220 Utilizing AI in Home Building Processes - Doug Newell Jun 5 2024

Greg Bray: [00:00:00] Hello, everybody. And welcome to today's episode of The Home Builder Digital Marketing Podcast. I'm Greg Bray with Blue Tangerine.

Kevin Weitzel: And I'm Kevin Weitzel with OutHouse.

Greg Bray: And we are excited today to have joining us on the show, Doug Newell. Doug is the CEO of Swarmalytics. Welcome, Doug. Thanks for being with us.

Doug Newell: Thank you.

Greg Bray: Well, Doug, let's start off. Just help people get to know you a little bit. Give us that quick overview and background about yourself.

Doug Newell: Sure. I am a serial entrepreneur, which means I'm a glutton for punishment. I have started seven companies. They [00:01:00] all have had to do with big data and the analysis of big data. And currently, I'm the CEO of Swarmalytics, which is focused in the building space using big data in building.

Kevin Weitzel: Sweet. Before we unpack all that though, Doug, could you do me a favor and break out one personal factoid about yourself that has nothing to do with business, work, nothing?

Doug Newell: Yeah. Let's see. I am currently surrounded by Borzois, also known as Russian Wolfhounds. I've had Russian Wolfhounds for 40 years. And I consider it one of the joys in my life to go out, I can't run like they do, they go at about 40 miles an hour, but walking them keeps me healthy, I hope.

Kevin Weitzel: That's cool.

Greg Bray: How big do they get?

Doug Newell: They get about 36 inches at the shoulders. They can stand on their hind legs and look me in the eye.

Greg Bray: All right. Big ones. [00:02:00] So Doug, tell us a little bit more about Swarmalytics and what it is, what you guys are doing because big data and home building is really high level. Let's go a little deeper.

Doug Newell: Okay, Well, first thing, the home building space, it looks ripe to me. And as any good businessman, you're looking for a place where there's a lot of potential and it hasn't been all used up yet. I started focusing on building when I heard that many builders use some fairly rudimentary processes to price their houses. So, you've built a house of 450,000, maybe, and then you're using some, as I say, somewhat simplistic methods to price it.

That rang of opportunity, I started digging in and learning a bit more about the industry. I found that there is in fact, one of the big players has entered with artificial intelligence and is leveraging that and that's Lennar. I've got to believe [00:03:00] they've probably invested millions into bringing AI into the market. But looking around, I don't see too much other competition and I think that smells like opportunity to me.

Greg Bray: All right. Well, you said those magic words that have everybody on their edge, AI. Tell us just what does AI mean to you? Because I think there's some different ways that term gets used right now, and we just kind of set a baseline.

Doug Newell: I was speaking at the Builders Conference in Las Vegas end of February, and I was fascinated that people just latched on to AI and they flocked to my presentation because I tried to explain it in simple terms.

AI, when you throw away all the math and all the spooky things and Arnold Schwarzenegger movies, forget about that. AI is about using information to make better decisions. [00:04:00] So, we gather up data, reflecting things that have happened in the past. We find the patterns in there and we apply them to future decisions. It's that simple.

So, I mentioned before pricing of homes. You reach into data, you gather up every house sale that has occurred in the United States in the last, say, 12 years. You have a ton of data about the house and then you come out with, and what did it sell for? So, you have what the interest rates were at that moment of time, and how many beds and how many baths and so forth.

And what we do is we bring more data to bear on the decision than anybody else. We bring about 4000 different pieces of information, including how close is this house to a golf course. How about to a lake? How about to a toxic waste dump? What's the crime rate? How good are the schools? Four thousand pieces of [00:05:00] information go into the soup and the algorithm just chunks away.

And again, I'm not going to bore anybody with the math, but the algorithm just chunks away and finds the underlying patterns that yields an equation. And the equation is the estimated value of this house is this times this plus this times this and so forth going out maybe 25 factors that would go into that.

Again, simplifying it, taking information from the past, analyzing it, and predicting something. That's the part of AI that we focus on. There's a whole nother side of AI that many of you have heard of and may have used, which is having to do with text and chat kinds of things. That's as big a mystery to me as it is to you. I don't operate in that space.

However, I think the basics are still the same. Take a lot of [00:06:00] history, analyze it, and predict something. So, the history might be everything that's ever been written about emus or something like that. It's digested all of that and comes up with a paragraph or a story about emus or something. Not my thing. Don't know how they do it. Fascinating, but we focus on predictive analytics.

 

Kevin Weitzel: Our little side note and a challenge to our future guests. Doug has mentioned both Toxic Waste Dump and Emus in the same podcast episode. I challenge all future guests to bring in any crazy things like that because that was amazing and it keeps me on my toes because I always look for the weird stuff, Doug. I really do.

Doug Newell: Well, it gets weirder in that earlier this morning I was telling my team to throw out the word hippopotamus if it comes into our analysis. [00:07:00] We are digesting data about short-term rental properties and part of our data flow there is all the descriptions. So, I was explaining that if the term hippopotamus comes in, in the million or so records we're analyzing, throw it out. The fact that somebody has a farm on their property that has a hippopotamus, it's not going to come in often enough to be statistically significant so we throw it out.

Greg Bray: Does it come in enough to even be mentioned?

Doug Newell: It might come in one time. I can tell you from analyzing rental housing data, the word stainless is important. The word stainless means the house will sell 30 days faster and at a higher price. People love stainless steel appliances, almost to the point where if you're selling a house that they don't have stainless steel appliances, you might do the math. It may be better for you to throw [00:08:00] away whatever appliances are in there and put in stainless so that you can be talking about stainless steel appliances. So, words are just variables that enter into equations.

Greg Bray: All right. That's fascinating that you can link one word or one little data element to that big of a difference.

Are there any others along those lines that just jump out?

Doug Newell: Yes. In analyzing MLS listings and things like that, the word fixer-upper, that's bad. It's not going to sell very quickly and it's going to sell at a negative premium. Something indicating repairs needed, people want to buy their dream, they don't want to buy a list of problems that have to be tackled. They do that only sort of under duress. So, sort of themes like that.

Now, back to, we need lots and lots of data. Hippopotamus might only come in once, if it does at all, in a million records or something. Stainless [00:09:00] steel may come in tens of thousands of times in a million records and so forth. This is one of the reasons we thrive on big, big data, why we have 4000 variables about each house, and why we have every house we can get data on all goes in so that these subtle variables enter into the equation if possible.

Kevin Weitzel: So, let me get this straight. So, if we take all these data sets and we find out that X and Y and Z and all these different columns that cross-reference with each other and it's all analyzed that, you know, like split second time, the home building industry could literally utilize this information to build the perfect house. So, aren't we, in essence, creating a scenario where all home builders will literally be building the same house?

Doug Newell: Great question. I'd question the premise. A great house for you and a great house for me is not the same thing. A great house in San Diego and a great house in Bangor, Maine [00:10:00] is not the same thing.

What we do and why we have the company called Swarmalytics, we use a proprietary technology where we explode the United States into 56 pieces. Each piece gets its own predictive model because I can't figure out any way to get Malibu, California, and Wilmington, North Carolina to fit in the same model. They might as well be, on different planets.

So, we explode the country and we don't just do it geographically. Lexington, Kentucky may have more in common with Burlington, Vermont than with Louisville, Kentucky, or something. So, our proprietary technology explodes the country into pieces, builds a model on each piece, which means we're building again, 56 models.

What we do is we score every house in the United States, about 100 million of them, four [00:11:00] times a month with these 4000 variables in order to predict things like, will this house sell? Will it be transformed into rental property? Will it be flipped where somebody buys it, fixes it up, and then sells it again? If it is turned into rental property, what will the rent be?

Anyway, we just proliferate predictive models that come up with a new piece of information, like the expected rent on this house is 2000 a month. That's the output from our process, which people then use in order to make decisions, like, should I buy this or not? Or should I flip it or things like that?

Greg Bray: So, Doug, as your getting into this business and creating these models, have you seen builders that have just totally missed the mark with their pricing, where you've come in and say, gosh, you're losing 20 percent of home here because you're not priced right?

Doug Newell: That's the interesting thing they don't have to [00:12:00] totally miss the mark. In fact, if they miss the mark by pricing too high, their methodology is usually that they start backing down. But as they back down, maybe lowering the price every couple of weeks, time has a cost. And sitting on a house, say, a 500, 000 dollar house for 3 months, because you priced it wrong until you get to the right price, those 3 months have insurance costs and maintenance costs and risk of vandalism and just the cost of the capital that's tied up and so forth.

The wonderful thing from my perspective on this market is the prices are so high that small improvements more than pay for my company services. If you think that the house is 450 and in fact, it should be 460, small percentage change, but that's 10 grand. I wouldn't be charging anywhere near that for a deep analysis where it's [00:13:00] again taking into account the local schools are subpar and therefore they ought to price differently or something.

Greg Bray: Okay. So, what you're saying is we tend to figure it out because if we start high, the market will tell us what they're willing until we sell it. But if we start low, we might be leaving money on the table and never even realize it.

Doug Newell: Exactly. Exactly. I mean, somebody can be very happy that the house sold in five days. They may be bragging about it. It may not be something to brag about. It could be that you could have priced it 10 or 20 grand higher and had it sell in 30 days, which would have been acceptable. That 20,000 dollars extra might have been a major improvement in your profitability.

Greg Bray: So, we've been talking about your model for pricing. In your presentation at IBS, you talked about a handful of ways that builders are utilizing AI types of tools. What were those key ways that you've seen builders use AI?

Doug Newell: First, it begins [00:14:00] with land. So, what we've done is built a model, actually, a series of models to predict which pieces of undeveloped land are most likely to be turned into residential communities or apartment buildings. So, there are, I forget how many million pieces of empty property in the United States, but we have used all of our data and scored each one of them.

This came out of an observation where we were doing some analysis and we were looking at the city of Dayton. And it looked like rings on a tree. You could tell what was going to be developed around Dayton based upon the distance from the center city and the distance from the last ring and things like that. In any case, we have scored every property in the US as to it's attractiveness to be turned into residential or apartment.

We also have another model. Is it likely to be sold? If it's currently owned by an 83-year-old farmer who is [00:15:00] sort of retired, that property has a much greater likelihood of being sold in the next few years than a property that has a different owner characteristics. So, we come at it from two different directions. One, is it attractive? Two, is it likely to be sold? And we could come at it from a third, which is how much should you pay for it? We have that as a product right now that somebody can go and log in. I think it's on our site and they can take it for a test drive kind of thing to play around and see the properties.

I live in the Asheville area and it's interesting all of the potential development is between Asheville and a city a little south to us, Hendersonville. Up in my area, which is north of Asheville, there is little or no development potential, and that's based upon again, hundreds of factors and looking at where development's been in the past.

Another product is [00:16:00] one that sits on top of a website and makes the website more interactive. So, when somebody is on the website that information, that data about what they're doing, how long they've looked at this house, did they go back to that house, did they come back on another day, and so forth, all that data flows to our data center. It's enhanced with additional data about where the person's coming from. It makes a difference between whether they're coming from Greenwich, Connecticut, or down the street kind of thing.

Then we trigger interactive behavior on the website. So, it could be that somebody has looked at a house, but then the website gets notification the house is under contract. So, our technology could pop up a, hey, that house is gone now, but would you like to see three others just like it kind of thing. So, it makes it more interactive and it also is [00:17:00] a funnel of data into the other models. The probability of a house being sold in the next 30 days, one of the things is maybe how many times it's been looked at on the website and for how long and by how many people and all of that website data pours into the models as well.

Greg Bray: So Doug, we have to unpack that one more because, you know, this is The Digital Marketing Podcast. so we need to drill into the website a little bit, because one of my first reactions to that is, well, that's only going to work well if the builder has certain content on their website. If it's a, we'll use the word crappy website, then this type of overlay isn't going to be able to capture the kind of data that you need to really make a difference, right? So, what is it that is your recommendation in that baseline of what kind of content needs to be on the website for something like that to work?

Doug Newell: Most builders will have this if they have houses that are for sale on the website, then it'll pick up more information. Like, again, how long somebody has [00:18:00] been looking at it and across how many days and things like that. If they only have sort of a crappy website that has some blueprints on it, we can't do much for them there.

I mean, all AI algorithms thrive on lots of data and if you've got a website that's so crappy that nobody goes there and it's got two pages of just a blueprint of a cape and a blueprint of a split level or something, there are going to be few visitors, which means we can't have enough observations to do any prediction and they're not going to stay long and they're not going to generate much data. If you've got a robust website, where somebody can wander here and wander there and interact in different ways, then the volume of data that flows into our algorithm increases.

By the way, I should emphasize we maintain privacy. At no point, are we looking at the name of the person visiting the [00:19:00] website or their home address or anything like that. It's got just, here's a key that identifies a record, and here's behavior about that record.

Kevin Weitzel: So, you're not diving into the IP information of this person earns this amount of money, lives in this specific spot?

Doug Newell: No, no.

Kevin Weitzel: Okay. But isn't that relevant to the analytical data collection?

Doug Newell: It could be, but you have to walk a fine line there. First thing, I believe in California, there's some legislation that you have to put up a warning on your website if you are in fact capturing personally identifiable information like that. We don't want to make it creepy. Okay. You want a website to be welcoming and informative and so forth. Technically, we could have something pop up that says, Hey, Kevin, good to see you. How's the wife and kids, blah, blah, blah, kind of thing? That would be creepy.

You don't want to drive people off the website. You want to [00:20:00] have the website, an initial place, that leads to probably a meeting. You could have something that says, Hey, you've looked at this house three times in the last three days. We have a broker ready on the phone right now. Push this button and you'll talk to them and you can arrange a viewing of the house or something. You can have that, but you can't get into, Doug, we know your wife's name is Debbie and you moved there in 2017,

you know.

Kevin Weitzel: How was your divorce last year? You know, hope that divorce went well.

Doug Newell: Yeah, exactly. Have you gotten over that?

Greg Bray: Doug, when you collect this kind of data on the web traffic, what are some of the interactions or actionable items that you recommend after you know a certain amount about this particular visitor?

Doug Newell: Well, to start with, some part of the population comes one time, stays for 30 seconds, and is never seen [00:21:00] again. Can't do anything about that. Okay. At a minimum, they've got to look at multiple pages of the website. A house is a big purchase. Often there's another party involved, like, a spouse kind of thing. I don't believe anybody, or I don't believe very many people, look at a house one time on a website and then go off and buy it or something. I think there's a, hey, Deb, come on down. You got to see this thing. What a deal. It's just what we've talked about kind of thing.

So, coming back on another day, I think is a strong indicator. Time spent on a particular property, and these are all variables that we have to manufacture based upon the stream. We simply get a stream that says this IP looked at this page number and they started at this time and ended at this kind of thing. So, we have to manipulate [00:22:00] that to, Hey, this is the same person that we saw yesterday. That's something we create. So, now we've got return visitor field, and that return visitor field goes into the algorithm.

Again, we want to trigger things that make it easier for them. Whether it's a recommendation engine, it could be as simple as a coupon. You give somebody a 500-dollar coupon on a house, it may move the needle. The good news is, we are doing an A/B test all the time. So, we have people who are receiving the stimulus and people who look just like them who do not. So, if you want to put a 500 dollar coupon on a house to try to get somebody to take a tour or something, we only give that coupon to a fraction of the people and then we say, okay, the ones that got the coupon, here's how it turned out. The ones that didn't, had a different behavior.

Greg Bray: Anybody doing a buy one, get one kind of [00:23:00] deal cause I'd go for one of those.

Doug Newell: That would indicate a greater margin than most of the builders have.

Greg Bray: So Doug, when you talk with builders about, okay, we're trying to move them forward, is there a particular step in the process that you're trying to get them to take next? Is it simply the contact, like reach out? What are you after?

Doug Newell: that's a wonderful question. Coming off of that speaking engagement in Las Vegas at the Builders Show, I realized where most builders are in the AI process. Don't mean to have a sales hat on or anything. We've come up with something called an audit, an AI readiness audit, where we go in and do a deep dive on what's their data look like. I can tell you if their data isn't captured electronically, if it's in a file cabinet somewhere, that's not a good sign for their readiness on AI.

So, we look at the data. We look at what processing they're using. We look at the [00:24:00] expertise by interviewing key members of the team. On the expertise thing, a lot of that has to do with framing the right questions. Okay. This may seem inside baseball or whatever, but if you ask AI the wrong question, it's going to give you the wrong answer to what you want.

So, if you try to get it to maximize revenue, you may not do very well from a profitability perspective. You really want it to be maximizing profitability. So, in any case, we look across five or six different dimensions and do up an analysis to tell the builder where they stand versus other builders and versus where you have to be in order to start getting the rewards of AI.

Greg Bray: To kind of dig a little deeper, if we are looking at a particular visitor on the website, we know this information, they've been back a few times, we're thinking about trying [00:25:00] to incent them with some particular messaging. What is typically your goal of what you want to incent them to do? What is that next step?

Doug Newell: For builders and for almost all the rest of us, there is, in my mind, a funnel. At the top of the funnel, there are all the people that might be interested in buying a house. Okay, and then they go through and maybe hitting the website. Now, they're a little way down the funnel. And then finally, at the very bottom of the funnel, they are signing mortgage documents. So, you want to figure out what this funnel looks like and where the sticky spots are. You want them to slide right down into buying a house.

So, in the funnel, going from the website, the next thing I would think is probably a view of the house. Getting them in touch with someone who can answer questions, [00:26:00] someone who can provide additional information, someone who can give them a tour of the house or work through finances or all of those services, that would be the next thing.

So, you don't want to be offering lots of incentives to people that are just bubbling at the top of the funnel. There are too many of them and it's too expensive. People's time is expensive and so forth. So, when somebody has entered and they've come back three times to look at this house and you've checked and at least they're plausible in that they are living in a zip code that has some money. Maybe they're living in northern Virginia and they're looking for a house in the mountains or something. You've worked through all that. Now you have people that cost a lot of money interact with them to try to close the deal.

Greg Bray: Well, Doug, this has been very interesting to hear how you're using this [00:27:00] data in some of the different ways. When you think about just this AI term. And again, because it's such a buzz right now, and you described already kind of some of the different directions and what it means, what would you recommend to people as a way for them to really understand where a particular tool fits in kind of the potential toolbox? Just because somebody calls it AI or whatever, it's like, there's a lot of who knows if it really is and what's really going on. What's your advice there on how to vet some tools like this, that people are claiming are AI?

Doug Newell: I would in fact start at sort of the other end of what are the opportunities if you had better decisions. The most obvious one for me in the builders is the sales price. There may be just as large of an opportunity in the estimating process, which is painful and time-consuming, and so forth, as I understand it. Where are the opportunities? Look [00:28:00] there and try to get your hands around it.

Well, if I could sell my houses for 3 percent more, what would that mean to me? That kind of thing and then go looking at what data do I have? Do I either know someone who has the algorithms and so forth? I would not expect any builder to become expert in AI. Why would you want to learn that? I don't know how the transmission on my sports car works, but it does. So, I'd probably try it in some sort of test period and so forth so that I can see that it actually works.

But the key thing is getting the question right. Here are 200 houses I've sold over the last three years. What should I have sold them for based upon the AI? Am I shooting too high or too low? How long did the AI expect the house to sit on the market versus how long did it? Do a [00:29:00] pilot project of some sort and then start making more financial commitments and so forth. But first, get the questions. What one thing, if you knew the answer to it, what one thing might be the most valuable to you and focus on that?

Greg Bray: No, it makes sense. Well, Doug, again, we appreciate your time today. Do you have any last thoughts or words of advice you want to leave with our audience?

Doug Newell: Swarmalytics believe it or not is spelled like it sounds. If anybody's interested in learning more about us, www.swarmalytics.com. Or reach out to me, you can find me on LinkedIn and so forth, and I'd be happy to answer questions you have whether it ends up being a business relationship or not, I'd be happy to pass along. I'm one of the few people who has about 25 years experience in AI. I started an AI company in 1998. I can answer [00:30:00] many of the questions and I'm happy to share my knowledge.

Greg Bray: Well, thanks again for spending time with us today. We really appreciate it. And thank you everybody for listening to The Home Builder Digital Marketing Podcast. I'm Greg Bray with Blue Tangerine.

Kevin Weitzel: And I'm Kevin Weitzel with OutHouse. Thank you.