Latest / Net Effective / Hot take: You might need machine learning, not AI

Transcript

Hot take: You might need machine learning, not AI · Net Effective

• 0:00
  SPEAKER_01: I don't know that leading width is going to take all your jobs is the best way to get people to adopt it.
• 0:04
  SPEAKER_01: As smart as we are, as much data as we had, as the algorithms were very accurate at a high level or on the average.
• 0:09
  SPEAKER_01: You have to admit to yourself that no one knows the value of the rental home more than the person renting it.
• 0:14
  SPEAKER_00: That's Garrett Van Paris, and I'm Jonas Bordo, CEO and co-founder of DwellZ.
• 0:19
  SPEAKER_00: This is Net Effective, the Rental AI Show.
• 0:21
  SPEAKER_00: Thrilled to have Garrett on the show this week.
• 0:23
  SPEAKER_00: He and I have known each other for many years and have always been impressed by his thinking and his operational skills.
• 0:27
  SPEAKER_00: Garrett's been a leader at Invitation Homes for many years, where he helped oversee, amongst other things, a combination of several companies into what we now know as Invitation Homes.
• 0:36
  SPEAKER_00: And he helped build the platform from the ground up, right from the first when they were getting started and trying to figure out how you can do SFR at scale.
• 0:44
  SPEAKER_00: Garrett probably knows more about how to use data and analytics to run operations than anybody I've ever met in the industry.
• 0:50
  SPEAKER_00: Garrett, I'm thrilled to have you on the podcast today.
• 0:52
  SPEAKER_00: Thanks for joining us.
• 0:53
  SPEAKER_00: Thanks, Jonas.
• 0:54
  SPEAKER_00: Good to be on and looking forward to the conversation today.
• 0:56
  SPEAKER_00: Absolutely.
• 0:57
  SPEAKER_00: So I want to start in the early days of SFR.
• 1:00
  SPEAKER_00: I feel like you were there right at the beginning.
• 1:02
  SPEAKER_00: What did analytics look like at that time?
• 1:04
  SPEAKER_00: What did data look like at that time?
• 1:06
  SPEAKER_01: Yeah.
• 1:06
  SPEAKER_01: Well, I started in 2012 at a company called Colony American Homes.
• 1:10
  SPEAKER_01: It was sponsored by Colony Capital.
• 1:12
  SPEAKER_01: This is one of the few funds that became eventually Invitation Homes through a couple of mergers.
• 1:16
  SPEAKER_01: And when I showed up, I think I was employee number 30.
• 1:19
  SPEAKER_01: And we had an access database and some Excel files.
• 1:22
  SPEAKER_01: And that was the Extent R Data Analytics capabilities.
• 1:25
  SPEAKER_01: No ERP, no systems like a YARTI or anything that people are familiar with in this space.
• 1:30
  SPEAKER_01: And so we were making do tracking all the activity that was happening, which was largely acquisitions at the time, knowing that we were going to have a lot more to track someday.
• 1:39
  SPEAKER_01: And we were working towards a solution there, but nothing existed when I joined.
• 1:42
  SPEAKER_00: Wow.
• 1:43
  SPEAKER_00: Access database, that really brings me back.
• 1:45
  SPEAKER_00: I remember trying to toy around with those and getting absolutely nowhere back in the day.
• 1:50
  SPEAKER_00: So you know, I feel like you were there right at the beginning.
• 1:53
  SPEAKER_00: You know, the industry was kind of being invented.
• 1:55
  SPEAKER_00: What did you have to build because it didn't exist yet?
• 1:58
  SPEAKER_01: Yeah, we had to build everything.
• 1:59
  SPEAKER_01: So what we had was a lot of capital and we had an opportunity to go and execute on, but we didn't have any systems or support in place.
• 2:06
  SPEAKER_01: And so when I joined the company, we were growing rapidly.
• 2:11
  SPEAKER_01: We had outsourced all the management.
• 2:13
  SPEAKER_01: All the emphasis and focus was on taking the capital that we had raised and deploying it and buying homes, investing in those homes and bringing them back online so people could rent them and we could put families in houses.
• 2:23
  SPEAKER_01: That was the initiative.
• 2:25
  SPEAKER_01: A lot of the focus, obviously, was on acquisitions, underwriting, and then making sure that we could take the CapEx dollars that we underwrote, put them into the homes, and make sure that they were ripped clean, safe, functional, and ready to move in.
• 2:36
  SPEAKER_01: That process took a lot longer back then than it does today.
• 2:39
  SPEAKER_01: Just knowing where the homes are and what status they were in was a huge challenge because it was a game of telephone with the people that were involved, whether they're affiliates or other managers, the acquisition partners.
• 2:50
  SPEAKER_01: Sometimes we bought the wrong house a couple of times.
• 2:53
  SPEAKER_01: We meant to buy a different house on First Street and we bought that one.
• 2:56
  SPEAKER_01: And we were flying really fast because again, the window was tight here.
• 3:00
  SPEAKER_01: We had never had an opportunity like this to reinvest in single family housing and make sure that we were buying in the right neighborhoods where people would be interested with you know the characteristics, good schools, clean, safe, functional, all that that we're seeking out.
• 3:12
  SPEAKER_01: And once we got the homes in our possession, we acquired them, they had to go through a rehab process.
• 3:18
  SPEAKER_01: And so my first job was to help uh a gentleman named Justin Anicone, who was working with Jay McKee at Vineyard Services when they started Colony American homes.
• 3:25
  SPEAKER_01: They did the initial raise of colony capital is to track all the homes and were they pre-rehab, were they in rehab, were they being inspected, or were they ready to deliver?
• 3:32
  SPEAKER_01: And by the time we had probably acquired about a thousand homes a month at this point, we were doing that clip.
• 3:37
  SPEAKER_01: We had hired an industry veteran named Fred Tuame.
• 3:40
  SPEAKER_01: And Fred had come for 20 years at Equity Residential, Sam Zell, and uh he had retired, and it took about six months for Colony to talk him out of that retirement.
• 3:49
  SPEAKER_01: And he joined and he looked at the numbers and he said, Who does the numbers for operations?
• 3:54
  SPEAKER_01: And I was doing it for Justin at the time.
• 3:56
  SPEAKER_01: And uh and Fred said, How many homes are we requiring a month?
• 3:59
  SPEAKER_01: And it was about a thousand.
• 4:00
  SPEAKER_01: He goes, How many homes are we having a month?
• 4:01
  SPEAKER_01: It was about a hundred.
• 4:02
  SPEAKER_01: And you can just do the inventory math there.
• 4:04
  SPEAKER_01: We're accumulating a huge backlog.
• 4:05
  SPEAKER_01: And so Fred said the thing with the numbers and said, You got to do a thousand next month.
• 4:09
  SPEAKER_01: And you know, I think the team about fellow their chairs.
• 4:12
  SPEAKER_01: And uh we actually did a thousand.
• 4:13
  SPEAKER_01: We trapped it every day, got to a thousand and four.
• 4:15
  SPEAKER_01: And he said, Good, do 1200 next month and then keep it there.
• 4:19
  SPEAKER_01: And so we had to eat through this backlog we'd accumulated.
• 4:22
  SPEAKER_01: And we could only understand if we were doing that through tracking things, building systems, and then uh analyzing it.
• 4:28
  SPEAKER_00: That is wild.
• 4:29
  SPEAKER_00: Uh, you know, I think about our friends in multifamily and the challenges they have, you know, but you buy a 300 unit community for 200 million dollars, it's pretty chunky.
• 4:37
  SPEAKER_00: It's hard to lose track of it.
• 4:38
  SPEAKER_00: But the same amount of money spent on 500 houses is is a very, very different beast.
• 4:45
  SPEAKER_00: 500 different locations, 500 different rehab challenges, very challenging to find scale in that kind of thing.
• 4:51
  SPEAKER_00: It must have been epic.
• 4:52
  SPEAKER_01: It was a lot of uh new challenges.
• 4:54
  SPEAKER_01: Uh, we hired a lot of veterans from the multifamily space who all said, this is harder.
• 4:57
  SPEAKER_01: This is more difficult, everything is more disaggregated, right?
• 5:01
  SPEAKER_01: And uh I think that uh again, for someone like Fred, who when he landed here, he wrote out on a big piece of paper, this is what I usually look at to run the business.
• 5:09
  SPEAKER_01: It was called the executive overview.
• 5:11
  SPEAKER_01: And he said, I need you to make it.
• 5:13
  SPEAKER_01: And I said, We don't have the systems to make this, you can't automate any of this.
• 5:16
  SPEAKER_01: And so he says, I need you to make it because I didn't know how to run this business.
• 5:19
  SPEAKER_01: And so over the next two weeks, we had this sprint to put the uh the simple things on the page.
• 5:23
  SPEAKER_01: How many homes do we own?
• 5:24
  SPEAKER_01: How many are down, how many being sold, what are the various statuses of the ones that are in the net?
• 5:28
  SPEAKER_01: And then he knew where the homes were, and now we had to go then build all the stats around what's occupancy level, what's turnover look like, reasons for move out, rent growth, all that sort of thing.
• 5:39
  SPEAKER_01: And when we finally stood that up, my job famously was to update it and I wanted it to look automated.
• 5:44
  SPEAKER_01: And so I updated that thing manually, pulling front-end reports out of Yardy, putting them into an Excel model every night, scheduling an email to go out at 1201 to a it started with Fred, and then it grew to an audience about 150 people throughout the organization over the next three years.
• 5:57
  SPEAKER_01: And I did it for about three years straight as a labor of love.
• 6:00
  SPEAKER_01: But it felt like we had a kind of a true north at that point to understand where the portfolio was.
• 6:05
  SPEAKER_01: And by the time we did the merger with Starwood all those years later, this was in 2015 to 16, we had gotten to 95% occupancy.
• 6:12
  SPEAKER_01: The red occupancy at 94 and change went green at 95.
• 6:16
  SPEAKER_01: We merged the company and we had all these new homes in the report.
• 6:20
  SPEAKER_01: And we were really able to manage and drive efficiency through something like that.
• 6:24
  SPEAKER_01: It's now automated, it still runs at imitation.
• 6:26
  SPEAKER_01: I think it's a BI tool, but those were the fun days, I would say.
• 6:48
  SPEAKER_00: Combining data sets, infrastructures, management tools.
• 6:51
  SPEAKER_00: Yeah, I got to be a part of both mergers and integrations at a young age.
• 6:55
  SPEAKER_01: I felt very privileged to do that.
• 6:57
  SPEAKER_01: And it was difficult.
• 6:58
  SPEAKER_01: But the good part of the silver lining about all that is we had a lot of great athletes to pick from.
• 7:02
  SPEAKER_01: Some people were, okay, I'm gonna call my number, I'm ready to go do something different.
• 7:06
  SPEAKER_01: Famously in the uh invitation, the colony and Starwood merger, rather, uh Doug Bryan, you know, he was gonna be the company, I think, COO originally.
• 7:13
  SPEAKER_01: And then he decided to go start mine property management and now as part of Roofstock.
• 7:17
  SPEAKER_01: And uh Ali Nazar was our integration leader.
• 7:19
  SPEAKER_01: He joined them at Mine for a little while.
• 7:22
  SPEAKER_01: And you know, Charles Young was their COO at Starwood, and he became our new company COO.
• 7:26
  SPEAKER_01: And I met him during integration when I was sitting in the tech department technically, kind of taking orders from Fred and helping build a lot of insights analytics.
• 7:34
  SPEAKER_01: You know, Charles was like, okay, after integration, don't go back to tech, come work for me in operations.
• 7:39
  SPEAKER_01: And uh, I saw where the business was going.
• 7:41
  SPEAKER_01: It wasn't an acquisitions challenge anymore.
• 7:44
  SPEAKER_01: I wanted to follow the challenge as it went through the business.
• 7:46
  SPEAKER_01: It was becoming an operational challenge.
• 7:48
  SPEAKER_01: And under the COO, that umbrella was about 75 to 80 percent of the business, uh, the organization center.
• 7:53
  SPEAKER_01: And so that's where the muscle was.
• 7:55
  SPEAKER_01: And uh, for those of you who don't know, Charles Young, he's a world-class leader.
• 7:58
  SPEAKER_01: And when he he says, hey, come work for me, it's an easy decision for a young guy like me.
• 8:02
  SPEAKER_01: I think I was 26 or 27 at the time, still getting my MBA and on the weekends in an executive program.
• 8:07
  SPEAKER_01: And I said, Yeah, uh, of course I'll come work for you.
• 8:09
  SPEAKER_01: And so we kind of moneyballed it in this way where we want to drive operational excellence.
• 8:13
  SPEAKER_01: Well, drive operational efficiency.
• 8:15
  SPEAKER_01: And 18 months later, we did the invitation homes merger and do it all over again.
• 8:20
  SPEAKER_01: And that was a really fun exercise because now you have 80,000 assets spread out across the country, a lot of overlap, some new markets.
• 8:26
  SPEAKER_01: And you had to decide, you know, just the first thing operationally is how we're gonna run this business.
• 8:30
  SPEAKER_01: And so, how many people do you need in which positions and you know, to what extent in which markets can we drive efficiency or are we gonna be running a little bit less lean now?
• 8:38
  SPEAKER_01: And so that's all a modeling problem.
• 8:40
  SPEAKER_01: But without data, without analytics, without information, we wouldn't have had a plan that came together as quickly as it did.
• 8:45
  SPEAKER_01: And we really were able to pick up and run with it and not see a big uh hurdle or a kind of a trip-up point in either of those mergers, which is pretty fascinating when you think about it.
• 8:54
  SPEAKER_01: We went from zero homes to about 80,000 homes in about five years, matrix really quickly.
• 8:59
  SPEAKER_00: Yeah, and kudos to you and the team because uh every transaction that I've ever been a part of has lots of hiccups.
• 9:04
  SPEAKER_00: It's never as easy to combine organizations as anyone wants it to be.
• 9:08
  SPEAKER_00: Uh, it feels like it should be straightforward to the people doing the deal.
• 9:10
  SPEAKER_00: And then uh when it hits the operation side of things, it's pretty messy.
• 9:14
  SPEAKER_01: When you think about Yardi, no one had put that many properties in Yardy before.
• 9:17
  SPEAKER_01: It's a one-to-many relationship, it's properties to many units for multifamily.
• 9:21
  SPEAKER_01: Yeah, one property, 300 units.
• 9:22
  SPEAKER_01: We had one property to one unit.
• 9:24
  SPEAKER_01: And so we were gonna have 80,000 properties and it's one single Yardi.
• 9:27
  SPEAKER_01: So we had to work with those guys and see like it's just gonna work.
• 9:29
  SPEAKER_01: And there's a lot of breakthrough points there along the way.
• 9:32
  SPEAKER_00: Yeah, yeah, I can imagine.
• 9:34
  SPEAKER_00: So changing gears a little bit, Garrett, when we were talking before the call, before the the recording, you mentioned that there's still a lot to mine in machine learning before we even get to AI.
• 9:44
  SPEAKER_00: I'm curious first, just for our audience, what's the difference between machine learning and AI?
• 9:49
  SPEAKER_00: Where's the line between them?
• 9:50
  SPEAKER_00: And then, you know, what do you think we have left to mine from machine learning?
• 9:54
  SPEAKER_01: Yeah, there's a number of ways you can take that, but I think machine learning has been here for a little while.
• 9:58
  SPEAKER_01: It didn't give as much shine as artificial intelligence.
• 10:00
  SPEAKER_01: It doesn't sound as sexy as you know, machine learning.
• 10:02
  SPEAKER_01: Um, but these are where we build our algorithms.
• 10:04
  SPEAKER_01: These are the things that are deterministic when we say if we run that algorithm, we're gonna get the same answer as long as we have the same data and the same status of things.
• 10:11
  SPEAKER_01: Or AI, you know, a lot of people are talking about modeling and underwriting with artificial intelligence.
• 10:16
  SPEAKER_01: Sometimes you can use the same data and get two answers, you know, when you push the button to run twice with artificial intelligence.
• 10:22
  SPEAKER_01: I think with machine learning, it's less expensive, it's a little bit more straightforward, it's more explainable most of the time.
• 10:29
  SPEAKER_01: And then you can show a clear return from this is what we're gonna do, and this is what's coming out the other side.
• 10:35
  SPEAKER_01: For example, pricing.
• 10:36
  SPEAKER_01: Pricing is on everyone's mind.
• 10:38
  SPEAKER_01: Underwriting is something that gets a little bit, I would say, complicated, overcomplicated.
• 10:42
  SPEAKER_01: Underwriting is basically a big calculation.
• 10:44
  SPEAKER_01: You can do it in an Excel spreadsheet.
• 10:46
  SPEAKER_01: You do hit limits once you hit volume levels in Excel, but the underwriting is really simple once you have the data to underwrite with.
• 10:53
  SPEAKER_01: And once the data is set up, I think machine learning is the first step that you should take after you set up your database before you start to endeavor into artificial intelligence if you want to solve some of the more fundamental problems in the company, especially in real estate investing.
• 11:07
  SPEAKER_01: We started with valuations, so valuations of homes, valuations of the literal home value or the rental value of that home.
• 11:13
  SPEAKER_01: And what's great about being an analytics professional in real estate is that there's not that much data, relatively.
• 11:19
  SPEAKER_01: There are 150 million units in the country or take.
• 11:22
  SPEAKER_01: That's not a lot of data, 150 million rows.
• 11:24
  SPEAKER_01: So we asked ourselves questions like why wouldn't we just want to understand the value of every home in the country that we have information on?
• 11:30
  SPEAKER_01: Information exists.
• 11:31
  SPEAKER_01: You can buy it, you can go out and reach it.
• 11:33
  SPEAKER_01: And once you understand how to build a machine learning algorithm, that you can scale that comprehensively across the market.
• 11:39
  SPEAKER_01: There's going back to the homes that we were buying at that time, all the way to where we are today, we build an analog towards like Amazon or UPS or FedEx, right?
• 11:48
  SPEAKER_01: Amazon does what, four or five billion packages a year that are moving constantly, that have a lifespan that's pretty short, the delivery schedule.
• 11:54
  SPEAKER_01: Our homes are static, they don't move, and we don't see their half-life is decades, probably, that we're going to hopefully own these homes.
• 12:00
  SPEAKER_01: So if they can do it with the packages, we thought we should certainly be able to do it with real estate assets.
• 12:05
  SPEAKER_01: And so that's where we saw a lot of return for less of an investment.
• 12:09
  SPEAKER_01: And I think that most organizations would be rewarded for going in and investing more time in machine learning first.
• 12:16
  SPEAKER_01: And then using specific use cases for artificial intelligence.
• 12:19
  SPEAKER_01: We can get to that level where you bifurcate that between back office and operational AI and where ROI sits healthy.
• 12:25
  SPEAKER_01: But machine learning is where I would start specifically around questions around underwriting and valuations.
• 12:30
  SPEAKER_00: It's so interesting.
• 12:31
  SPEAKER_00: You know, a lot of the work we do here at Dwellsey is very machine learning oriented and has been for a lot of years.
• 12:36
  SPEAKER_00: But one of the things that feels very different about it is that I can start up Cloud or ChatGPT on my computer.
• 12:41
  SPEAKER_00: I, as a non-technical person in the industry, I don't know how to access machine learning.
• 12:47
  SPEAKER_00: Are there tools out there that would allow operating executives to be able to use machine learning?
• 12:52
  SPEAKER_01: Yeah, I'm privileged to have very talented people working on the team that uh focus on this.
• 12:56
  SPEAKER_01: And so I would say that it's not always as easy as firing up Cloud and saying, build me a machine learning algorithm because you don't know what's good or not good.
• 13:03
  SPEAKER_01: And I think that scalability, durability are things that talented individuals know.
• 13:08
  SPEAKER_01: And machine learning and artificial intelligence are two really great tools that just got added to our toolkit and are more democratized than ever.
• 13:15
  SPEAKER_01: But the best artists will be able to use those tools better than people that are not artists at all.
• 13:19
  SPEAKER_01: You may be able to take a paintbrush and paint.
• 13:21
  SPEAKER_01: It doesn't mean that you're gonna be a great artist with that paintbrush, right?
• 13:24
  SPEAKER_01: These tools are the same in the exact way.
• 13:25
  SPEAKER_01: You have to have great artists on the team, people that really understand not just mathematically or coding or the computer science behind it.
• 13:34
  SPEAKER_01: You have to take those field individuals, you have to teach them the business and they have to be interested in learning about the business.
• 13:38
  SPEAKER_01: And when they understand the problem that they're trying to solve, or maybe even problems that you haven't thought of yet as an executive or leader, they'll come with unique solutions by using these tools in ways that you can't imagine.
• 13:48
  SPEAKER_01: And I think that using like a clot is an interesting way to bounce ideas off of how would I solve this problem?
• 13:53
  SPEAKER_01: Give me four or five different ways to think about it and really test yourself versus the dogmas that you've probably come up under and experienced or trained under.
• 14:01
  SPEAKER_01: Maybe there are new ways to approach a problem.
• 14:03
  SPEAKER_01: And so we are a constraint team.
• 14:05
  SPEAKER_01: Our teams have always been basically challenged by great constraints.
• 14:08
  SPEAKER_01: The first one we gave you our own valuations and pricing.
• 14:10
  SPEAKER_01: Can we price our entire portfolio of 80,000 assets?
• 14:12
  SPEAKER_01: No, can you price all the 90 million single family units of the country?
• 14:16
  SPEAKER_01: That's a constraint.
• 14:17
  SPEAKER_01: And now all of a sudden, that 20,000 line algorithm that you build now has to be leaner and run more efficiently.
• 14:23
  SPEAKER_01: And we iterated and iterated and actually got that algorithm down to about 200 lines of code.
• 14:27
  SPEAKER_01: Wow.
• 14:27
  SPEAKER_01: It's really simple, runs really quickly, and it runs uh the right way all the time because it would have to at that skill.
• 14:33
  SPEAKER_01: And so that's those are the ways that I would think about rather than how do you reach machine learning?
• 14:37
  SPEAKER_01: It's like how do you connect with people that want to solve these problems and recruit them and retain them at your firm to use these tools to their maximum potential?
• 14:44
  SPEAKER_00: Yeah, no substitute for the people.
• 14:46
  SPEAKER_00: That's for sure.
• 14:47
  SPEAKER_00: Expertise that's always required, no matter how good the tools are.
• 14:51
  SPEAKER_00: That's 100% clear.
• 14:52
  SPEAKER_00: So as we move into this new era, you know, where do you think the opportunities are with AI?
• 14:58
  SPEAKER_00: Yeah.
• 14:58
  SPEAKER_01: So there's obviously the headlines, I think AI has probably been marketed about as poorly to people and consumers as any other unique uh technology iteration.
• 15:07
  SPEAKER_00: Yeah, you're telling me just here's so many people who are afraid, worried, concerned.
• 15:12
  SPEAKER_01: I don't know that leading with is going to take all your jobs is the best way to get people to adopt it.
• 15:16
  SPEAKER_01: I think that we have, yes, sure, some jobs could be replaced or augmented heavily with AI.
• 15:22
  SPEAKER_01: I think AI has proven to be good at task-level things.
• 15:25
  SPEAKER_01: I don't know that it's been job replacement level in a lot of areas yet.
• 15:28
  SPEAKER_01: And that's something that you have to understand of where it's good and where it's not good.
• 15:31
  SPEAKER_01: But I would say that any of us could, if we're being honest, could admit that in the grand scheme of all the scope of work that we do every day at our jobs, and specifically in real estate where I'm experienced in, there's a lot of things that we weren't doing yet.
• 15:42
  SPEAKER_01: We just don't have the time to do it.
• 15:44
  SPEAKER_01: Maybe they're not the top priority, and we have to have some sort of operating margin that looks attractive to shareholders and investors, right?
• 15:50
  SPEAKER_01: But with some of these tools, you can now do some of those things more efficiently.
• 15:54
  SPEAKER_01: And so it's about doing more with the same amount of potential overhead or spend, right?
• 15:59
  SPEAKER_01: Sure, you can make a little bit more of an investment if there's a healthy ROI.
• 16:02
  SPEAKER_01: But an example here, if we look back at our customer data, whether it's the calls that we record that they know are recorded for quality and assurance, right?
• 16:09
  SPEAKER_01: What do you do with those?
• 16:10
  SPEAKER_01: Are you going to put a human on that to listen to every call?
• 16:13
  SPEAKER_01: You can't do that.
• 16:13
  SPEAKER_01: It doesn't scale.
• 16:14
  SPEAKER_01: You can obviously put that, translate all those into text, transcribe them into text, rather.
• 16:19
  SPEAKER_01: You can then put them through a language model, measure for sentiment, measure what is the topic of that conversation.
• 16:25
  SPEAKER_01: And you should do that persistently.
• 16:26
  SPEAKER_01: That's incremental work that wasn't being done before, that's done in an economical way.
• 16:30
  SPEAKER_01: So now you can understand really what's happening in those conversations.
• 16:34
  SPEAKER_01: Did your representative do a good job?
• 16:36
  SPEAKER_01: Is the customer leaving that conversation happy or satisfied?
• 16:39
  SPEAKER_01: Are those conversations happening right before you're about to go, oh, uh send a renewal offer and potentially ask for more money when the customer is not happy right now?
• 16:47
  SPEAKER_01: So I think that we can take a lot of the interactions that we have, the information that we're gathering through the processes, and we can analyze it persistently now.
• 16:54
  SPEAKER_01: Instead of having someone go and do that, it's just happening in the background.
• 16:58
  SPEAKER_01: And a language model is a lot better at that than natural language processors used to be for those of us who had to use them before.
• 17:03
  SPEAKER_01: It's a much more scalable way to understand your customer.
• 17:07
  SPEAKER_01: There's a lot of interactions that you're having that weren't measured before.
• 17:09
  SPEAKER_01: And now, if you can persistently have a temperature on that customer, they're like uh 100 degrees here, running a little hot.
• 17:16
  SPEAKER_01: It's literally in Phoenix, we would have customers that, you know, with enough homes, this would happen sporadically where AC goes out in the summer, they've got three kids at home, you got to get there fast, right?
• 17:25
  SPEAKER_01: And solve that problem.
• 17:26
  SPEAKER_01: Maybe don't send a new letter without a phone call attached to it saying we're gonna solve that problem, but we also have to send you this notice of renewal because of the time constraints we have there.
• 17:35
  SPEAKER_01: So I think that's where the first use cases of AI should be customer-centric in my way, because that is the hardest thing to solve in just about any business.
• 17:43
  SPEAKER_01: And the more you can understand what your customers really want, then the more harmed you'll be to solve those problems.
• 17:48
  SPEAKER_00: I think that's fantastic insight.
• 17:49
  SPEAKER_00: You think about the nature of customer relationships in an older era, it's a one-to-one thing.
• 17:54
  SPEAKER_00: An individual landlord owns one or two units, they have a personal relationship with that renter, they know not to send a renewal letter with a 5% increase, you know, on the heels of the air conditioning failing.
• 18:03
  SPEAKER_00: That's incredibly obvious.
• 18:05
  SPEAKER_00: But once you scale up to 80,000, 90,000, 100,000 units and you've got different people dealing with every piece of that puzzle, that is it's impossible to do without some sort of AI or something like that, uh, helping sort through all the information and put it in the right places for the right people.
• 18:21
  SPEAKER_00: Those instances are critical.
• 18:22
  SPEAKER_01: We wanted to talk about yes, we have this many thousands of homes, but we should treat it as every home is equally as important as the next.
• 18:29
  SPEAKER_01: And nothing matters more to customer like that value in rental real estate than a longer term stay.
• 18:35
  SPEAKER_01: So you're talking about a transaction to lease a home, and the next transaction is renewing.
• 18:40
  SPEAKER_01: And every renewal you can get, it's on average about 12 more months of rental income at a certain margin.
• 18:46
  SPEAKER_01: And so in single family, specifically where we spend, you need at least one renewal before you get the break even with NOI.
• 18:51
  SPEAKER_01: You're not going to make money unless you get at least one renewal and you get closer to 18 or 20 months.
• 18:56
  SPEAKER_01: And so obviously you want more time than that.
• 18:58
  SPEAKER_01: I think the average industry is about three or four years in single family.
• 19:01
  SPEAKER_01: But if you were out to five to six, six to seven years because customers were happy and they were choosing to stay with you because you didn't trip up at one of these interactions, that's incredibly meaningful.
• 19:10
  SPEAKER_00: There's no greater ROI than that in the rental real estate business.
• 19:13
  SPEAKER_00: Yeah.
• 19:13
  SPEAKER_00: So I'm so curious.
• 19:15
  SPEAKER_00: I got in a debate with somebody over the weekend as to what their heuristic is for vacancy and what it costs defined by months of rent.
• 19:24
  SPEAKER_00: What's your shorthand?
• 19:26
  SPEAKER_01: I don't think that there's a mathematical equation that says 95% occupancy is better than 100% occupancy, so long as you're pursuing relatively close to market rate, right?
• 19:35
  SPEAKER_01: So we understand the relationship.
• 19:36
  SPEAKER_01: If you're charging market rate, the home's going to turn over a certain amount of time, or they're going to take a certain amount of days on market to rent.
• 19:42
  SPEAKER_01: If you're driving a discount to that, you're probably going to reduce days on market, or you're going to draw longer retention periods.
• 19:47
  SPEAKER_01: The moment pops have been doing this for decades.
• 19:49
  SPEAKER_01: They are more likely to hold your rent flat for several years and then mark to market when the home turns over.
• 19:54
  SPEAKER_01: I think that we should always be pursuing, knock those dogmas out and say 95%, 96% is not more right or wrong.
• 20:01
  SPEAKER_01: It depends on everything else that you're doing.
• 20:03
  SPEAKER_01: Are you getting the market rate?
• 20:04
  SPEAKER_01: Charging more than market does not help anything.
• 20:07
  SPEAKER_01: We've determined that mathematically.
• 20:08
  SPEAKER_01: It only creates more problems with the customer.
• 20:10
  SPEAKER_01: It only leads to more turnover, more expensive management.
• 20:13
  SPEAKER_01: And so you should always be trying to solve for probably market or a bit of a discount to it because vacancy is very expensive.
• 20:20
  SPEAKER_01: Where we were at an invitation about 80 bucks a day, you could do the math.
• 20:23
  SPEAKER_01: And so why am I chasing another 10 or 20 bucks of rent per month?
• 20:27
  SPEAKER_01: It's costing me$80 a day for two or three or four more weeks to go and pursue someone to pay that.
• 20:32
  SPEAKER_01: And so there's always a balance between the rate and occupancy.
• 20:35
  SPEAKER_01: Vacancy is very expensive.
• 20:37
  SPEAKER_01: And with some of these tools, I believe that we've we've proven in times of great demand and low supply, like we saw after 2020, especially in the single family space.
• 20:46
  SPEAKER_01: You saw you can track uh with this homes public results, you can track them in AMH, everybody's occupancies went way up.
• 20:52
  SPEAKER_01: And all of a sudden they're floating at 97, 98% occupancy.
• 20:56
  SPEAKER_01: And that's not because we're, you know, just giving huge discounts away.
• 20:59
  SPEAKER_01: It's because, again, the market was yielding that.
• 21:01
  SPEAKER_01: But our efficiencies went through the roof as far as how well we were able to serve the customer.
• 21:07
  SPEAKER_01: We're getting less phone calls during vacant period because there's half the amount of vacancy in the portfolio.
• 21:11
  SPEAKER_01: Our turn times were shrinking because we had half the amount of turns to do, because we were 95% occupied and more 97.5 or 98.
• 21:18
  SPEAKER_01: And so with the same amount of overhead, we were able to do a better job on less work.
• 21:22
  SPEAKER_01: That's pretty intuitive.
• 21:24
  SPEAKER_01: Uh, but I think we proved out that with less vacancy, uh, you can do probably a net better job if you have the same amount of human capital or resources to do with.
• 21:33
  SPEAKER_01: And the question is do you need more resources or less resources?
• 21:36
  SPEAKER_01: I think that what we saw was a very uh healthy business when the balance was struck.
• 21:40
  SPEAKER_01: I don't understand what people say, oh, your occupancy is too high.
• 21:43
  SPEAKER_01: It's like, well, if you're giving rent away to get higher occupancy, sure.
• 21:46
  SPEAKER_01: But if you're managing your rent the right level, you should always be driving towards higher occupancy and keeping residents in the homes.
• 21:53
  SPEAKER_00: Yeah, no, that makes a ton of sense to me.
• 21:55
  SPEAKER_00: Gets me thinking about revenue management.
• 21:56
  SPEAKER_00: You know, you ran revenue management on three billion dollars in revenue, not very many.
• 22:00
  SPEAKER_00: People have had that experience.
• 22:01
  SPEAKER_00: Would you let AI into that picture?
• 22:03
  SPEAKER_00: How would you run that today?
• 22:05
  SPEAKER_01: Yeah.
• 22:05
  SPEAKER_01: So we in revenue management, the most important variable that you had to know is where was the market for your home?
• 22:11
  SPEAKER_01: And we had to dispel some of this.
• 22:12
  SPEAKER_01: As smart as we are, as much data as we had, as the algorithms were very accurate at a high level or on the average, you have to admit to yourself that no one knows the value of a rental home more than the person renting it.
• 22:23
  SPEAKER_01: They can open up apps like Zillow or the other ILS apps, Dwellsy, and do their research.
• 22:28
  SPEAKER_01: And they can see what else is on the market.
• 22:29
  SPEAKER_01: And they can spend hours researching, looking at pictures, those kinds of things.
• 22:33
  SPEAKER_01: And so while we can understand what beds, bath, square foot location looks like, where AI can come and do it is, well, are we looking at the pictures?
• 22:40
  SPEAKER_01: Do we know what our house looks like compared to what the other houses look like on the market?
• 22:44
  SPEAKER_01: And so AI is a way that you can somewhat economically, it's still a little expensive today to analyze pictures because they're not standard on the ILS channels most of the time, is just get an idea for what kind of uh shape the house is in, and because that that does drive a premium or a discount to the the other comps in the market is what do your fit and finishes look like?
• 23:04
  SPEAKER_01: What are the appointments in that home?
• 23:05
  SPEAKER_01: How bright is it?
• 23:06
  SPEAKER_01: You know, what's playing well in that market?
• 23:08
  SPEAKER_01: Is it bright white kitchens?
• 23:09
  SPEAKER_01: Is it you know kitchens that are a little bit more dark, hard surface flooring areas?
• 23:13
  SPEAKER_01: So all this you can understand through pictures out there.
• 23:15
  SPEAKER_01: I think that's where AI can come in and give you a little bit more context and not just, again, the big variables that will mostly drive accuracy and home coping at an average level, which is bed bath, square foot, and most of all location, but there's more context to be had there.
• 23:29
  SPEAKER_01: And customers have that opinion as humans.
• 23:31
  SPEAKER_01: They can look at pictures and form that opinion.
• 23:33
  SPEAKER_01: And you should be afforded that information as well if you can.
• 23:36
  SPEAKER_00: Yeah.
• 23:36
  SPEAKER_00: Thanks, actually.
• 23:37
  SPEAKER_00: You know, use of pictures and revenue management, I think, has huge potential.
• 23:40
  SPEAKER_00: There, I think the SFR players have a huge advantage because they've actually imaged every unit.
• 23:44
  SPEAKER_00: And the multifamily folks usually have not.
• 23:46
  SPEAKER_00: They've got beautiful pictures of the amenities and the model unit, but they don't know what unit 2C looks like at all.
• 23:53
  SPEAKER_00: And it might be incredibly divergent, even just because of the way the light falls on that unit and the experience somebody gets in there as a result of that.
• 24:00
  SPEAKER_00: That's right.
• 24:01
  SPEAKER_01: They have an advantage in there's less four plans.
• 24:03
  SPEAKER_01: Uh, so there's only a few per property, and so they can kind of rinse and repeat and have a model to show off where single family almost every home is a little bit different.
• 24:10
  SPEAKER_01: But you're right, there's a lot of images out there as the models get more efficient analyzing those images that are non-standard, right?
• 24:16
  SPEAKER_01: If you have a perfect picture taking panoramics, it's really clear what's available to you there, and you can analyze it efficiently.
• 24:22
  SPEAKER_01: Training on you know, pictures that were taken maybe on a old BlackBerry or an older phone or the lighting's worse, right?
• 24:29
  SPEAKER_01: It can be um less accurate, but understanding things like is there stainless steel appliances inside there?
• 24:34
  SPEAKER_01: Is there a is there an actual backsplash and and what drives what in different markets?
• 24:39
  SPEAKER_01: So in certain markets, this is important.
• 24:41
  SPEAKER_01: In other markets, this is really important.
• 24:43
  SPEAKER_01: And so that kind of specificity is what's on the frontier.
• 24:46
  SPEAKER_00: Yeah, absolutely.
• 24:47
  SPEAKER_00: A couple of quick questions for you, Gar.
• 24:49
  SPEAKER_00: First, you know, thinking about folks getting into AI, where would you advise somebody to start if somebody is in the industry AI curious but hasn't tried anything yet?
• 24:58
  SPEAKER_00: Yeah, yeah.
• 24:59
  SPEAKER_00: Tread lightly, right?
• 25:00
  SPEAKER_01: You can always spend more money if you need to, but you can't take it back once you've spent it.
• 25:04
  SPEAKER_01: So I think that we're in a really interesting position today on the timeline of what AI has, where it's been and where it's going.
• 25:11
  SPEAKER_01: I think that I've got a personal opinion that open models are probably gonna be a big part of the future in two or three years.
• 25:17
  SPEAKER_01: I think that the vast majority of use cases will be solved with open models either locally or on-prem, and you'll be able to scale that out and not have to worry about what the tokens are going to cost in two or three years when they're maybe not subsidized anymore, right?
• 25:28
  SPEAKER_01: That's a cost that every CEFO and CEO is worried about, is we've committed to this model, this approach, this infrastructure that we built around.
• 25:36
  SPEAKER_01: And what if the cost of these tokens goes up and we're beholden to that and we can't pivot away?
• 25:41
  SPEAKER_01: And so I think that some of these models have gotten really good really fast, and the incrementality of every iteration is less and less material to most of the problems being solved.
• 25:50
  SPEAKER_01: In real estate, this is not a typically complicated business.
• 25:54
  SPEAKER_01: It's not an easy business, but it's not complicated.
• 25:56
  SPEAKER_01: We're not trying to trick answer, right?
• 25:58
  SPEAKER_01: We're not trying to catch rockets out of the sky.
• 26:00
  SPEAKER_01: We're trying to serve a residence and make sure these homes are available and you know, occupied.
• 26:04
  SPEAKER_01: And so I would say that most of the models that are out today are good enough, more than good enough.
• 26:10
  SPEAKER_01: And that's a decision that's got to be strategic within the business.
• 26:12
  SPEAKER_01: And then how do you want to deploy them?
• 26:14
  SPEAKER_01: How do you want to support them?
• 26:15
  SPEAKER_01: And everybody should want to drive some level of positive ROI.
• 26:19
  SPEAKER_01: And right now, I think there's challenges to that.
• 26:21
  SPEAKER_01: And we've seen the research that says most of these endeavors are probably not net positive yet.
• 26:26
  SPEAKER_01: And so you either have to reap more benefit or you have to do it more efficiently.
• 26:31
  SPEAKER_01: I would start with efficiency.
• 26:32
  SPEAKER_01: Make sure that the problem you're pointing at is vast, that it needs to scale.
• 26:36
  SPEAKER_01: It's a big problem.
• 26:37
  SPEAKER_01: And uh, it's not something that you're doing today.
• 26:40
  SPEAKER_01: I would not advise anybody to take a human out of the equation unless this is really a job that a computer does better and the human's or your employees' life gets better as a result of this.
• 26:51
  SPEAKER_01: And you can put them on more human problems where we need more of their time and bandwidth.
• 26:55
  SPEAKER_01: That's where I would start first.
• 26:57
  SPEAKER_00: Yeah.
• 26:57
  SPEAKER_00: No, it makes sense.
• 26:58
  SPEAKER_00: I'm seeing, you know, a lot of folks getting, you know, basically extensions of the team members they've got.
• 27:02
  SPEAKER_00: You know, has anybody ever had enough development capabilities?
• 27:05
  SPEAKER_00: Has anybody ever had enough IT capabilities or marketing capabilities?
• 27:09
  SPEAKER_00: Like we always want more, we always want better of everything.
• 27:12
  SPEAKER_00: So how can the AI help the people involved get there?
• 27:16
  SPEAKER_00: Yes.
• 27:17
  SPEAKER_01: And the biggest challenge for technical leaders, especially in real estate, is and I've said this out loud many times, the real estate business is typically not nice to CTOs or CIOs in the business.
• 27:27
  SPEAKER_01: They're coming into a room.
• 27:28
  SPEAKER_01: Most of the time, they didn't start the business, right?
• 27:30
  SPEAKER_01: This started from somebody who has an investment background or just a real estate background in general.
• 27:35
  SPEAKER_01: So they know more about the business than the technical leader does.
• 27:37
  SPEAKER_01: And they have their generals, whether SVPs or other C-suiteers, that have been in this business for a long time as well.
• 27:43
  SPEAKER_01: And they have opinions on what the technology platform needs to do.
• 27:47
  SPEAKER_01: And usually there's several different leaders in the room over finance, over operations, over investments, asset management.
• 27:53
  SPEAKER_01: They don't really agree, believe it or not.
• 27:55
  SPEAKER_01: And so, how is this technical leader going to solve all the problems if they're being pulled in different directions, especially with some of these problems requiring real investment?
• 28:03
  SPEAKER_01: AI is a real investment.
• 28:05
  SPEAKER_01: And so I feel sympathy.
• 28:06
  SPEAKER_01: I've never been a technology leader in the business.
• 28:08
  SPEAKER_01: I feel sympathetic for these individuals that are leading, but I would challenge everybody on the business side in real estate to make sure that you're taking a step back and thinking about the total stack of the platform, how your part dovetails in with the other parts, and we can get a unified vision of where we're going with an understanding of where we are today and how do we bridge the gap?
• 28:28
  SPEAKER_01: And until we do that, I think the real estate business is always going to lag around those businesses that are maybe more technical in nature now, e-commerce, the digital space, right?
• 28:38
  SPEAKER_01: They've always led with technology and then figured out how to operate it around that.
• 28:43
  SPEAKER_01: And I think that there's a litmus test out there.
• 28:45
  SPEAKER_01: One famous one is Amazon.
• 28:47
  SPEAKER_01: How did they just pick up a grocery store chain and keep moving?
• 28:50
  SPEAKER_01: I'm sure it was harder than it looked to all of us.
• 28:52
  SPEAKER_01: That's a pretty big sidestep.
• 28:54
  SPEAKER_01: I don't know that real estate platforms can make that kind of sidestep into, let's say, a different asset class.
• 28:58
  SPEAKER_01: We don't have a big, large multifamily and single family operator yet.
• 29:02
  SPEAKER_01: Uh, someday we will, but it's gonna come down to can your platform uh handle that?
• 29:06
  SPEAKER_01: Because you don't want to run dual platforms all the time.
• 29:09
  SPEAKER_01: And so no one's there quite yet.
• 29:11
  SPEAKER_01: I'm sure we'll see it, but that's that's some of the challenges that are facing the investment related to it and can you execute it?
• 29:16
  SPEAKER_01: It's gonna come down to can the team coalesce.
• 29:19
  SPEAKER_00: Yeah, yeah.
• 29:19
  SPEAKER_00: No, this can be an operationally very complex business.
• 29:23
  SPEAKER_00: And I do agree the CTOs often get the brunt of that, and unfortunately, it can be a real challenge to manage as a result of that.
• 29:29
  SPEAKER_00: So last question for you, Garrett.
• 29:31
  SPEAKER_00: How are you using AI in your personal life?
• 29:33
  SPEAKER_01: Yeah, personally, uh I'm trying to make sure AI is is uh challenging me in a certain way.
• 29:38
  SPEAKER_01: So I almost treat it as a contrarian um, you know, in my life.
• 29:41
  SPEAKER_01: And so I'm I'm always everybody who's got ideas, they think they're probably they're a little biased about their own ideas.
• 29:46
  SPEAKER_01: And so I like to challenge and make sure, like, tell me where I'm wrong, tell me where I'm right, I guess maybe, but like really come up with an additional opportunity or or tell me all the reasons why this is gonna fail.
• 29:56
  SPEAKER_01: And so I think that that helps you strengthen your idea before you you bring it anywhere, whether it's to your boss or it's to your wife or your partner, whoever it is.
• 30:04
  SPEAKER_01: It's like, I've got an opinion here.
• 30:05
  SPEAKER_01: Where's my opinion wrong?
• 30:06
  SPEAKER_01: And give me some examples as to why.
• 30:08
  SPEAKER_01: And I think that's that sped up some of the processing on my side as to getting to the right answer, whether it's personally or professionally.
• 30:14
  SPEAKER_01: So it's always helpful, but understand that AI is uh at its core just a prediction model on what's been said already and it's trying to predict the next word.
• 30:22
  SPEAKER_01: So it's not a no-all and all be all.
• 30:24
  SPEAKER_01: It doesn't have psychic powers or anything.
• 30:26
  SPEAKER_01: So be careful about what you listen to it because it as these models were trained.
• 30:30
  SPEAKER_01: We all know they were trained to tell us what we want to hear, so we keep using them.
• 30:34
  SPEAKER_01: So it can become a bit of a sycophantic exercise.
• 30:37
  SPEAKER_01: And I think we've all seen that come to light now.
• 30:39
  SPEAKER_01: So use it to challenge yourself, but uh don't just listen to it blindly, of course.
• 30:44
  SPEAKER_00: Yeah, I think of one of our earlier guests, uh Adam Siegel, uh, who's on the pod a few weeks back, talking about how AI tells him that he's the smartest man in history.
• 30:52
  SPEAKER_00: And uh it's easy to believe that.
• 30:54
  SPEAKER_00: So I I love the use case that you've got there.
• 30:56
  SPEAKER_00: Uh, I think that's fantastic to use it as uh somebody who can help keep you honest, if you will.
• 31:00
  SPEAKER_00: We all need people like that or machines like that in our lives.
• 31:03
  SPEAKER_00: Yeah, yeah, of course.
• 31:04
  SPEAKER_00: And hopefully it's not too much smarter than us soon.
• 31:06
  SPEAKER_00: But uh we'll see where that goes.
• 31:08
  SPEAKER_00: Yeah, we'll see where that goes.
• 31:09
  SPEAKER_00: It's gonna be interesting one way or the other.
• 31:11
  SPEAKER_00: Yeah, for sure.
• 31:12
  SPEAKER_00: Garrett, it's been such a pleasure having you on Net Effective.
• 31:14
  SPEAKER_00: Thank you so much for joining us.
• 31:15
  SPEAKER_00: Really appreciate you coming on.
• 31:17
  SPEAKER_01: You bet.
• 31:17
  SPEAKER_01: Thanks, Jonas.
• 31:18
  SPEAKER_01: It was a joy.
• 31:19
  SPEAKER_00: That's Net Effective, new episodes weekly.
• 31:22
  SPEAKER_00: If you're running rentals and figuring out AI, hit like, subscribe, follow, or whatever your app makes you do to get more of the show.
• 31:28
  SPEAKER_00: And if you'd like to come on as a guest or subscribe to the weekly updates, go to neteffective.show.
• 31:33
  SPEAKER_00: I'm Jonas Porto.
• 31:35
  SPEAKER_00: Thanks for listening.