
• When the Appraisal Comes in Low: What Buyers and Sellers in North Texas Should Do Next is what the entire process above is designed to prevent.

If you are thinking about selling in Sachse, Wylie, Murphy, Rowlett, or anywhere in Northeast Dallas, I will build all five of these for your house and walk you through every one of them, including the ones that disagree with me. You should know where your number comes from before you commit ninety days of your life to it.

Jeanie Marten Real Estate is a brokerage, not an appraisal firm, and nothing here is an appraisal or a guarantee of value. The example above is hypothetical and illustrative. Third-party valuation tools and their published accuracy figures change over time, so confirm current figures with the providers.

Visit MartenTeam.com or book a consultation.

Jeanie Marten Real Estate 972-414-0719 | MartenTeam.com

Sources referenced

• CoreLogic RealAVM product documentation
• CoreLogic Realist for MLS
• Zillow: What is a Zestimate and how accurate is it
• McKissock: making adjustments in the sales comparison approach