Inside AirDNA Adapt: Building the Future of Revenue Management

Welcome to the STR data lab. Hello and welcome to the STR data lab. I'm Jamie Lane, chief economist at ARDNA, and I'm joined today by two guests. Uh actually my good friends and colleagues at ARDNA. We've got Mark Marino, uh, head of data science, data science manager at AirDNA, and we've got Jordan Meyers, the senior product manager at ARDNA. Guys, welcome to the show.

Thanks for having us, Jamie. Yeah, thank you for having us. Yeah. So, we were all just in Denver uh uh celebrating uh what has been an exciting last but like half year 12 months of of building um AirDNA Adapt. And we're here to talk about not necessarily the feature functionality of it because I think that'd actually be a pretty boring podcast. Uh that may be uh better for a live demo. Uh but uh what I do want to talk about uh is sort of what it took um and with two guys that and you were leading the teams that were really building a large portion of what it is. So for those of you that are regular listeners uh we don't t typically talk about AirDNA product. Uh this one is going to be more about uh what it took to build a brand new product in today's ecosystem. Uh so if you're building today uh using a lot of the AI tools, you may find this interesting. Uh if you're just interested in like how someone goes about building a revenue management dynamic pricing tool from scratch, uh that is what we're going to talk about.

Um and uh and dig in some of the trials and tribulations of product development today. So Jordan, I see you smiling. Uh [laughter] and maybe just start uh with a question for both of you is and what's been maybe the the hardest part of the past 12 months in terms of uh this sort of product development journey maybe not and what's the best and what what's what's been the hardest part.

Well Jamie I'm smiling because this has been uh quite the challenge for the team. Um, it building the age of AI is totally wild. I'm sure we'll get more into that. But what has been the hardest thing is really tackling the scale of what we're trying to build with AirDNA adapt. We are trying to build a holistic revenue management system for folks. And not not only for uh the smaller operators. Maybe you have one, two doors, you're an owner operator. Maybe you're a growing property management company or you're at scale with thousands of listings. And so trying to tackle all of that and build the right tool that can work for everyone has been a huge challenge. Really exciting though.

Yeah. How would you answer that question, Mark?

Uh the data like getting the right getting the data to a right place. Um more so like we're talking about 12 million active listings across three channels uh where there's a lot of noise and so you could have the best model in the in the industry but garbage in garbage out, right? So making sure that you have like that data polish that the feature engineering and everything that goes into a model is as clean as possible was a big challenge like it it has been a big undertaking and I feel like we reach a good place. Uh it took us a while but we reach a good place.

Yeah. So Jordan part of the sort of leadup to building this was you and Luke and others uh doing hundreds if not a thousand like operator interviews. uh some of those deep dives, some of those uh sort of and and feedback sort of questionnaires uh and interested in maybe what you would expect or what you expected to hear in those conver conversations versus what sort of rain through all those conversations.

Yeah, I was amazed to find out I I think the number was like 43% of the people we talked to don't use any kind of dynamic pricing. uh they do it manually whether they're going into their their property management system, they're going to Airbnb or they have a spreadsheet. Like we're still in the world where people are managing their business off of a spreadsheet. And that I think was the wildest uh thing to me. I did not expect that. But but talking to people, there was a a common thread whether you used a pricing tool or you didn't. There was a common thread of of frustration. frustration, doubt, this like lack of trust that existed because those folks who weren't using the tool or were using the tool, they had some confusion around, well, how do I best optimize this tool, right? There's a lot of buttons, there's a lot of levers and knobs and and dials that I have to that I have to tweak in order to make sure I'm really getting what I want out of out of this uh out of this revenue management tool.

Uh, and there was doubt in the data and the recommendation. It was well, why this number, right? and and a lot of a lack of clarity around how did you arrive to this number? It's it's been just a receipt of plus 3%, minus 4%, well, what do these numbers mean? You know, how did you really arrive at this? And so that was kind of the the clear line of feedback that we got.

Yeah. Mark when you heard that of like the problems trust not really the trust in the number not necessarily the number and that you were getting back did that change how you thought about it as a data science problem like and you could get to the best rate possible and but if you're not able to I'm sort of build trust behind it for the user like ultimately and it wouldn't sort of resonate. So how did you think about that when going in and and starting to develop the models?

It changed a lot uh because that basically like that explanability and having to be able to build that trust discards any model that can you cannot explain. So for us basically that meant that we had to discard any blackbox model. Uh we went from have like considering these options that were very sophisticated, very complex but had very little explanability. So all of those were out of the door. we had to go for more simplistic approaches that still yielded good rates because that part is still important, right? Like you still have to get to those good rates, but um now I feel confident that we've gotten to a point where like we have those uh very strong rates that we can explain thankfully.

Yeah. and and maybe for both of you um like as you've sort of dug into the existing tools that are out there into understanding sort of customer needs like do you think it's like incremental advancement in pricing tools or do you think fundamentally like something different needed to be built for the industry? So,

oh my goodness.

Yeah, good good question. And I'm gonna take the cop out product manager answer and say like it depends or it's both.

Um I I think that there are small things in that in that explanability that are important and don't require huge swings in technological advancement or innovation. Um but I think if you are going to build something for 1 to 10,000 listings that requires that that uh innovation that requires building something totally new and so I think the approach we've been taking is can we build something that's a little bit better to start and think about the infrastructure and then how we leapfrog uh technologically. Yeah, you Mark

for me was focusing on like what are some of the like based on the customer feedback that we had gotten like how we could get to like leap frog everyone. I remember hearing a lot about it's not that uh the rates are bad, it's just that I have to manually go in and change a lot of things. And so working with Jordan on like providing like those rates that are good that they don't have to go in and manually change a lot the explanability behind it and also uh trying one that we came up a lot during the interviews that uh Jordan probably agrees is uh events like I cannot tell you how many times we uh people kept mentioning events. So like we were trying to gather all of those thoughts and just make sure that we accommodated them because we wanted to reach out as many people as possible.

Yeah, maybe let's let's jump into that because events is one of those things that I we had as a line item at the beginning like oh yeah we need to figure out events like that seems like something we should we should have. What what actually went into figuring out events? Like was it an easy problem? Was it a hard problem? like is an ongoing problem like what's up with events? It's a hard problem and it's ongoing but we're still working. Like I I think uh we have one senior data scientist that is working on it and I think her life now revolves around events. Like she wakes up and all she sees are events. But uh basically the hard problem was that a lot of people brought this um point and it's that many of these tools are reactive to events, right? They wait for the demand signal to show up to then like start pricing accordingly. And we wanted we wanted to avoid that because yeah like anyone can do that like once the demand signal picks up it's very easy for me to tell you hey there's something going on in your city and so the hard problem was like how which signals can we capture to be proactive about it and it was a lot of like brainstorming with you Jamie and the rest of the team and like in the end we ended up in a very good place where like we are capturing other signals that are not exclusively short-term rental data to be able to capture events proactively and make sure that when you're pricing You're pricing when people are booking and not two weeks late.

Yeah. So, if Taylor Swift announced a concert today in Atlanta, how quickly do you would you pick up on that event?

Uh, pretty soon at most in a week. So, like I feel pretty good about it.

Yeah. I one of the debates I know we had going in or or maybe throughout was I mean what things were sort of more a UI problem versus a model problem. Um, so we talk about sort of model explanability and sort of rate explanability. Do you think that is a a data science problem or a UI problem?

I think it's an explanability problem more than a model problem because again like we have 12 million listings. We have all of these data and we have like your historic data and we have all of this and based on all of that we're going to make a suggestion but then there's a psychology behind it that a model cannot capture. there's like uh we think that you are uh midscale uh listing and then the person comes in and they're like I disagree I think I'm upscale or luxury so you need to bump my rates or the other way around right and so rates like it's one part of it but like I think that the user experience and psychology behind it is a big one

how do you feel about that Jordan

I totally agree I mean I'm still going to forward all of the uh anything that that comes in from users around feedback about the model but uh agree ree that it's it's an explanability and a a confidencebuilding exercise more than anything. I think that's what that's what we're here to do is to make people feel confident in the decision that they're making even if it's not the one that we or our model recommended. And so like to the point of of comp set or price tier, it's we can give you a recommendation and try to explain, hey, here's how we came up with this decision, but we have to make sure the user feels comfortable. Maybe they want to change that price tier. Maybe they want to change their comp set. And so we should make it easy for them to do that so that they feel confident in whatever rate is uh is going out there.

Yeah. So Mark, when we were talking earlier, you said like because explanability was such a a big uh feature that it sort of threw out potential blackbox models that we may have used. Do you think that forced us to make tradeoffs in terms of model performance because I mean we sort of threw out all these potential models or do you think ultimately we got to a similar point?

I think ultimately we got to a similar point just with a very different route. I think that in the end we build like because we didn't just use one single model, right? Like we use an assemble of models to try to get to the same answer. It just took us a bit longer because like the blackbox model can do it all in one. Like the blackbox model gets to the same answer. It just does it very fast. And the problem is that you don't know what's behind it. But by building the ensemble of models, I do believe we got to the exact same or like a very similar answer. It just took a bit more time and a bit more thought and like how we treat the data and like everything like the feature engineering part. I think it's was key.

Yeah. So, so pretend you're explaining this to my mom. like how would you sort of walk her through like how you get to a a rate? Like I don't need we don't need to get to like the the deep dive of like models and things like that, but like helping her sort of build confidence like what would you say is sort of going into an in any given set of rates that we're putting out there? So I would tell her that when into this model we take into account both like uh signals of every single thing property that is around you that we think is a compatible property and then also market signals and then all that the model does and like when I talk about feature engineering fancy word is just like basically like breaking down the data in a way that we are giving strong signals to a model. So, we're just trying to uh say, hey, in this market, there's a booking window that usually happens 90 days away. So, that basically means that 90 days away, your price should be at a premium, and you should be trying to get as many bookings as possible.

Right after that, you should start discounting heavily because after that, your prime booking window uh has been missed, and your probability of getting booked is way smaller. The and the other way around, right? Like if uh your the booking window of a market is like 90 days out to 120 days out, if you're beyond those 120 days, like what we don't want to do is like price you way too uh way too low because if you're getting too many bookings outside of that window, it basically means we price you wrong. So when I talk about the feature engineering and getting the data right, it basically means that we the model all it does is read a story and the story that we're telling is with the data that we have and we just have to put in a clean format so the model understands it and then gives you a good rate.

Yeah. So if we're pricing in potentially in every listing in every single market around the world and a wide variety of property types, a wide variety of and and different listings, what do you think is the hardest thing that the model is still getting wrong?

Markets where there are a lot of unique properties. Mhm.

unique properties like I and this is something that like has happened even when we had uh our previous dynamic pricing model. Uh I remember a client came in saying that they had a castle in the middle of France with 20 bedrooms, 12 bathrooms and I'm like how how am I supposed to price that? Like so those properties will always be problematic because like it's it's like we don't have it in the like it's barely maybe there's one or two properties like that worldwide. So like it is very hard to place those. So I think that is one that is a challenge that we're still trying to figure out.

Yeah. Uh Jordan, thinking about sort of this process and and over the past year, I'm interested in from you like what are some of the ideas? Like I know there's lots of things we've killed and lots of things that are still ongoing, but what is like one thing that you had conviction in from the beginning that survived like all the cuts, all the betas, all the like all the iteration around the product that I and uh yeah, you had conviction in from the beginning.

Wow. Um only one thing now. Um, [laughter] if there's one thing I I had and still have conviction for, it is the listing setup is walking users through the recommendations that we're making across all these like key aspects, right? The the base rate. Okay, what are we recommending? What does that mean? How do we get there? How can we help you choose the base rate that's right for you? Because that's where everything else that's where all these other rates stem from. Looking at what is your min and max? Can we walk you through that step? Choose something that makes sense for you. can walk you through a a strategy or do none of the strategies we created make sense to you and you want to chat with our AI agent and build your own. We are like we are releasing that.

And so I have a a really high degree of conviction that that we need to before the user even toggles on their rates. We should make them feel comfortable, make them feel confident in the in the rates that they're seeing and the decisions that they're making that we're making uh through our model. And so even though I've had a high degree of conviction, I've also been wrong about how we do that. Uh but that's the beauty of of you know building in the age of AI and being able to go through a beta is that I I pushed out a lot of ideas and I had a lot of strong opinions. Some of them didn't work. But the beauty of where we're at and how quickly we're able to build is that we've brought it through different phases and I'm really excited for for the next one. And I'm sure there will be things that are wrong there too, but we'll just continue to improve it.

And so, yeah, I I think I think it's that it's that that setup, that onboarding of a listing, uh, I think is just incredibly important.

Can you give an example of something that you were just flat out wrong on?

Yes, Jamie, I can. Um I so I took some let's call it inspiration uh from some of the other tools out there and we were getting some feedback that uh or or we were see we were noticing that someone would click into a listing after they connected they would just see a rate chart or see their calendar they see the rates they say no these are bad you know too high too low there was no like Goldilocks moment and so I thought okay before we even show them the rates we have to make them uh choose a base rate, right? Like let's put them through. Let's really get them bought in and confident in what they're seeing. And they can't see their rates. They can't see the calendar. They can't see anything else uh until they make this decision because it's that important.

Boy, was I wrong. And that was a frustrating experience for users because while I do think it's important that you adjust your base rate or you make sure that it it feels right, make sure you feel confident in it before syncing rates, putting that kind of technical question in front of someone before they can even explore the tool was just the wrong thing. It it was Yeah, I was just uh I was off. The instincts were were correct, but the the execution had to be had to be improved. And now we've we've uh revamped that.

Yeah. And and the other thing and you did talk about like the conversations with users and how helpful there was and are there any like specific pieces of feedback that and was sort of instrumental maybe in and some of the ways that and the product got shaped?

Yeah, absolutely. Uh I think the biggest thing and and this wouldn't have happened if we were building this two years ago is how people talked about the fact they wanted to chat with their data. They wanted to set recurring tasks. Uh they wanted to be able to take action on their pricing, their revenue management strategy from wherever they work. And that's become codeex uh claude co-work. become uh homebuilt tools using agentic engineering. Uh that's become I have a tech team and they've built me this uh this HQ right where I'm analyzing all this data and this is where I want to make all of my uh pricing and availability decisions. And so what that meant was that we shouldn't or didn't have to focus on all the bells and whistles in the UI. Like we didn't need to build every screen. We didn't need to build every the ability to see every single you know data point in a particular way because what we really need to focus on was the infrastructure and making the data and action and operations accessible from wherever they want to work because I think that that's now and that's going to be the future is the the data is is the product and and building something that makes it accessible uh for these agents is imperative.

Yeah. And having listened to a lot of these user conversations, I feel like the AI conversation was one that went back and forth of some people really wanted AI forward product, other people just like as soon as you say AI, they have true distrust in anything you're talking about. Um, so as a as a product manager, I mean, what was your system for sort of distilling those conversations of sort of competing feedback in terms of and then what you would suggest that we go out and build?

Yeah. I I think it's about finding the the themes because everyone has a different idea. Everyone wants a slightly different product, different way of viewing things. And so I think the the point is to distill it down to like what's important. Um, and the word I keep coming back to is is confidence. And so, how do we build confidence in that decision became my guiding light for everything else that we built? Because some folks want to, to put it bluntly, just be told what to do with their rates on any given night. They want an agent to send text to their phone and say, "Hey, change these rates." Okay, that that that's fine. Like that that's one way of doing it. But then there are other people who say,"I just want the access to the data and I want to be able to analyze it myself, have my agents analyze it because I've built these skills. I've built all this infrastructure." And so I think it's about I just keep coming back to that that infrastructure piece and providing the data the explanability that we that we brought up earlier like the explanability is the is the huge part because then the folks who do just want to be told what to do well great they can go into claude and they they can build a skill that's going to give them a daily digest and say you should do x y and z and for the folks who want to do the more uh rigorous analysis on their own great you can ingest this data it's explainable it's an easy to uh access format.

And another thing to like add on to that is I I really want to be careful about agentic or AI recommendations. Like that's why that's something that we haven't really gone incredibly deep into. Obviously, we're testing and we're trying things and we're designing behind the scenes, but I want to be confident that what we're putting in front of people is is more on the deterministic side and is in line with what their goals and their strategy is. Um, yeah. So, so I think it's it's about distilling all that feedback and finding those like key themes to guide your development.

Yeah. And when Jordan just said deterministic, it sort of brought back uh memories to I'm sort of building out our our comp set models and where we can leverage AI and where we can't. So, and sort of tease up a question for you, Mark, of and what were some of the hard calls when building this where like two smart people and you've got a lot of smart people on your team can have and sort of differing opinions on how and why I'm we should build something. Uh any any good examples there?

There's one that for example was uh revolving around coms. like coms is I think a great example for that. When we started designing coms, we thought okay this could be a problem where AI is very good at and so we started with this idea to let like AI do a lot of the heavy lifting and like do a lot of the first initial selection and then ranking results came back and it was a disaster. It was so bad. It was and like the problem is what you mentioned the deterministic part of it, right? Like you cannot give a different concept every single time or a different price recommendation every single time. And so that one was a challenging one because like we thought it would it would do good. And so then we had to rethink about it and then we figured out a good flow where like AI can give a little flavor without having to do any of the heavy lifting and it it ended up yielding a good result. And we had similar experiences with like now we're working a bit more on the rate explanability and rate recommendation soon to come and all of that like it's the same thing where like you have to be like be able to understand strengths and weaknesses of AI or every single model and so it's been a challenge but it's been very fun.

Yep. We're in the third phase of the adapt launch. So, we started with a private beta, then we moved into a public beta, and now we're into full public launch. And some of the thinking behind doing this sort of phased approach was that and we expected things to go wrong and we expected there to be problems in the model. We expected there to be bugs in the UI and but we wanted to get those users that were willing to be beta testers and to get in and start hammering uh the product. Uh so and both of you maybe we'll start with Mark like what's one thing like you really and that the beta really I illuminated and that we had to go back to the drawing board on

base rate. [laughter] That one is always fun that like just how pe how people because people get so attached to it. It's one number that it's not I don't think it's that I I didn't think it had that big of a significance and I didn't realize how much people get caught on to like a number that they have in mind and and they keep going back to it and it's like I understand why but like I I I I like let us help you that one. And then overall like just like the studio onebedroom segment uh like this like how noisy it can be because of the hotels hotel data and like hotels these guys in a short-term rental data. That one was a cute like I knew how that it would it was noisy. I didn't know how noisy it was and it's one that it has been a lot of fun just going through it and like cleaning it up.

What about you Jordan?

Oh gosh. I mean I wanted to steal Mark's answer on base rate. That was the when whenever we launched and so for context, right, private beta around June 17th or so, 25 people we we picked for kind of the first wave and almost immediately it was where's my base rate, right? Like how do I pick my base rate? How did you end up with this base rate? I'm confused. I'm frustrated. There was a way to change it, but you just had to go to this tab at the end called settings and then find base rate. And then there was nothing to explain like how we got there or give you any context. Uh and so that was probably that was the biggest thing that we were like, "Oh, shoot. We need to we need to go fix this. We need to um provide more context uh around what the the base rate is and how we got to the recommended one." Uh another thing I would say is probably the price rules in general. A lot of what we heard in the feedback was, "Oh, it's so complicated.

all these all these knobs and levers. And so my initial thought is, wow, this sounds like an opportunity to not build all of those, right? Like we can we can be really simple. Uh we can just give them a couple of strategies uh which are important like those those are still key. I believe that. But I thought we could do that in lie of building all of the the knobs and dials and levers and uh that quickly um came to surface that that was not uh true. There are people who really want to get into the weeds and make those configurations and they're not scared of the complexity of of getting in there and really um configuring things to their liking, what makes sense for their operations. And so we realized we have to build out this kind of full tool set while also providing the simple u kind of out of the box solutions, the easy button if you will of selecting an adaptive strategy. Was there any moment where likeh this isn't going to work?

Are are you are you asking me to tell you like every time I go to bed at night I'm wondering like I'm wondering are we building the right thing right like is there something that we're that we're missing spoiler alert there's a lot that we we still want to build. I mean, we we're day one on this, but no, I I think that I got a lot of confidence in the that private beta. I told you we started with 25 handpicked users over seven weeks in that private beta. We're inviting more and more people. We're getting really good feedback. Uh we're improving the product every day, whether it's something small like a UI tweak or change of copy or releasing new features like our inapp uh AI agent. Eventually, we ended up with 11,000 connected listings after two months of being like very quiet about it. Like maybe a couple LinkedIn posts.

I know you're a celebrity, Jamie, but you know, it didn't drive that much traffic. Uh, and so it was really cool to to see, you know, over 10,000 connected listings. And so, I've just become more and more confident over time. But early on, yeah, I'm I'm new to the revenue management world. We I we're all kind of we have some experience with revenue management, but none of us are experts. And so it took talking to a lot of people, a lot of customers, a lot of people who face these problems every day um and really hearing them and trying to build the thing that's that's going to um help their lives and make make them better at revenue management.

Yeah. And you you talked a a bit about sort of AI um and how it did sort of change the way we were able to build. So hoping if you could talk and a little bit about the specifics of it of like and let's say today verse like if we were building this in 2023, how has product development evolved and like how specifically is is AirDNA sort of viewing product development today?

Oh my goodness, great question. AI has introduced uh a lot of speed as well as some complexity to be totally transparent. Uh to get to your question about 2023 for context, the front-end team like the team who's you know building what you see in the app and a lot of the data engineering making sure all you know all the data is piping through. We've been doing that for six months and and we released what you see in adapt right now uh in 2023 to get to where we're at today over a year of of just the front development. And that's that's not even taking into consideration all of the integrations that we have built and are building. That's not even thinking about the the data science involved and connecting those worlds together like between the app and data science. Uh so it's it's wild like I I never would have before AI. I don't think I could have said yeah Jamie we can we can ship this in six months.

Let's go. We started back in March when I put the team together. Uh yeah, so that has just been incredible and what that enables us to do like at AirDNA we are taking a very MVP iterative approach to how we work with AI like we're testing things we're trying stuff and that translates not only from how the team works together but how we are developing adapt and developing the product is that we're able to just try stuff and and that's what we do because I've been in product management long enough. I've been software long enough to know that your first shot at something isn't going to be the best version of it. You're never going to get it right. It requires iterating. It requires listening to customers, moving quickly, getting it in their hands, and getting real user feedback rather than just letting your your opinions get in the way. And so we are we have a very experimental uh mindset, a very, you know, move quickly, let's ship it, let's learn and iterate like that.

that's become our model in the age of AI and we can do that with a a smaller team. Yeah. And so that that's been incredible.

So like I've I sort of sit on I'm product adjacent, right? Like we chat [laughter] uh and it and it feels like it has changed a lot. like if we were sitting a meeting together in 2023 of which we were like sitting in meetings it was like let's ideate let's talk with the designer let's then wait a few weeks and get some designs back and then fight about those designs and then get new designs back and then I mean two months later like the devs sort of get a v1 of what we're doing and then we debate on that because it like it doesn't work exactly how we thought it was going to and then we ideulate more and we sort of move iron like maybe just if I'm off on that process maybe and fix it and then maybe explain a bit more of like what the like iterative process looks like now in this new environment.

Yeah. No, you're absolutely right that that's what it was. It was hey this is our best idea. let let Jordan or any of the other product managers go write a you know fivepage essay what we call a PRD a product requirements doc and then me you seuite all argue about that and then exactly like you said okay let's pull in designer we wait a while u you're totally spot on and how that's changed now is I actually I built most of what we see in adapt uh in a prototype using cloud code like it it's it's gone gone from that that very like long kind of throw things over the fence process or like this hand off. Okay, time to hand off to design. Now they bring it back. Now we hand off to engineering. It's become more of oh you have an idea Jamie. Oh, Jordan has an idea. Let's mock it up. Like we can we can really look at the user experience.

We can connect it to to Snowflake or you know connect to our data sources. We can play with real data and see does this idea make sense? We can put that in front of users. like we can we can just uh coal so much of the lack of confidence so quickly because we can kind of build it. Um obviously we're not you and I aren't shipping stuff to production. We still want to bring in our we still want to bring in our designers. We still want our engineers to really own uh delivering the best solution, but we're not starting from a five-page document anymore. We're starting from what does this really look like? How does this feel? Does this make sense? Uh, and so that's just incredibly powerful

and and I do want to highlight that of what what you said of like there's this iterative process with cloud code where Jordan's designing, Jordan's pulling in data, doing different visualizations and and I saw some of your original mocks like you're you're showing overlays of rules and how they would work, how they'd visualize, but then that's not what gets shipped. there's then a handoff process to real software developers that are then going and writing the code and and building out the process like it is not a claude it's not Jordan by coding an app this is still real software development it's just happening at a pace that's so much faster because we're able to get to what to build the right thing to build and from the beginning then and through more sort of longer process. Is that right?

Oh, absolutely. And so that's what speeds us up on the the product side and then design is also using it to to try things out quicker to mock things up faster using our design system, right? They can kind of put all the pieces together in a in a more professional probably polished version. And then engineers are also using a aentic engineering to speed up their process, right? And so everyone's still staying in their lane, quote unquote. Uh but it's just gone so much faster. It's like we all have superpowers to do the thing we're best at that much faster and we can accomplish so much more now.

Yeah. Maybe maybe Mark you and same question to you on how sort of data science has evolved uh in this sort of new AI realm. how you guys think about and sort of iterating how you think about building models, how you think about testing models, how you're even like I'm getting a sense of like what's working and not working sort of the QA testing around the models that it does feel like there's been a whole lot of new approaches taken to make sure the the right data is sort of coming out of the the system.

Yeah, 100%. Like for us AI has played a big part first of all first of all in the research phase. So like in the research phase before you had to go through every single link and find a lot of uh different uh research articles how how we build this uh like manually. Now you cannot just ask cla or chd research project. Can you start sending me some uh some articles and some literature on all of these topics? And so that already speeds up the process by a lot. We're talking like probably like days or weeks. maybe highlight that a little bit because Jordan talked about like once the dev started about six months of work but the data science and modeling and sort of initial research started well before that where I remember all of us yeah sharing back academic articles and doing chats with PhD data scientists of like I what are the new approaches today and sort of going about I mean the the ultimately optimization models that we're building that that research now feels like so long ago but I do feel like it was uh very important towards building a modern data science sort of ecosystem for this to live in.

Oh, it was massive because like it as like it helped us first of all put together all the ideas we could before like same thing it would have taken us forever to do this like now we could move way faster because we could have an aggregator for us and then as we were sharing uh ideas and research it was also good that you could use AI to like do summaries and understand where the main idea behind the algorithm was and see like is this going to work with our data and not only that like then when it came to like building time. It was also great to just be able to do the validation tests because you could connect it to our data or get a sample of our data and just like before I we had to talk to the engineering. We had to try to get something spin span up like it would take weeks and now for us it was just so much faster. And what was very interesting about all the research that we did to is just talking about the different models and how they interact and how it also interacts with all the different settings that yielded so many future ideas and that yielded some of the work that we have on the pipeline that like you're going to like we're we're currently working on and we're going to be working on in the next few months.

That's that's awesome. So each of you what would you do different uh if we started on adapt starting today? uh read the Airbnb paper on probability of booking from the from the get-go and and understand that probility of booking is not the answer to all of your problems. It it's a great help but it is not the answer to all the problems. Uh I that one would have been of great help. Uh, and just like the biggest thing I think we we would have spent more time like cleaning up the data which like we discovered later if we still solved it before going live but I think that maybe at the beginning spending more time like cleaning the data would have been like a it would have yielded better results from the get-go. Like it would have helped us like see better rates when we were iterating and instead of at the beginning we're getting frustrated because the rates we're seeing were not great. And then that's when we understood, hey, it's not the model itself, it's the data behind it.

Right? It's one of my my favorite books, uh, signal and the noise of there's a signal in the data and you got to sort of clean out the noise to, uh, pull that signal out

100%.

What about you, Jordan?

Oh my goodness. Uh, nothing. I think we've crushed it. Now, [laughter] I I do believe that. And again, it's easy to say that with this iterative it iterative approach with a team that is incredible and works so quickly, this might be a hot take that I don't think I've shared with any of you, but I think I would have started with API first. Build something that is maybe more akin to like a Stripe and just a very simple dashboard uh and give people API and MCP access and see what they do with it. I think if I could go back and again we're building this so it's like I'm still gonna get to see what happens with it but I I think I would have done that first before even bringing in the kind of like front-end team or or could have could have used fewer uh resources, fewer engineers to kind of build the MVP for it.

Yeah. May maybe looking about now into the future and we talked about how much things have changed in the last three years and maybe getting you guys thoughts on and I won't maybe not next three years but maybe next year or two uh a bit further in the future. Where do you think revenue management's going? Um, and everyone sort of claims this sort of AI layer now and we've got MCP, we've got and and chat within the with the within the app like and Jordan, where where do you see this evolving in terms of what you're going to be able to do and and or how you're going to perform the sort of task of revenue management?

Yeah, absolutely. I think that right now, even as other folks in the space are starting to build additional tools, I don't think anyone's built a true holistic revenue management platform. And I think that's where that's where we're going to go. I I think that's where you need to go. uh is building something that again not only in the UI but from a data perspective you have access to understand not only how am I doing with my my rates my availability but being able to dig in deeper and see what's happening higher up the funnel what's happening in terms of the the listing optimization where am I ranking and how does that affect whether someone books with me or not because that's going to play into your your rate as well are you do you have a a really restrictive midnight stay. Okay, great.

Now you're not going to show up if someone's looking for a stay that's one night shorter. It's like there's so much nuance and granularity involved and I just don't think the pieces are kind of out there, but no one's brought them all together to give you the the data points that help you build a truly uh holistic revenue management strategy. And so I think that's that's where we need to go. Like that's where it's going be beyond midnight stays and and rates and daily rates. It's going to get so much more nuanced. Even getting into we love talking about this, but channel optimization very specific, right? Like because if you discount your rate, it's not going to show up the same on Airbnb as if you actually use a a promotion, right? That gives you the strikethrough price and there's a psychology for the guests there and it changes how you're ranked and there's just there's so many variables.

And so I think it's tackling that, bringing all those into one place, whether it's adapt the UI or adapt the data layer that's coming into your your own um you know internal tooling. Uh I think that's the that's the path forward and that's where we need to get to.

Do you think maybe in two years that the majority of users will be actually working within the AirDNA UI or do you think it'll be happening outside?

What a question. Um, it's so hard to say because there's different levels of sophistication, right? There's different levels of comfortability with using these AI tools. And so, I think there's going to be a lot of folks as the industry matures. Um, and maybe we see more of these uh more property management companies spinning up with the what like 20 to 60 listings. Uh, we see more uh more maturity. I think they are going to adopt AI tools at a a much faster clip where I think you get a lot of these owner operators, folks with maybe, you know, one listing. I think it's going to be hard. There's going to be some some folks who are technical enough to to adopt the AI tools, but I think we still have to build a friendly yet comprehensive uh UI for those folks.

Does it bother you if someone would come in and subscribe and never use our UI?

No, that'd be awesome. That'd be great. I Yeah, I I think that like we have the the best data in in the market. And so if you want to use us as just the data layer to pipe into into your HQ, into your internal tools, heck yeah, I'm I'm all for it.

Awesome.

Mark, what about you? Where do where do you see all this going? I in terms of dynamic pricing, I think that the one that is going to win all of this, which is going to be us, uh it's going to be the one that like is able to fit right in between the segment of what a property manager wants and a revenue manager wants. And so, let me explain that many revenue managers are thinking that AI is going to replace them. And then there's a property manager that wants to get rid of every single revenue manager. And so, I think I think there's a balance in there that doesn't have to be exactly like that. I think what we need to build and where we're going and what all of this is going is we need to build a tool that speeds up like helps revenue managers make their their job way easier. And instead of u managing like a 100 listings, they can manage a thousand. And so if we build a tool and we give them those like that power like that superpower, then like property managers are way happier, revenue managers are way happier.

And then you have found like a very niche spot in in the middle of all of this. And I feel very confident that we can do it because like if you look at around the market like what the figure I mentioned before, right? Like 12 million active listings. We scrape them daily. We have super accurate data. We have a unified data set across Airbnb, Verbal, Booking.com. So all of that I think that fed into an a and model plus AI done well like I feel it can yield a great result and that's where we where I think we're heading.

Yeah. Uh Jordan, I know you've got what 10 different dashboards you're looking at and looking at sort of usage, trying to understand how people are are leveraging the product. What are you most keen on and what do you not necessarily know today that you you might know in six months to understand if uh the product is resonating or not?

Wow. Uh, that's a great question because right now, so we've been going through beta and we provided a really compelling offer to those people to give them 60-day free trials, right? And so then once we got to public beta in the beginning of August, we gave them 30-day free trial. And this is at a at a discounted rate, too. And so what I don't know is how long they're going to stay with us. How long are they willing to pay for this product? And so that's something that the signals look great. Like the people who turn on the the price sync, who you are sending rates to their to their PMS or Airbnb, they're sticking with us over 70% which is incredible. But we haven't gotten to the the point that every business has to reach and that's hey give us money for this or we haven't reached that at a large scale yet. And so I'd be really curious to see how that uh the conversion and retention uh hold in the next six months.

It's like by the end of the year, what conversation are we having having?

Yeah. What about you, Mark? What are you What are you tracking? I I'll seed you on one like and interested if this bothers you or not, but like of like seeing people override rates. Like we talked in the beginning about like oh like everyone complains about need to override a rate or not. Like I know we I mean we're we're tracking that. We're trying to get a sense, but like is there is there anything you're looking at any anything you're trying to understand and looking at on usage behavior? Um I try to gather like insightful things from the feedback that we get. We always get uh that person that says my rates are too low but they don't say why or my rates are too high and they don't say so like that one to me like feels a little bit noisy but like when someone is for example a couple people start talking about hey uh I am seeing that there's a uh going back to the events there's this event in this market like I think it should be priced much higher like to me like all of those singles that can make the rates better are the ones that I look for like more than just like general feedback of like rates hard to high to low like or the other one is like hey I have a new property uh I think we should be offering way higher promotion like the promotion should be way more like way stronger like very aggressive so I can get those bookings in so that is another one that we like I remember like I I'm keeping a close eye on because I think it has come up a couple times and so just trying to understand all of these points that are coming that uh beyond just the noisy ones.

Yeah. All right. So to close uh I've got two questions for you or a question for you each and you can p and pick the persona uh that you're sort of going for whether it's a individual host small to midsize property manager or large property manager uh that has maybe been burned by pricing tools before and if you and we talked with so many people Jordan are just like and I'm never using a pricing tool again and I don't trust them. Why should they trust adapt uh when they've been burned in the past?

Yeah, I will go ahead. So, this go this uh ties into what Mark said earlier about we abandoned the black box. We wanted to focus on explanability. We wanted to focus on building confidence in whatever decision you made whether you go with our recommendation or not. I want to give a specific quote from a customer if if I might. This is my favorite quote I've ever heard. This is from a uh a midsize uh product management company, you know, small to midsize. They have uh 30 plus listings. They came over from, you know, using another another tool they were unhappy with. It wasn't giving them the uh level of insights that they needed, the level of control and guard rails. Like this is someone who wanted to get into the weeds. This was a revenue manager and the property manager on the call uh with me, the CEO. and uh and he said, and I quote, "It's like I'm the person dying of thirst in the desert and I just walked into an oasis.

You just handed me a glass of water." And so, I mean, if that quote, that's a real like that's a real quote from this guy. And uh I think that it's just a testament to the team being really thoughtful about what we're building like being really in tune with what has worked and what's not worked in in the other solutions in the space whether it's another software or it's a spreadsheet and and so I think coming to AirDNA you can trust that we care deeply about what we're building. uh we have a high degree of confidence in the the data and the recommendations. We're going to explain it to you. We're not hiding anything and we are going to continue to build so much and build fast and and hear our customers and build what they need.

What about you Mark?

Uh for me it starts with the data. Uh like people should trust adaptab because we've been doing this for uh as our DNA has been doing has been in the short term mental space for more than 10 years. Uh we are extremely accurate. We have the best data out there like we're trusted by multiple s by multiple sources and very different ones and multiple people have based their investment decisions on our data. So that to me is like the main reason why and then the other one is that I think that what we're building is going to help both the small host and the middle and big uh mid and size host and big host. The reason why I think this is because we are not over complex in it like we're not doing something overengineering it. We are giving you the tools and then you can use them as you want. Uh we are building you like an MCP or an AI that you can then like ask on like why is my rate like this on this day and then it breaks down all of our historical data and tells you it's exactly like this because the money is softer this year.

Uh we're seeing that like we're ping a bit lower. So this is why your uh rate is like this. And so to me that is extremely helpful beyond just trying to touch 18 knobs and ended up in the exact same place that you started at.

Well, since I know there's no one else that knows our data better than you, Mark. Uh it it it says something that you your number one reason is is the data. uh and I know you and your team and spend a ton of time and effort making sure the data it is clean and accurate and consistent uh over time as possible uh and complete as possible. So uh thank you guys for that and thank both of you for spending an hour with me to talk about uh adapt uh the process of building it and for all your hard work and in making it happen.