17. The Smartest Ways to Use AI as an STR Operator

AI can do far more for short-term rental operators than write messages or summarize notes.

In this episode, Steve and Petar explain where AI creates real value across hospitality operations: turning reservation data into useful insights, personalizing guest experiences, removing repetitive work, and choosing purpose-built tools instead of building fragile DIY systems.

Key takeaways:
• How LLMs can personalize guest experiences using customer data
• Why clean reservation data and reliable pipelines come first
• Where traditional machine learning can outperform generative AI
• How to use AI to grow revenue and improve hospitality, not only reduce costs

Learn more at https://yada.ai/demo

Transcript

SPEAKER_01 0:00

Hello, everybody, and welcome to the Fully Booked STR podcast. I'm Steve Hasi. I'm here with my co-founder at Yada, Peter Oydrovich. And today we're going to talk about the smartest ways to use AI as an STR operator. Peter, are you ready?

SPEAKER_00 0:16

Yeah, this is uh there's a lot to unpack here, and I think it's a really interesting topic. Um there's a lot of a lot of different ways to use AI in any sort of business. Um it's just a matter of finding like where the best use cases for it are and what the most intelligent way to use it is, and not get stuck in you know buzz loops and these sorts of uh gimmicky things that may not provide a lot of value. So there's a lot to unpack here and a lot to go over.

SPEAKER_01 0:47

We were thinking about what to cover in today's episode, and the idea of what is the state of AI, what can STR operators do to leverage it was most uh exciting to us. So that's that's what we're gonna open up. Um it comes from a lot of experience on our side. Uh, we are power users of AI, not just for um uh creating Yada and you know giving our customers the best uh access to these tools and to this intelligence, but also in how we build our company. Um, we're constantly listening for what is the latest, how are these things being used? So, Peter, let's start with uh if someone is an operator, where should they begin? Let's let's say they're kind of at the early stages of experimenting, what's the best place to start?

SPEAKER_00 1:41

Well, it really, really depends on what the primary objective is. And if the primary objective is to save some time and and like kick your feet up and be at the beach, then the only real like use case for AI is a chatbot to handle guest messaging. Um but I don't want to talk about that, I don't think it's as interesting of a topic, mostly because the focus on using AI in those situations is more on um like work, like workload reduction versus optimizing for growth and forward thinking and being able to scale up. Um so at at Yada, I don't think we we have a single use case for AI where we're using AI to automate customer support or customer success. Like we certainly use AI to summarize calls and transcripts and things like that to um to keep our team on top of what's going on, but all of the interactions with our users, with our customers are generated by humans to provide the top, like the best possible level of service and support uh any hour of the day. Um, so we're probably not the best people to ask about that. Um but when it comes to using AI uh to help you grow faster, there is so much out there that can be uh that can be implemented pretty pretty quickly. And one of the biggest places that I see for for like really smart usage of AI is in just plugging in all of your like, I mean, we can start like super high level, um, you know, exporting like any sort of calendar and reservation data from your property management system into a CSV, into BigQuery or Databricks or something like that. Uh, but probably CSV at this scale. Like I don't think anybody has like billions of rows, so these are probably tens of thousands or hundreds of thousands of rows, and and most machines can or most AI systems can write scripts to parse them and and you know, understand them pretty quickly. And use like very easily available AI tools without any sort of additional like external tooling or anything like that to just understand the data that powers your business at a fundamental level, because understanding where your revenue is coming from, where you have leaks in your revenue, where you're not optimizing um, you know, why parts of your portfolio are performing in one way or another, this is the um like the first step in the ladder of really professionalizing. So if you look at how hotels operate, airlines, restaurant chains, like every square foot is accounted for. There is metrics and visibility and understanding on every like lever that drives performance anywhere. And I think for STR operators, before we start getting into the tactical stuff of like how to use AI to grow faster, one of the key things that I would really recommend is like using the machine to understand your data in a way that's that was previously only accessible to much larger teams that had internal data teams to help them figure out occupancy and seasonality and things like that. So don't rely on the analytics that the PMS gives you. They're really like minimal at best, depending on what system you're using. Pull your data out, run it through the machine to get a good understanding of what your performance looks like year over year, and um you know, try to find some gaps in there or something in there where you can uh you can improve it.

SPEAKER_01 5:34

That's a good call. And I mean, even if you are on a PMS that is really good in terms of analytics, it's like a second opinion from a doctor. If you're just taking you know your system's word for it and getting the views that your system offers, you're missing out on potential insight. Um, and the interesting thing with that also is let's say you get the same information from you know uploading a CSV to uh Chat GPT versus you know what your PMS is telling you. Your PMS isn't necessarily going to tell you, so what? What should I do next? Whereas with an AI system, you can see that data and say, what's my biggest lever here? What can I change to you know grow in in this certain way that I'm that I'm seeing from the data?

SPEAKER_00 6:29

I would also add to that um, you know, one caveat, and that is you know, the advice that the machine gives should never be taken as gospel. That being said, it is better than nothing. And certainly until a certain level of scale is reached where you have the resources to run these types of analytics yourself, uh, it is a phenomenal way to like start understanding what's going on. Um and I think it also helps with a mindset shift around using the tools and the technology that's out there to just focus on how do we grow as much as possible and as efficiently as possible and as quickly as possible, instead of saying, and instead of thinking we should use the tools out there to save ourselves as much time or as much money as possible. I think that goes to like a deeper conversation around you know what sort of psychology or mindset drives business results more. Um and like growth focus will always like growth focused companies will always outperform cost-cutting companies and like out like outsourcing like critical business services companies. So by like reframing your usage of the machine as something like not to save you time, but like enable you to grow faster is I think also super important. And this like initial exercise of like talking to your data and understanding the numbers and the metrics that power your business in a human, understandable way is a great place to start with that.

SPEAKER_01 8:11

I love it. Yeah, I love that shift from let's save as much money as possible, which is good, and absolutely, you know, put the machine on that. Like, where are there places where we're leaking revenue, where um, you know, we're paying too much for a service, for an employee, for a certain role. Um, my overall advice in terms of mindset, like how to use AI, is find out where it can help you.

SPEAKER_00 8:41

Yeah.

SPEAKER_01 8:42

Because there's no way that you know the full scope of how this thing can help you right now. Yeah. Um, I'm always blown away when I say, well, let's see if it'll do this. And I put in a prompt and it does. Like, I did not expect that result.

SPEAKER_00 9:00

I just do want to add something there, Steve. For us, it works really well because we have really good pipelines to get the data out of the systems that we use every day and into a format that's really accessible by the machine. So another like really important point here is that AI doesn't replace the software and the solutions that you're already using or that you might be thinking about using. Because the stuff that's really, really hard to build are the data pipelines, data cleanup, like managing the raw information that gives the AI the context that it needs to make judgment calls and deliver insights and so forth. Um so, as you know, I started this by saying, like, you know, pull your data from your property management system or whatever other tools you have, like whether that's you know, Yada for marketing or or um the uh like wheelhouse or like price labs for pricing, you know, your first step is to extract the data that exists within these platforms, um while also recognizing that these platforms are the only place that you can really like create and store that data. So if if you can never use like the machine to like roll your own property management system or your own marketing solution or your own dynamic pricing solution, because um for a lot of the the platforms out there that generate, handle, and manage the data, there's years and years, if not decades, of institutional knowledge on like how to do things really well at scale. So like I'm always like really uh amused when somebody says, Oh, I built a Slack clone. Like you built a Slack clone that can handle like two users concurrently. Um, but you certainly didn't build an infrastructure that can handle like 50 million messages a day and like zero dropping of messages and just 99.999% uptime and things like that. So the tools that you're using, like you need to continue using those tools because that like institutional knowledge of like handling, managing, storing, working with data, building the integrations, building the pipelines is incredibly valuable. Like I would say, like more valuable than ever because it allows you to do all these sorts of insights and actions with the AI that previously um were just not possible. So that's the other thing that I would you know really want to drill down on is like it's it's not really a question for for a lot of like usage of AI of like build versus buy, like I'll split spin up Clot or Codex and roll my own property management system and roll my own marketing solution and roll my own dynamic pricing thing. Um, it's more of a matter of let me go out and spin up and buy the the right solutions for what I need. And I think for most property managers, it's pretty basic. Like you need a property management system, you need someone that's really, really good at pricing. I think super critically, you need someone that's amazing at marketing because that's one of the most direct ways that you can turn those AI insights and that data into future results, and then maybe somebody for like turnover and task management and and cleaning management and so on. But without these tools, like the AI lives in its own bubble of hallucinations and assumptions and things like that. And we all know that assumptions make an ass out of you and me. Um, so like to do all of this really cool stuff with the machine, you need to have the ability to like own and create and store that data that powers the machine in a really stable and in a really good place.

SPEAKER_01 12:35

Yeah. I was speaking with uh an operator not long ago, and his take on it I really loved, which was you need the tools that are gonna do the job, whether it's set the price, send the marketing message, block the calendar, charge the credit card, send the message, whatever it is. You need the tools on the outer layer that are gonna do the work of managing your company and you know, taking care of guests. And you need a big brain around the whole thing.

SPEAKER_00 13:14

Yeah.

SPEAKER_01 13:15

And that framing of it, I really loved because it's not enough anymore to have a tool that will help you do those things. You need intelligence across the whole organization that can bring together how I am doing with my turnovers, how is my staff, how is my occupancy, how is my marketing, how is my database, so that you can actually know what to do.

SPEAKER_00 13:40

Yeah.

SPEAKER_01 13:41

And so we recently changed our own CRM system from a tool that did not enable that, actually made working with the AI pretty cumbersome, um, both in terms of building out our website and uh you know working with our um contact database uh to one that made it really easy. Like this was built for the machine to be able to give us insights and like have that big brain right in the middle of the whole thing. And it's a process to switch your CRM, it's like kind of the backbone of your business. Yeah, we did it in order to have access for the machine to have access to that data so that we could have more power in how we grow.

SPEAKER_00 14:26

And what's really important there is I think our own experience highlights the fact that the biggest use cases for AI are not things that are gimmicky. Because we've tried, like I think as a company, we we really focus on trying every possible tool out there because you never know where something is gonna just blow your mind and make a lot of things faster. But the fact of the matter is that like the vast majority of those tools that we've tried and and piloted and demoed and looked at are very gimmicky, like they don't actually solve the problem. So I was looking at one uh the other week, it was like uh um essentially something that listened to all of the signals across social media, across SEO, across your Google Search Console. And like it couldn't turn that into any sort of viable insights or content. Um, you know, same as we we tried this tool uh that lives within Slack, and it's supposed to be like your, you know, it was a really interesting concept. Like, you know, it connects to all of your external tools and lives within Slack, but it couldn't really do anything. So it was like super gimmicky, like it would, you know, high five and thumbs up and things like that, but it couldn't actually solve a problem.

SPEAKER_01 15:44

It was a heck of a hype, man.

SPEAKER_00 15:45

It was a heck of a hype, man, but it couldn't solve a problem. Um meanwhile, like just the simplest workflows and the simplest tools that we've in a way been using since day one have continued to be the bedrock of how we use AI. They just get a little bit more sophisticated and a little bit more capable every day. Um, and that's literally just having, in our case, a really, really good database of customer data, interactions, customer relationship data within the CRM, and then phenomenal analytics data from for how people are using our product. And that's all we really need, right? Like everything else that like like the actions that come out of that are like they fall into like a relatively small number of buckets. It's either you know working on something that's growth related, like a marketing campaign or a sale, it's something that's you know customer success related. So debugging and triaging and figuring out why something went wrong, or it's building code and building content that goes out online. And when you realize that it's like the number of verticals and the number of buckets in which the machine needs to take action in is actually quite small. Um the need for or like the temptation to start using gimmicky tools becomes considerably smaller uh because the output is very like tightly scoped. So for instance, the other day, um you know, I I have a lot of admiration for the work that they're doing at Box, or sorry, not Box Square, uh, but they released this like quote unquote um Slack killer, I think it's called Buzz. And I tried it out for a few days, and everyone that I've seen trying it is not actually building anything that works or that's in the public or that has paying customers or that has you know the need for stability. Um it's like a really cool, but it's kind of gimmicky. It's like it does something that a lot of other things already do anyway. It's just a way to like orchestrate agents and things like that. Um so I would say that you know, as you're as you're looking at how to use AI in your short-term rental business, uh, you know, don't be like impressed by this. Like, there's a lot of like I wouldn't call it snake oil, but just like very like uh hopeful uh work out there around like all sorts of different like orchestrations and managements and layers and things like that. When in reality, like the thing that's gonna like move the needle the most is having incredibly good data for everything that your business does and making that data incredibly accessible to the key tools, which is your Claude or your OpenAI instance, your email inbox, and like perhaps some external tools for like task management. Within the tools that you use, like there should be a lot of AI built into it. Um and I think the platforms, like I'll I'll die on this hill, but the platforms that are gonna be the most useful are not the ones that are like completely AI native, which is just like a wrapper around data where the agent then and Claude or ChatGBT or whatever goes out and does the work. Like it's the platforms that integrate machine learning, AI, like analytics, knowledge, like generation into every single activity and layer and action that takes place within the platform, because that's like a much more purposeful use of AI versus like, oh, we're just gonna build a database and then throw like slap an MCP server on top of it. Like this is like that's like a very intentional, like thoughtful, data-driven way of like, you know, using AI, like the platforms, building AI within them to you know, minimize the number of clicks or or make actions that would otherwise have taken 30 minutes, take 30 seconds. Um, you know, as an operator, you have to get access to your data, use that data in conjunction with Claude code, or doesn't need to be clawed code, but like Claude or Chat GBT to understand that data and like derive insights from it. And then when you're picking out tools to use, make sure that they're really good at like working with the data, but also have like very deep AI layered in everywhere to make life easier.

SPEAKER_01 20:23

Yeah, because if we think about the smartest ways to use AI, it it's a bit of a misleading statement.

unknown 20:30

Yeah.

SPEAKER_01 20:30

It almost makes you think, well, okay, as an operator, just you know, do everything in-house, build it all myself, let AI, you know, go nuts. That doesn't work at this point in 2026. You can't just say, you know, create all the software that I'll need uh so I can cancel my every subscription. Um, sure, maybe some of the things that AI will spin up for you are things that you could cancel. If that's the case, those things were probably pretty lightweight anyway. Um, usually there's enough technical requirement and infrastructure requirement that whatever the AI would spin up for you would not, if it is on par with another tool, it certainly isn't going to develop on par with that other tool. Um if if there is a company that is devoted to solving that problem, like we are uh at Yada for solving the marketing and hospitality problem, um that's a significant advantage that you have having that company on your team, having us on your team in this case, um, because we're spending cycles, we're spending energy, we're using the AI to build that feature set so that you don't have to, right? And so in any of these areas, um you could roll a lightweight layer to solve each of those problems and you know cancel those subscriptions, but it's not going to be durable. It's not gonna help you really scale. So the smartest ways to use AI as an STR operator is to have on your team. Those systems that are using AI in a really smart way that are getting you phenomenal results for pricing, for marketing, for all the things that we've been mentioning. That way you don't have to worry about it. You don't have to be, you know, figuring out if the machine is hallucinating or not, or how to become an expert at building marketing software or at uh you know property management software. You can actually focus on your hospitality. You can focus on what makes your brand unique. We've talked a lot about that on this show is you know, building out your own experience for your guests. That really is the smartest way to use AI as an operator is to get the bullshit off your plate, let other people handle that. Use AI to say, hey, how are these things going? Is my pricing okay? Is my occupancy up to, you know, up to snuff? Um, let it give you that second opinion, let it give you that insight on where to grow, but then focus your energy on your secret sauce, focus your energy on your brand, on those guest experiences, and that is going to create uh an incredible combination for you.

SPEAKER_00 23:12

I also want to add here that if um when you look at the most valuable applications of AI right now, and let's just take short-term rentals as an example, uh, a very small percentage of it is actually generative. Um so I'll name you know two examples that that really highlight this. Uh at Yada, we have hundreds of millions of data points that represent heat maps of calendars and availabilities, and you know, how far before a trip does somebody book, and how does that overlap with demographics and where they're coming from, why they're traveling, what their their their demographic profiles are, etc. Um that in a way is is AI that's that's far more useful than just throwing a couple sentences into Chat GPT and getting a response back because it leverages you know what what a lot of what used to be thought of as AI up until two or three years ago, which is like deep learning, neural nets, like machine learning, et cetera, to take hundreds of millions, if not billions, of data points and divine some sort of patterns and truth and like future-facing um uh behaviors that are like very, very deeply backed up by the data that has already existed in the past and that looks past like five, 10, 20 years. Um so this is why companies that like use AI or that build AI not just to you know have like a generative feature in a chat bubble in the bottom right window are really providing a lot more value because they're leveraging like all of these incredibly sophisticated data pipelines and machine learning technologies to find insights in that data to drive actual, like actionable insights and suggestions and automations for their customers. And the other example is price labs and wheelhouse and all of these dynamic pricing solutions, which are AI. It's not generative AI like Claude that's like writing emails or writing code. It's using the vast data set that they have around pricing and dynamic pricing and surge pricing and all of these different types of things to get the operator the best possible price, right? Like Uber's AI division, like even now, if you you know, Kalinick spun up a new company uh eight years ago, they just came out of stealth two months ago. Um, like when he left Uber and the stuff that he's been working on for the past eight years is almost entirely AI, but it's not like Gen AI, like large language models or anything like that. It's just incredibly sophisticated applications of like computer vision, uh pricing, map optimizations, like all of these different types of things that like Gen AI can never do. Like Gen AI is really good at like understanding text and kind of figuring out like broadly what's going on, but it's not like the be-all end-all of AI. So, you know, if there's one like parting thing that I want to uh you know really hammer home is that like AI is is is a vastly broader world than just Chat GPT and Claude. If you look at like the business result like returns on AI, I was reading somewhere that like generative AI is like less than half a percent of like the total like global dollars that are generated by like AI activities, like Meta. Um I think if you take like the total revenue and like the like I think the the dollars like for for most enterprises and corporations that use like like uh Claude and and and and uh the GPT models, like for every dollar that they pay OpenAI and Anthropic, they get something like 70 cents of value out. So they're actually losing money on that. Um whereas like for every dollar that they spend on like Google Ads or Meta ads, which is like phenomenally sophisticated AI for like getting the right message or the right ad to the right person, the ROI is sometimes like 10, 15, 20x. Um so for a lot of these like traditional ideas of AI, which is just analytics, machine learning, neural nets, like deep learning, like figuring out how to do the right thing at the right time most efficiently. Um as an STR operator, like I would also be thinking about that as well. So, not just, you know, um, how can I use Claude or Codex most efficiently, but can I find the platforms that are really at the bleeding edge of, you know, taking the millions of data points and the billions of data points that are being generated and using that data to drive like actual business, like measurably drive business results and business outcomes?

SPEAKER_01 27:58

Yeah, that's awesome. I am gonna stand up for the LLMs here though, not as though they need any uh, you know, standing up for they're certainly still the the bell of the ball at this point that most people love them. Um, but as an STR operator, one thing that you could do as an LLM, and and uh spoiler alert that this is what Yada does, but it's the thing that I am most pumped about right now is you take all the information you have about a single guest, you take every conversation you've had with them, you take every reservation they've had in the past, you take what you know about where they live and their social graph and you know, their profiles and all of that stuff. You you summarize it into who is that person, who is Steve, who is Peter, and then you take what you know about your business. You've got this whole you know Venn diagram and the overlap between who they are and who you are, and how you want to message that invitation can be so personal, it can be so specific, it can feel so hospitable, like you're really but seeing them, and that is what the LLM enables.

SPEAKER_00 29:17

But I don't it's not that I'm pushing back here, but uh that that's only possible because of all of these existing um like pipelines and and technologies that are out there. So if if you're an operator and you have uh a customer profile, let's say like Steve Hasi, um first off, like if I'm using if I'm using something like Yada, you know, I might have like 150 or 200 data points about you from all the way before you booked, and you're looking at the website and checking things out through your visit and what you interacted with in the property and and all the semantic data, et cetera. Um that stuff only really becomes useful to the LLM uh at scale if you have these systems that can manage that data and and run a layer of learning upon that data uh at scale. So if I have like 50,000 Steve Hossies in my database, like I'm surely not gonna like you know burn through billions of tokens to like analyze every Steve Hossey individually. Like I'm gonna start looking at trends and the customer data that's existing there and try to compress all of that like numerical data, activity data, et cetera, because running your profile through a or like running 20,000 Steve Hossee profiles through an LLM to find the best fits for an offer that I have is gonna be multiple orders of magnitude cheaper than sorry, more expensive than using something like a regression model, which like once you build the model, like every subsequent like ping against it is like infinitesimally cheap, to like first filter like out of those 20,000, like let me find the 500 and then use that LLM layer uh to out of those 500, find the 100 or the 80 or the 75 that are most viable for receiving this marketing message at this period in time. So the best actually is to combine the both, like combine both types of technologies, like use traditional AI as you know, what's traditionally known as AI, like regression models, like deep learning, et cetera, to get to like literally like point your your your spy glass in the right direction and zoom in as much as possible. And then once it's zoomed in as much as possible, extract the relevant data from there and then hand it off to to a generative model who can use that and look at it like not just like numerically and and time series and things like that, but also incorporate like any sort of semantic data that exists about that person to find, like really just fine-tune the right audience and the right message to to reach out to them with. Um so I think we're like kind of converging. Like, I'm coming from you know my obsessive background on like just deep learning and and and machine learning and just you know being super optimized. You're obviously like the biggest uh LLM cheerleader. And I'm a huge LLM cheerleader as as well. But I think the moral is that like combining these two types of technologies and like really finding people to partner with who who uh who combine these two types of technologies is like the winning strategy overall.

SPEAKER_01 32:27

Yeah, I love it. You know, you're talking to an engineer when they start getting fired up about regression analysis and uh deep learning.

SPEAKER_00 32:35

Well, listen, I mean, like at scale, there's no other way to do it.

SPEAKER_01 32:38

So you know that's that's the deal, is it's uh it enables scale, which is what is really exciting. I mean, that's that's what we're up to here. And a lot of the operators we talk to, you know, they're interested in scale as well. And you know, you cannot get to a great marketing program for a hundred properties, a thousand properties if you're not thinking about some of these elements of scale. Uh so yeah, surround yourself with folks and intelligence that are thinking in that same way, and that's gonna be uh a smart way for you for you to use AI as an operator.

SPEAKER_00 33:14

Absolutely. I'm willing to bet, and this is a hill that I'm not yet ready to die on.

SPEAKER_01 33:22

You and your hills, man. What the hell?

SPEAKER_00 33:25

Within okay, like I think right now, like Databricks and Snowflake and BigQuery are making way more money than OpenAI and Anthropic. Like they are significantly better at printing money than these two companies, because you know, if you're an enterprise with like 10,000 employees, like you know, having a good way to store exabytes of data and run models and queries and things like that against it at light speed is is incredibly difficult. But I think at some point, like as these two technologies converge and they will converge, like Palantir is a great example of that, like using like um, you know, not a great example from a you know, it's it's a cool product standpoint, but definitely a great example from like a uh from a standpoint of like merging these two types of technologies, like generative and really good machine learning. Um it's gonna become less like distinctive, like which type of technology is used to do what, because the platforms and the partners that really like deliver results are gonna be the ones that seamlessly and almost invisibly combine those two types of technologies in a way that you don't have to like think about what part of your your business decisions and business logic and business activities live on your traditional like data warehouse, data lake stack, and what lives in the real-time um LLM stack, it it just delivers the results, delivers the actions, and delivers the insights automatically.

SPEAKER_01 34:59

Yeah, that's great. So we're always happy to geek out about this. If you have ways that you're using AI that we haven't touched on, that you find really powerful, really high leverage, let us know. We'd love to hear from you. Uh, and if you want to see what Yada can do for your data and turn it into great marketing, we'd love to have that conversation with you as well. Just head over to yada.ai slash demo. We can line up a time to uh have a chat.

SPEAKER_00 35:28

We can't wait to hear from all of you.

SPEAKER_01 35:30

Take care.

SPEAKER_00 35:31

Take care.