Nest’s Sweat the Details podcast continues. We just released our 67th podcast, and we’re excited to bring you interesting conversations that touch on real estate — and how we look beyond real estate for inspiration.
Listen to the podcast here, and subscribe to the podcast here.
Summary:
Elephant and the blind mice; Divided housing market; Algorithmically noting whether a house has been smoke in or lived in by cats; AVMs and their algorithms; Zestimates, assessments and Zillow’s accuracy Zillow becoming more right; Jonathan tracked the effect of professional photos on a Zestimate; ChatGPT hiring humans; Take professional photos early; Value of local knowledge about housing stock and inventory.
We hope you’ll join us for the next episode of Sweat the Details. View the full transcript below.
Jim:
It’s Jim Duncan with Nest Realty. Sweat the Details. We talked about Zillow, Zestimates, pricing, strategies, algorithms, artificial intelligence, touched on ChatGPT. We talked about how to use the data and how to allocate it and how to use it in your business and your practice.
Jonathan, Keith, how are y’all doing?
Jonathan:
Great. It’s a beautiful spring day, the weather’s amazing, the leaves are popping on the trees.
Keith:
And the pollen is on the car.
Jonathan:
The pollen is on the car.
Jim:
I have had to wash my car twice now because the eyesight won’t work on my Subi, because the pollen is clogging the sensors, which is really fun, because I hate washing my car.
Jonathan:
Do you wash your car or does somebody, like a machine wash your car?
Jim:
I go into the $7 car wash, and just get the pollen kicked off.
Jonathan:
Because as we lead into the conversation about machines today, it is amazing to see the new car washes that are coming into the…
Jim:
The Tunnel Wash.
Jonathan:
Charlottesville, small town USA, that have been everywhere. And now we are coming here and just seeing the lines of people that are there to have these machines wash their cars.
Jim:
Machines excite people. What are you going to do?
Jonathan:
Machines excite people. And so we’re here today to talk a little bit about machines, and the hot term recently across all the, probably in every part of our lives, has been AI. And AI’s been in real estate really, to a certain extent for, gosh, 10, 15, probably 15 years.
Jim:
Yeah.
Jonathan:
15 years. But now with ChatGPT and some other services that are coming out, it’s coming to the forefront again. So we want to talk today about AI and pricing, and humans and pricing, and how we price homes and how computers can price homes.
Jim:
Who do you trust?
Jonathan:
Who do you trust? That’s a good question.
Jim:
Yeah. So I think that just a quick look. Nationally, I was doing the research for this podcast, the thing that kept coming to my mind was the parable about the elephant and the blind mice, because I know our market. I know the Schultz Admiral Central Virginia market really, really well, but I’m just a mouse looking at the toenail of the elephant, because a national market is vastly different. And I think that’s where the algorithms really do come into play.
But nationally, home sales are down 2.4%, March to March. Which, from my lens is, it makes sense in a lot of ways because we don’t have the inventory. And this is the conversation of the last couple of years and especially this year, is that we’ve got new inventory. But it’s made it so that in our market, it’s harder than ever to price houses because when you list a house, you don’t know that demand. You know it’s going to be high in our market, but you don’t know the demand until it actually happens. And when you get one offer, two offers, three offers, four offers, et cetera. I hate to tell my clients, but I’m honest that when they ask me what the price of a house is, I’m like, “In two weeks, I’ll look at the data” because the data drives the second, third, fourth week on, the emotions drive and the desperation drive the first few days.
Keith:
Well I was at a meeting yesterday at our local associations, and one of the agents was discussing that we really needed to do a better job locally of combating the national media in regards to the housing market, because everyone in the housing market is saying, “Oh, prices are off and they’re not locally.” And I think that’s obviously the case, always, that your local market differs in some degree from a national market. We all trend in a similar manner, but they are specific, and so we talk about how real estate is local. But at the same time, the numbers can be deceiving in terms of what you’re looking at. So in Virginia, sales are off by 24% year over year from the first quarter. But the question is, is that because there is no demand, which is what we saw during the great recession, or is it because there’s simply nobody willing to move?
Because as we’ll talk about, I know we’ve had some numbers on the number of homeowners that have interest rates right now that are so far below what’s available, that they view it as, we can’t afford to move. And so people are sitting on houses. So the question then becomes, what is the real meaning behind the number that you’re looking at and how does that affect things? So if prices are up but sales are down, the demand is still there. And I think that’s what we’re seeing in some of our markets, but again, not in every market. So Phoenix, Houston, or not Houston, but Denver, Austin, these markets are off, way off, on demand. And we’re seeing dramatically different pricing from the Virginia markets that we’re heavily invested in.
Jonathan:
Yeah. And I think going back to your analogy of the elephant and the mouse, as we know a little bit about other markets, we know a lot about the markets where we operate in. And really, what things like the Zestimate do is make people feel like they’re experts in every market. I can pick a market right now in a country that I’ve never been to, and if you told me to go on to Zillow every day and just track prices within a fairly tight geographic area, I’d probably say, “Look, I’m a decent expert,” not the case. It’s probably not the case. So it’s kind of like a false belief in my mind that I may be better than I am.
Jim:
I think it’s a false confidence that buyers get when they spend the time on Zillow and they’re learning the market, they’re coming from another area, as you’re talking about, and then they get here like, “Oh, this is a lot different than I expected. And they look at this…
Jonathan:
That’s what that’s priced that way?
Jim:
They walk in the house like, “Oh, this smells like smoke,” or “This smells like cats,” or “This is next to something bad,” or it’s a functionally obsolescent house. Keith, do we know if the algorithm is there yet, that ascertains what that value is?
Keith:
Is it there? Probably. Is it being publicly disclosed to this point? Probably not. Right? This is the other part of AI, and we will get to that, is that so many of these models for, whether it’s for bank desk appraisals or whether it’s the Zestimates that we’ll get to, we don’t know what they’re based on. We’ll never understand their underlying piece. We can try and find them out, and Jonathan’s done some undercover sleuth work on his own houses to see what drives it. But I remember the last home that I sold before I got into the business, I was working with my realtor and I had actually gone down to the city tax office and had pulled every house that had sold within a year and a half of my neighborhood and gave her a full spreadsheet, because this is who I am.
Jim:
You would be a terrible client.
Keith:
Oh, I guarantee you, I was the worst client there. But my whole thing was, here’s where every single house is sold against tax assessments. Because that was the only third party free spirited kind of number that wasn’t based on somebody telling me, “Oh, your one house is based on this. What’s the averages that I can find?” Was to look at the tax estimates and figure out where people were selling. I have not been asked that by a client a long time, but what percentage above or below assessed value or house is selling for. We’re talking about on Zestimates now, we’re talking about a more readily available, more frequently changing, because tax estimates only get changed once a year or once every other year in the state of Virginia.
Jonathan:
But I will say that if you go on the Zillow website, and I’ll actually preface this right now saying, I greatly respect Zillow, I think they’ve had an amazing impact on the industry in a lot of ways. And so this isn’t a conversation ragging on Zillow.
Keith:
No.
Jonathan:
It’s just analyzing what they’ve done with this groundbreaking term, Zestimate over the last 15 years. So that being said, side note, kind of get back to it. If you go on Zillow’s website, they say… There’s an FAQ about Zestimates. And they say, “If you think your estimate’s wrong, you should go talk to your county tax assessor.” And so clearly that’s a sign right there…
Keith:
It’s a piece of their algorithm.
Jonathan:
That is a piece of the algorithm. And you said earlier that we don’t know what the algorithm is. We probably know 90% of it.
Keith:
We probably know 90% of the components of the algorithm. We don’t know…
Jonathan:
We might know a hundred percent of the components, but we don’t know how they’re weighted. Right? Who knows. And so that’s the interesting part is, it’s probably always changing on their end. We have a pretty good feel, and that’s why there’s been so many other AVMs that have popped up across the country that you can go onto these real estate data websites right now and they all offer as a service, some sort of AVM. Now, how accurate is it? We don’t know. Everybody’s throwing some spaghetti against a wall to see what works.
Keith:
Well, there’s a great book that Zillow published and it’s probably been a decade now, that’s called Zillow talk. I think it was, Stan Humphreys was their chief economist at the time, who published it, that was looking at all kinds of odd data points and how they affected, and you know that whatever he found on each of these individual chapters were being plugged into the larger algorithm in some way. But it was everything from, was the street name of a presidential last name or a tree name or a body of water type name? He had grouped everything, but it was also, how close are you to a Whole Foods? And they looked at and understood the impact of opening a Whole Foods or a Starbucks near a house, what that would do to the immediate surrounding area. And you know that became a permanent part of their algorithm, because it does. It controls what people are looking for right now.
Jim:
And I think that was the book where they talked and said that if you used the word unique in the description, that had a negative impact on the value because everybody wants to be a unique, delicate snow flower.
Keith:
Every house is unique, period.
Jim:
Correct.
Keith:
Yes.
Jim:
They’re not. I think that if you have four beds, two and a half baths, 2300 square feet, they’re all the same in a lot of ways. So I think it’s something that, when you’re looking, when you’re marketing a house and you’re pricing a house, you’re pricing it for the broad brush market. You’re not pricing it for Keith, who’s going to look at every single number known to man to figure out the value. You’re looking at the person who’s going to say, “Oh, okay, I need four beds, two and a half baths, and I need a bedroom on the main level.”
Keith:
220 a square foot.
Jonathan:
You would think though, based on, Jim, what you just said, you would think that a Zestimate of two houses that are identical and next to each other or say identical, like 99.9% identical.
Keith:
Same floor plan.
Jonathan:
Same floor plan, same square footage.
Keith:
Same builder.
Jonathan:
Same neighborhood, same builder. You would think that’s Z estimate would be pretty identical. So here’s a story. I’ve told you this too, but I would tell it again. I have a house in South Carolina that we bought a couple years ago, we rented out, but I started paying attention to the Zestimate on it, just out of curiosity to see what would impact it. And so what I did, about a year after we bought it, I went and I took the photos that we had professionally taken, and I uploaded them into Zillow. That’s the only changes I made. In the neighborhood where we’re in, there’s 22 houses, eight of them are almost identical to ours. And then I tracked the Zestimate. The only change I made is, I took this nice bright photos with new linens and…
Keith:
New paint.
Jonathan:
New paint and a little bit of new flooring and put them up in there and tracked the Zestimate.
And what did I find? I found that the month after I did that, the other comparable houses, those ones that are the eight similar, the Zestimates went up by one to 1.7%. My house went up by 3.6%. The house that’s next door that is literally identical, identical maybe with the exception of the paint color and the door color on the front door, went up 1.5%. And I tracked it for a couple months, and our house continued to outperform the other houses. And the only thing that changed, none of the houses sold, nothing else changed. The only thing that changed is, I put photos in there. And that goes back to 2018, I was at a Zillow conference that was invited to a Zillow broker forum, and they were talking about AI and how they were using AI to look at photos of properties to see what had a wolf range, what has hardwood floors, what has this, what has that.
Keith:
Looking for cats in the pictures?
Jonathan:
Might be looking for cats or animals in the pictures, and that’s probably…
Keith:
It sees a litter box.
Jonathan:
Yeah. Sees a litter box.
Keith:
Ashtrays.
Jonathan:
So I look at that as too as, that maybe is amazing. But it also, the question is, is there a way to gain the system a little bit?
Jim:
Well, I think when I go look at a house to put on the market, the top two things that I tell sellers they need to do, clean and paint. It’s not hard and it’s fairly inexpensive and a high ROI, but I would imagine that the AI can determine whether a house is clean. It can see dust bunnies, it can see cobwebs. It can see clutter.
Keith:
So the other, the third piece, if you go to the third one is probably going to be landscaping. It’s maintaining the exterior of your home. Which does make you wonder, if you maintain a home well, and you happen to have it looking perfect the day that the Zillow car or the Google car drives by and uploads, are they looking at exterior for general maintenance… Are they using it to look at the general maintenance condition of a home to say, this person takes better care than that person?
Jonathan:
Did your kid cut the lawn the day before or did they forget to cut the lawn for two weeks and all of a sudden…
Keith:
You got dinged by 2.2%?
Jonathan:
Timing is everything.
Jim:
Well, yeah, and I think goes back to, how does the algorithm, when Zillow, if a house comes on the market at X price, the Zestimate’s going to be X, and the Fannie, Freddy, they all have their own AVMs, but how does it account for the emotion of that buyer and the desperation of that buyer who lost six houses?
Jonathan:
It can’t account for that.
Jim:
Right. But when that house goes up for 15% of our asking price or whatever that number is, does that then impact the rest of those houses?
Jonathan:
Well, I think that if you go on a Zillow and you look at a listing that is actively for sale, you will see how many views it has and how many saves it has.
Keith:
How many people are following it.
Jonathan:
That probably is part of the Zestimate algorithm, right?
Keith:
Okay. But let’s talk about the human side of that. When we were agents prior to electronic lock boxes…
Jonathan:
Yes, good point.
Keith:
There was a sign-in sheet. And when you walked in with your clients and you saw that no one had looked to the house in three and a half weeks, your willingness to offer below asking by a substantial amount this went up. If you went in on the third day and there were 29 people who’d seen it, you told your client, if we’re going to get this, we got to go in and ask. So now the information is in a different location.
Jonathan:
Yes. And showing [inaudible 00:15:46]
Keith:
Hey, and you’re seeing how many people are…
Jonathan:
Well, I’ve got an idea. I just list my house and I go to all my friends on Facebook and say, “Hey, go and like…”
Keith:
Save it.
Jonathan:
“Save it and like this house and tell your friends, do it, and I’ll send you a $5 Starbucks gift card.”
Jim:
I’m going to go ahead and patent that idea.
Jonathan:
Gaming the system. Yeah. We just started a new company. I’ll see you later.
Jim:
Fantastic. One of the challenges with the AI is, it’s evolving so fast and it’s impossible to know where it’s going and how… Every three days, it seems like there’s a new story about how AI has been implemented and how ChatGPT is now hiring people on TaskRabbit to get people to do a task.
Keith:
Its hiring itself.
Jim:
Yeah. And I think I read one where it said that the ChatGPT or whatever it was, it was given a task, in order to beat the security, it hired somebody on TaskRabbit to come in and lie to accomplish this task for him.
Keith:
So yeah, the story was that they basically gave the chatbot a set amount of money and said, “Go out and make money.” And so it went and hired somebody. It tried to do a website that it was going to make money on an like an eBay type website. It was going to try and sell something. I can’t even remember what the site was, but it was going to try and monetize its value. And the website had a capture and it could not answer the capture. So it placed an ad on Fiver or whatever to hire somebody to go out and be a human to do it. And when the human said, “Are you just a chatbot trying to break the system?” It responded back, “No, I actually have difficulty seeing things and I’m blind, and therefore I need to hire people to help me.” And the people helped them and then broke in for it.
Jim:
I’m just saying, this is not going to end well.
Jonathan:
Yeah,
Keith:
It’s amazing. But to Jim’s point, it is changing every single day where it’s going. And I think I’m now sitting here thinking, I’d love to have our Zillow rep give us somebody within their AI group. Not to tell us how this is working, but for us just to throw out 50 random, do you use this? And for them to either say yes or no. Yes, that is included.
Jim:
I think it’s also, I think that understanding how the AI is crawling and looking for functional obsolescence and open floor plan, which now is going out of fashion in a lot of ways. Where does it value these things? And what should a seller do when they’re prepping their house? How far in advance should they get professional photos taken?
Jonathan:
Usually it was like, let’s get photos taken on Tuesday and we’ll list it on Thursday. And the question is, all right, you’re thinking about selling in March of 2024? Maybe you should put your new photos up in September and give it six months.
Keith:
Well, we’ve been advising clients for a long time that if you’ve got square footage in your house that’s not being counted by a county, to go back and come clean at least a year before your listing so that they can increase your tax assessment. Pay the higher taxes, don’t worry about that, because people do look at what the tax value, and we know that Zillow’s looking at the tax value. So the more accurate the information, the more current the information, the better it’s going to be for that valuation.
Jonathan:
Yeah. And now that Zillow is getting all these… People or realtors and sellers are uploading the 3D tours and things like that on to Zillow, now they’re starting to get a peek, not just with 2D inside the home, but with 3D inside the home, which is giving them eventually, probably the ability to look at functional obsolescence and what’s the layout is… I know there’s a house in my neighborhood, and you both know this, that I was involved with the developer of our neighborhood years ago, and the first 150 houses I looked at on floor plans with the builders before they were built. And there’s a house that’s down the street from me that is bigger, it’s bigger, beds, baths, square footage, everything. So the Zestimate is drastically higher than my personal residence.
But I know that the floor plan inside that house is so bad, that if someone came in right now and… The house is not for sale, someone came to me and said, “Hey, I like that house,” my first reaction would be like, “Do you want to have to walk 73 feet from your kitchen sink to go watch TV in your family room? Down a hall and around the corner?”
And they’re probably like, no. But so that’s the human right aspect of, when a realtor is in the trenches and they’ve seen thousands of houses and they know the builder and they know the problems that a builder has… Maybe things that a builder does well or the builder doesn’t do well, they have that…
Jim:
Innate sense of knowing what the intrinsic value is.
Jonathan:
Right.
Jim:
But at what point… And I’ll ask for projections here. At what point does the AI take over? How many data sets or how big of a gigabits or whatever, do they need? Or terabytes or whatever to understand that house that you’re talking about that is a higher Zestimate, is a trash house?
Keith:
It will never understand the motivation battalion the individual that is buying the house, and that’s where the difference comes in.
Jim:
But the commodification of housing is likely to happen with AI.
Jonathan:
Maybe.
Keith:
It’s getting better and better.
Jonathan:
It’s getting better and better. I think it almost takes, does it take a couple iterations of selling a house? So let’s go back to my example. If that house comes on the market today and where the market’s pretty hot, and everything else around it sells in three days, and that takes six months to sell and sells at 90% of list price, and then next time it comes in the market, the same thing happens. Then 20 years from now, when that house comes in the market, the Zestimate or the AI flags that as a, something’s off in this house. It’s got a little asterisk next to it.
Keith:
A couple of things to that. Number one, we always, as agents, we go back and we look at past histories of homes. We know that if a house sat on the market 120 days when the going rate was 20 at the time, we know there was something weird about it and we tried to figure that out. Whether it was the price or whether it was some other… The parkway was being built behind the house and nobody didn’t want to own the house during the construction or something. So there are things that we’re looking at. And yes, it’s going to pick up on those more and more.
I think it also goes to look at the old model, the old pricing piece that we used to talk a ton about was the case Sheller Index. Did not just look at median prices within areas and within MSAs, it looked at the same house, point A to point B, and looked at the change within that time. So it looked at specifics. And I think the K Schiller is the early precursor to what we’re dealing with right now, and yes, it is going to look at what outlying anomalies existed on that house.
Jonathan:
Yeah. Well, I think the Zestimates are really interesting to track. I’ve got some problems with it. One of the problems that I have is the fact that it reprises when a home is listed, so you see these stats out there that, a couple that I’ve read is 95% of Zestimates are within 10% of the final sales price. Well, that’s not that impressive a stat when you know that a house was… The Zestimate was 500 today, it’s listed tomorrow for 600, and all of a sudden, the Zestimate changes miraculously to 598 200. So there’s some things out there that, I’d call that cheating a little bit.
Keith:
Sure.
Jonathan:
But as we continue to go down more and more of this rabbit hole and the tech continues to improve and the AI continues to improve, I would really anticipate that pricing becomes… The Zestimates and the AVMs become even more accurate.
Jim:
Again, I think the more data, the better the algorithms, the better it’s going to be. I’m looking at the Zestimates now for the metro areas, and Atlanta is within their median errors 6.49%, Baltimore 6.74, Miami is 7.3%, and this is from November of last year. I would wager that algorithm has not been updated sufficiently to capture the March 23 market. But I think that also, when you’re looking at them, if it comes on the market at 500 and it doesn’t sell in 30 days, that Zestimate will drop. And so they do recognize that component to it.
Keith:
The interesting thing, I’m looking at the same page right now, Jim, and looking at, they’ve got 98.3% of homes in Atlanta are priced ready within 20% of the sales price, which is an enormous spread of 20%. So the question is, with that, is the 98% because 1.7% sold for more than 20% of asking price? Is that just the irrationality of the buyer market right now in some of these areas? Or is that because the Zestimate just wasn’t well-thought-out at all?
Jim:
Both. But I think that, to the that point, again, real estate is inherently a human-based business, and there’s no way to capture emotion of agents right now. There are the same number of agents that, if not more, in far fewer home sales. So there’s an emotional component to that that you need to be mindful of. But also the buyers out there who are desperate, who are looking for homes. And part of what we do is help pull that emotion of them back. And one of my clients asked me recently, it’s not AI, it’s human, obviously. And they said, “Do you have an ethical obligation to tell me if I’m overpaying for a house?” I don’t know, look at the estimate. But I think the answer is yes, but it’s also understanding that my role is not to talk them out of it, it is to give them that guidance. And they’re the ones making that decision. But no algorithm as of yet can give that guidance and can give that accuracy of pricing or that or that future pricing model of what it’s going to, accurately.
Jonathan:
Not for housing, I don’t believe. But we go to flip it a little bit to something that’s a little more of a commodity and it’s automobiles. And how much do we all trust the quote unquote Blue Book value.
Keith:
Constraints owned by the Automotive Dealer Association?
Jonathan:
But people trust it. You’ve had a conversation with somebody and they’ve said, “I got this for Below Blue Book.” Right? If someone said, “I got it below Zestimate,” I’d be like, big whoop. Right? Big woop. But if somebody says, you’re at a party and somebody’s like, “I bought it for 5,000 below Blue Book,” it’s like, oh. Your initial reaction is, well, good for you. Yeah.
Jim:
It’s a trusted benchmark. When I’m looking at bicycles, if my wife’s listening, they’re all under $500, I promise. But you look at bikes, and this is a bicycle blue book. And so when trying to ascertain what that value is, that’s my first thought. At least to see if it’s within the conversation. So again, for now, I think that it’s still human based, but the AI is going to get there, man.
Jonathan:
And they’re still human based, and the last thing I’ll say about the Blue Book is, I sold a car a couple years ago and I was putting it into the Blue Book, and there’s other companies out there that are offering these values, but it’s asking me what the condition is, right? It’s asking me all these questions.
Keith:
But it gives you very specific ways of…
Jonathan:
What’s a condition? Good, fair, great? What’s the difference between good and fair?
Keith:
Good.
Jonathan:
Everything that I have is at least good or great.
Jim:
Exactly.
Jonathan:
Yeah. There’s look ways to game to game the system, right?
Jim:
No, again, I think that it’s something that, when I was looking at the AVM, every single company has their own AVM. Insurance companies use them for replacement costs. And if you’re listening, might want to check your replacement costs on your house because those things have gone up significantly over the last pandemic years.
Jonathan:
Yes.
Jim:
So just double check what your replacement cost is. But they’re all using AVM to a certain degree. Just be mindful that they’re out there. For everything that we do, it’s data driven now, and the chat and the LLMs are going to be, more data every day, is going to be part of our continued existence.
Jonathan:
Agreed. So my recommendation to Zillow as we wrap this up is just, put something in there for the Zestimate of how many cats, and has someone smoked in the house, and we’ll see what happens with that.
Keith:
So I guess the only follow-up I would have is, does Apple look for and search our text to see if we reference Apple as a podcasting listening venue or Spotify to increase our ratings? In which case, we can Spotify this all day long.
Jim:
There you go. There you go. Well, cool, y’all. I think that the lesson of the day is, the machines are going to take over. Just be mindful of being nice to them.
Hey, y’all thanks for listening. We really appreciate you taking the time to hear us and to spend time with us. If you have not already, please subscribe. You can find us wherever you find your podcast.