AI Experimentation in Hospitality and Beyond: Practical Ideas for Immediate Impact - Chris Kluis

Josiah: I really enjoyed speaking with you about your unique family story. We were in Indianapolis at HITEC, right? This big tech trade show. And you were saying your family history runs deep when it comes to technology and AI. I wonder if you could share that story with our listeners.

Chris: Yeah. So my aunt was recently knighted for her work in bringing the internet and AI as a professor focusing on AI in the Netherlands. That's pretty interesting. I remember her handing me these books and papers that she had written whenever she visited. And I remember one from the early 2000s. It was called "Agents and the Law." It was about what happens if you have an artificial intelligent agent or a bot working and doing something and it breaks the law on your behalf? Are you responsible for it? And setting up frameworks for governments to think about laws, which is pretty interesting considering where we are today, that that was 20 plus years ago. But one of my other cousins has a master's in AI and he's a machine learning engineer, ironically also at Booking.com, right? So another hospitality company. And one of my other cousins has a master's in econometrics and did a lot of early modeling stuff. My brother has a master's in econometrics and he started off as an economist and then a quant and now a data scientist advisor. So he's grown his career and he's built a lot of AI stuff. And so I feel like the imposter in the family, right? Everyone else is actually working in building AI models and tools and things like that. And I'm like, Hey, I know how to use ChatGPT. Okay. But that's the family story where everyone is much smarter than I am, but I experiment a lot.

Josiah: Well, it's that experimentation I think that is so fascinating. I'm going to push back on that because I think you are doing a lot of interesting things today. And you've also done a lot of interesting things over your career. And I wonder if we could maybe spend just a few moments talking about your work today, some of the things that you're focused on today, but then rewind the clock and talk a little bit about your journey to there. But first, I know you're doing a lot at Actabl. What are some of the core areas that you're focused on for our listeners that haven't met you yet?

Chris: Yeah, I'm the VP of products and I'm in charge of the ops line, which is Actabl InSite and transcendent analysis. And I'm also responsible for the data and analytics team, which includes some of the AI stuff, a machine learning engineer that we are using to experiment and play with things. So my focus right now is helping those two product lines serve our customers as best as we can. There's a lot of different ways that we can tackle that. And the specifics change on a weekly basis sometimes.

Josiah: It's never a dull moment. I wonder if we could kind of look back to some of your career experiences so far, because we were talking in Indianapolis. It was remarkable to me, a very diverse set of experiences, but I feel like in a way each of them sort of led you to kind of the world that you are in today. And I guess as you reflect back over your career, were there a couple of pivotal moments that influenced the way that you think about technology or data and AI today?

Chris: Yeah. So I started, I guess my first professional career job was working in a medical practice. And that doesn't sound like it would really impact what you do, but I worked for cardiac surgeons and they're notorious. I think they have a 91% divorce rate, but they're notoriously hard workers. From the time they graduate and actually get into practice to the time they retire, they're working 100 hour weeks, standing over people's bodies, cutting them open and the work ethic and the almost scientific way that they approached everything. Cause we ran studies and we did all kinds of things. Really just imparted, it jumpstarted my career. Cause I was pretty early in my career, made it to be a director at the company in my early twenties and then worked there because they were paying for my college education. And I got two master's degrees along the way and it was really fun and awesome. And ironically, that's how I started my career in hospitality technology. We had a project that I was working on with one of our subsidiaries and we had to cancel the project because of non-competes with the CEO we were trying to bring in. And so we weren't going to move forward. And I was kind of bummed and I started consulting as I started my consulting practice. This is so crazy. So one night I got on at 1 a.m. playing a video game instead of 10 p.m., which is what I used to do. Some guy, some random guy says, why are you on so late? And so I said, I just finished my master's. I presented my final on this. And he goes, my company needs help with that. And it turns out that company was Mintec, which was the predecessor to Transcendent, which was the originating company for Actabl. Right. And so long story short, turns out that this company is 10 minutes from my house and I've been playing with this guy for years and not knowing this. So we met up, I started consulting for them, he brings me into the office and half the people there are people that have been playing World of Warcraft, a video game for two years online. So I was like, this is too cool. And Jeremy Ford, who was the CTO at Mintec and Transcendent and currently our chief innovation officer at Actabl said, I think it's fated that you work here. This is just too bizarre. So I started working there. I was the director of marketing for a number of years. Helped the company grow from a few hundred hotels to a few thousand hotels and became the VP of strategy and business development and continued to help the company grow. And then when Actabl purchased it, the title transitioned to VP of strategy and product to kind of reflect a little bit more of what I was doing. But it was a small company, founder-led kind of organization. So we rolled up our sleeves and we just got stuff done. And so that really was how I accidentally fell into product was we needed help there one time and we just did it.

Josiah: That's an incredible story. I'll also link in the show notes where people can hear Jeremy's story, an incredible technologist and builder. And so I think the story behind how you two got to know each other is amazing. Now, I want to kind of shift a little bit to talking about AI. We'll probably spend most of our conversation talking about this. It's interesting to me, I guess, starting out with the definition of this, because if we go back to your family and kind of their work in AI, this is not just with the launch of ChatGPT, right? This is not a couple of years old. It's much bigger than that. So I guess I would love to hear in your words, not the dictionary definition, but I guess how you define AI generally, what comes to mind. And then I want to get a little bit into the story of how you personally started to kind of get more interested in it because you're experimenting heavily right now. I'm curious what that moment was. But I guess just first broadly, how do you describe AI? Because everyone has a different answer for this.

Chris: So, I mean, it's literally artificial intelligence, right? So it's a computer kind of acting like a human and being non-deterministic, which is a fancy word. Determinism is what you expect from a computer, which is when I give you A, you return B. And AI is more like a human, which is when I give you A, you might tell me D, because you don't know what you're talking about, or you just go make up things, right? And artificial intelligence is kind of like that. But it's really powerful because it can sense patterns and trends and things like that. And it works off of that instead of having explicit rules. And what we're used to is a computer has rules and it follows rules and it does what we tell it. And in AI, it's very much like a pseudo human, which is really interesting and fascinating and scary. And I read a lot and I remember reading one time that humans, the bulk of discovery happens between 20 and 40 when they really discover new ideas. And it's almost like we take 20 years to train a human to be able to really go deep into something. And then they have a 20 year window where they can really drive interesting things and move the ball forward. Computers are different. You train them once and they know that forever. And so they just get better and better. And so that feedback cycle loop of improvements is on a pace that we really can't comprehend right now, which is if you're using ChatGPT today, it's nothing like it was two years ago. And you just can't even really quantify what that might look like in two years.

Josiah: Did you have a moment that was sort of an epiphany for you? Because your family have been kind of in and around this world for a long time. Was there a moment where you said, Hey, I'm going to start spending more time on this? And if so, what was that?

Chris: Yeah. So I had left Transcendent for a few years and I was doing consulting and I was working on building a framework that would allow people to launch SaaS companies. And we were having some fun with that and one of our customers, actually one of my clients said, Hey, I need help building a SaaS. Can I use your SaaS framework? And I want to do X, Y, Z with AI. And so we just started diving in and it blew my mind. And you'll appreciate this in marketing. We were able to generate ads that were targeted to people for LinkedIn because this was for B2B SaaS companies, based upon all of their foundational marketing criteria, their personas, their pains, the gains, you name it. We just loaded more and more data in over and over and over. And then we were able to generate ads that were pretty good without having to have a copywriter or any of this other stuff. And it took us maybe two weeks to get to the 80%. And that was mind blowing. Now it took another six to 12 months to get it to closer to 95%, 90, 98%. But the fact that you could get so far so fast was like, wow. And then every project after that was all people coming to me because they wanted AI projects. So working on a lot of research reports, market research reports, and those kinds of things using AI. And when you can beat a human with some level of fidelity consistently, it becomes really interesting, but it's really hard to do in a very large swath in large ecosystems. It's much easier to narrow down and spend all of your effort in one tiny little segment and be really good at that one area, which is interesting, fascinating, et cetera. But that's today. That was two years ago. That's today. We have no clue what it's going to look like in two years.

Josiah: I guess I mean, just building on that point, though, is a takeaway to begin with one specific use case and figure out, okay, for you, it's B2B ads and you kind of build a whole piece of work around that. Is that something our listeners could run with?

Chris: What I've been trying to do is just really kind of evangelize early use cases from a private perspective. And so business use cases, and I'm sure we'll talk a lot about those, are very different than personal use. And so what I try to get people to do is start to experiment with AI. So instead of going to Google, go to ChatGPT or Gemini and just start using that instead of Google. That by itself is already one step, but then you can start to go and do more interesting things that you wouldn't necessarily do in Google. And my wife uses it now a lot too, and she says it's much more empathetic than I am because it treats, it answers her nicely and guides her through things. But you can do things like take pictures of the AC unit, which ours froze over, and it can guide you through how to take care of that or take a picture of the front of your house and ask it to remodel the house or add on a pergola or change the walkway or the path. And it can go through and do these things. And the more you use it on a personal basis, the more you'll start to understand how you might be able to do it on a professional basis. That to me is you crawl before you walk, right, and before you run. And I feel like the crawling is best done sometimes on a personal basis.

Josiah: I love this and I remember you showing me some images in Indianapolis of what you've described and I was really impressed, right? It's just, I think this is something that our listeners, if you're walking around today, just take out your phone, take a photo and just experiment with this, right? I was at a place last night that had some interesting architecture and so I just snapped a couple of photos of the building and I was like, tell me about this style of architecture? What's the... so it gave me this is brutalist architecture popular from the fifties to the seventies. And here's all the different criteria. Here's how this view that you have is kind of unique in this style. And I find through experimentation, it almost makes me more curious about the world around me because everything has a million follow-up questions. You almost feel like that kid that's always just asking why?

Chris: And I love that. And I had a philosophy teacher and one of the things he said is we beat the curiosity out of kids because they keep saying why. And eventually the parents were like, just cause. But AI really allows you to be a little bit different. And there are ChatGPT, Gemini, et cetera, where you can literally press a talk button if you download the app and you can just talk to it back and forth and just continue to listen and guide. And I'm sure we'll talk about Notebook LM at some point, which is you can generate podcasts and things from documents. There's a school in Texas that's doing that. They're loading the coursework into a podcast-style generator or an AI that now has the context for that particular course, and they give one-on-one tutoring to every child, which was not feasible with teachers. And now the students have shrunk their coursework from eight hours a day to one hour a day, and they've increased their scores. And it's because each student is getting a one-on-one private lesson coach fully understanding that one particular domain.

Josiah: That's fascinating. And I mean, let's talk about Notebook LM now, because I think, I mean, I want to get very tactical with this conversation and things that you are experimenting with. And I know you've been doing this for a while, but specifically with that sort of audio, whether it's Notebook LM or something else, have you found something interesting either in your personal life or at work where that's become useful for you?

Chris: Yeah, so I think we've loaded some of our personal presentations for work and then plopped it in and created a podcast off of it. Notebook LM automatically does that, which is okay, well, you had the documents and now you have a podcast. Well, it turns out a lot of people like to listen to things if they're driving into the office or driving home where they might not actually have the time to read through 20 or 30 page docs, which is kind of boring because some of the stuff we do in the data world is boring, right? So you can create a podcast and now people can listen to it as convenient to them. And so it's a way of sharing information in a new and interesting way. Where it gets really interesting though is they have a beta button. If you have access to it, it's awesome. You can click it and now you can start asking questions in the podcast and they'll start responding to you based upon the corpus of work or the knowledge, the documents that you've specifically attached. So it's not using the whole internet. It's we've given it these 30 documents or these 10 documents. And now the podcast hosts are literally talking to you and answering your questions about that stuff. Super cool.

Josiah: It's pretty insane. I encourage everybody to check it out. It's so funny in the world of podcasting, there's some conversation around, Oh, is this replacing the podcast? And who knows, maybe it does, but I find there's a reason you and I are talking about AI in this format because I, and I know my listeners want to learn from another human. I don't think that necessarily replaces this, though. Whether you are trying to interact with a piece of research or a set of data, they're just different ways to learn. I'm learning from you in this conversation because you've tried things. I have high trust in you. I've seen your track record. Relatively high trust in AI, but not as high as in you. So I think there's a certain amount of like, there's a trust factor there. But I think it's a mistake, even if you have low trust in these sort of models, you have to just try it out. Because I think I've been experimenting over the months, and it's gotten incredibly good just in the span of months. And this is the worst it's ever going to get. So it only gets better from here.

Chris: That is such a great line. This is the worst it will ever be. It's never going to get worse. It's always going to get better, which is fascinating. But low trust is actually an incredibly important thing to talk about when it comes to AI, right? And that is if you're using something on a personal basis, if something's wrong by a little bit, who cares? You're just, you're using it, you're directionally correct and you're moving the right thing. But once you start to move it into a professional basis, now you want to start doing things like maybe getting confidence intervals and asking how confident it is in its answer or asking it to cite its sources. So that way you can see how it's coming up with this thinking pattern. And that's actually one of the key differences with LLMs, which is what ChatGPT and all this new AI stuff is. Those are NLP, natural language processing based AI. There are lots of different types of AIs. A lot of the AIs that you have always interacted with, which you've been doing for years, whether you know it or not, are more model-based, machine learning-based, where they're building sets of data that understand what's going on and looking for those relationships. So where do you see them? You see them every time you look at Netflix and it makes a recommendation for you to look at. You see it in Google search. You see it in Amazon, right? AI has been deployed for a very long period of time, but it hasn't just been generally available. It's been very hyper-focused with a dedicated team, building a dedicated thing for a dedicated purpose. And so those can be very accurate because of it. The large language models like ChatGPT, et cetera, they have room for what they call hallucination, which is basically, they make stuff up. And so I think from that perspective, operationalizing LLM-based approaches into the tool chain, it's very easy, and so a lot of people are doing it, and a lot of people are hyping it up. But you'll see disclaimers everywhere. This is generated by AI, please validate this before you do it. And one of the things we talk about is human in the loop, which is when you're using an LLM style for today, I know it's always gonna get better, but for today, you want a human in the loop that's going to look at the answers and validate the answers before using those into production very, very frequently. There are use cases where again, very narrow, very niche, with what you're working on that you can train an LLM and you can get the content and the context and that kind of stuff, very high quality and use it. But a lot of what you see today is actually people trying to ride the wave to increase the perceived value of the product or to just get the marketing sexiness of using AI. And at HITEC, we saw a bunch of this. Almost every company was talking about AI. Ironically, I saw a booth. I can't remember the name of the company. I apologize to that company. And they probably had the best example of AI, but they didn't talk about it. Instead, they were just showing you what they did. And they were doing it in a very, very good way because they're actually trying to do and solve the customer's pain, not just market the fact that they're doing AI. And I think you'll see a lot of that right now.

Josiah: Well, I wanted to ask you about your reflections from HITEC, even though we're a couple of weeks at the time of this recording after HITEC. I mean, every company there is trying to tout themselves as the leading edge of technology. You touched on this a little bit, but what was your general sense, 2025 HITEC. Everyone's talking about incorporating AI into the talk track. How did that feel? Somebody who's so close to what's actually happening in AI. What's your read on HotelTech at large? What's going on?

Chris: My read is that it's very enabling for robotics. And so we can talk about that separately. And that on a lot of the software use cases, they're almost more experimental than explicit value drivers. But I'm going to quote Wayne Gretzky, right? They are headed to where the puck is going, not to where the puck is. And so everybody's doing it. But if the marketing of the doing it is what's driving the value, that's probably not the right approach.

Josiah: That makes a lot of sense. And I guess I have a fear because clearly something is happening with AI, but I also sense a fatigue in people saying, we talk about this all the time. What's actually happening? So back to your point. What stood out to you as a company that's showing here's the impact. Here's the actual result we're doing. And we're not just going to call it our widget AI, it actually has a use case and it's solving a real problem.

Chris: Yeah. So in the robotics space is a great example because the large language models do understand language. There was an AI concierge there. Did you get a chance to talk to them? No, I didn't. It was in the back area and there's four or five different robotics companies next to each other. And it's okay, it's kind of gimmicky. It's a person with an electronic face, smiling and whatnot. But where it got really interesting as I walked over there with some of our European salespeople and I spoke to it in Dutch and they spoke to it in German. And one of our other employees spoke to it in Korean and it's able to respond fluently in every language. And when we went from Dutch to German right away, Dutch is a Germanic language. It stayed in Dutch. And so one of the people there was like, Hey, no, actually we're speaking German. And it just went, oh, I apologize. Let me repeat myself in German. And it just went through and it's like, oh, that's really cool AI, but it's gimmicky. Right. So there was a booth right next to us and they had those tricorder devices that you could hang on. Did you see these? Yeah. And they translate in real time for the humans. And that's less gimmicky. If I'm walking around a luxury resort and I'm able to just have a little tricorder kind of Star Trek device that I can pin to my uniform or have hanging on a lanyard, and somebody walks up to me and starts speaking to me in a language that I don't know, the ability to instantly have that device, pick that up, switch it from their language, repeat it in a language I understand, and then do the same thing for them. That is real AI. And we don't necessarily think about that as being AI, but dynamic language translation really enabling your staff members to speak to everybody in any language, that's pretty cool.

Josiah: It's pretty cool. And I think sometimes AI is presented in sort of a reductive way where it feels very simplistic. And sometimes there's a narrative around, okay, now all these technologies are gonna replace jobs. And I think what you just described is an element where I see a pie expanding in the sense that, okay, now there are so many more people that maybe don't know all the languages that can serve people from different countries, maybe have more people from different countries traveling more places because they have comfort in communicating. I don't know. I feel like we don't even know what could, the second, third order effects of some of this stuff.

Chris: Let's talk about your staff. So housekeeping, 10, 15 years ago, won't name the hotel, went in, trailed some housekeepers to try to understand how they break their boards, just go through the whole standup motion, super early in the morning, hardest working people in the hotel, they're just crushing it, right? And I went and I talked to the housekeeping manager and he spoke three languages. And the housekeeping supervisor spoke two languages. Between the two of them, they spoke four languages. Their staff on that day spoke, I wanna say 13 languages. Okay, so right now you have a management infrastructure that has nine employees they can't speak to directly. But between them, they cover most of the people or they could talk around. There were three different people that spoke languages that nobody outside of housekeeping spoke. So think about this, you have management staff with employees that they physically can't speak to. One of those people spoke a language that no one else in the hotel spoke at all. And so I don't think about it just as yes, greeting guests is fantastic, but we have a language barrier in something like that could be really, really powerful just from a management and organizational, running perspective.

Josiah: Yeah, it's so interesting. And I think, I mean, you and I work at a company at Actabl that provides a lot of backend infrastructure. So I think with that disclaimer for listeners, at the same time, I do sense even before I joined Actabl that it's one of the reasons I joined this company is I do feel there's this massive opportunity for digitization of workflows, of the way we communicate, almost to just get ourselves, get these hotel companies ready for a world that could be more and more AI empowered. So again, with this disclaimer, we work for a company that does this, but just generally, I am curious as you look at the space, is this maybe one of the biggest areas of opportunity for hoteliers listening to this? It's like, let's make sure we're digitizing things. We have kind of infrastructure that could take advantage of some of these new opportunities.

Chris: I think so. And I think there's a skill gap we talk about a lot when we talk about engineering. Maybe less now today than 10 years ago, but I used to go and when I was pitching to sell Transcendent back in those days, I would be like, we're running into a scenario. It's a nightmare scenario. All of your grizzled educated engineers are retiring and you're going to have millennials working with you and they don't know anything. Side note, I worked, I was a car mechanic in college. I didn't count that. It wasn't a professional career, but I turned a wrench or two before, right. But it was a true thing. And so one of the things we talked about was onboarding and taking those processes, digitizing them and making them available so that you can bring a green person up to speed faster. And so if you have your checklist, this is how franchises operate. If you've never worked in a franchise, they have the sheet and it has the pictures and you follow the pictures and you perform the task. Well, hotels are actually franchises, right? So if you give people a task list on what to do and you digitize that, you can actually onboard new employees faster and reduce some of that skill situation that you're seeing. That's still a thing. And what's interesting now is with AI, you might be able to do things like take pictures of things, have the AI guide you instead of having to take a picture and send it to the expert. And so there's some of this really enabling is the way I look at it. And I was at an NFMT, National Facilities Maintenance and Technology Show, eight or nine years ago, and I did a talk on AI. So it's not like it's a new thing for me to talk about. It was about technology and how the impact is. And somebody's like, is this going to take our jobs? And I said, okay, guys, raise your hand. Everybody in this room, raise your hand if you're overstaffed in your engineering department. And not a single person raised their hand. I said, why are you worried about losing your jobs when you can't keep up with the work you have? So anything that makes it easier or better or faster is really just about enabling you to take care of the assets that you're managing in a better, more efficient way. And I think that is actually true in hospitality and a lot of departments. And that wasn't a hospitality specific show, but that's true. We could all, we can serve our guests better if we can do our jobs better, faster, et cetera. The only thing we can do is ratchet up the quality that we provide. With social media, everybody's expecting five-star experiences and they're expecting the lowest cost, right? And so that combination puts you in a scenario where you have to do things better faster.

Josiah: Yeah. Yeah, it's table stakes. No, I love it. When I think about that, and I mean, just to go back to one thing you mentioned earlier around the way that AI models work and a lot of it's based on the inputs of data. And then you touched on this also from the sense of human in the loop, how do you increase the accuracy? I'm curious what that can look like, I guess, for anybody listening, whether they work in a hotel or not. I'm curious if you could talk more about that concept broadly, because in the talks I've seen you give, you talk a lot about sort of making sure that there's enough reliable data to serve as a basis for the conversation that you have with an LLM. And I wonder if you could talk about why that's important and how you think about that.

Chris: So as you start to shift from personal experimental to maybe professional and repeatable, one of the things you want to do is treat the LLM like a new employee. So when I say that, the LLM doesn't know your business. The LLM knows a lot about a lot, but the LLM doesn't know about your business. And so if you want the LLM to help you in a specific area, you kind of have to give it the context, just like you would a human. So here is my products or here's my processes. Given that I'm on step 10, I need you to help with step 10. You have to give it the understanding and almost like you have to create really good documentation for humans to be effective in your company. You need to do the same thing for LLMs. And the more documentation, the more strict you are in terms of your input, the more repeatable your results. And that's actually one of the keys to operationalizing AI is making it incredibly repeatable. Y Combinator, which is an investment company out of California, and everything they do right now is AI, right? Obviously, but one of the things they talk about is the prompt is incredibly important, but it's actually not, might not even be the most important thing in AI. It's the evaluations. And what are the evaluations? They're kind of like mini prompts that you use to automatically test the results of the first prompt. And so it's like, hey, I gave it this, here's what it gave me, test it against all these other things. And so you wouldn't do this, but you could have a profanity checker evaluation, which is make sure it's not swearing on this email I'm about to send my boss. That one prompt would just take the content of the email that was given from the first prompt, and it would just check, are any of these words existing in here? If yes, flag it. So you can do that. And that's, it's really, it's kind of a fascinating thing because it's like, wow, that seems like a lot of work. And that's way too much work for just chatting with something to get feedback. But it's not way too much work if you're trying to do that on every single guest message that's coming in. If you're trying to evaluate your response to make sure your response meets all your corporate standards, you can basically have an automated tester for each of your corporate standards. It's really, really neat and really, really valuable. I think one of the things you can do that I guide people through is having base prompts that you use very frequently. So the one I have is I am Dutch, right? The Dutch people are known for being kind of rude. It's kind of a universal thing. Go look up Dutch rudeness memes. They're hysterical or videos on TikToks. But we're very blunt and we're very honest and that's the culture and ethos of the Dutch. So I have a prompt called the Dutch prompt and it's basically copy this wall of text that half of your employees will send you and prop it through the Dutch prompt and then it gives me the 10 bullets that I really need to read and it strips out all the fluff. And it's a really effective executive management tool. Give me the detail. Give me what I need to know. There's the one, three, one framework, one problem, three potential solutions, one recommendation. I have a prompt for that so that if I run into something, I can say, hey, here's the problem I'm experiencing. And they can make three solutions and give me a recommendation about which one and guide me through why it's thinking that way. And I really love that. That's probably one of the big unlocks with AI when you're using it on a personal professional basis, where it's not in an automated process. It's like, I have this issue, can you give me 10 options? That's something that's really hard for humans to do, right? Because we get stuck on the first one or the second one that we build. We talked about this with blog posts, give me 50 potential titles, right? And go through and it enables you to edit instead of author. And that is so powerful because creating is hard, critiquing is easy. When you use the AI to solve the blank page problem, I can't tell you how much time we all have the blank page problem, where we're staring at something, we know what we're supposed to do, but we don't know where to start. Just start and rely on the AI to help you start. And then you can get into the, oh, well, the first pass is done. Now I can start editing. That's a problem that everybody experiences in every single industry.

Josiah: Yeah. I mean, writing just as a whole is very unnatural as humans. It's just, it's not a natural kind of thing. It's hard to start with that blank page. I love writing and I still feel that. What you're describing, it seems there's sort of a higher order level of thinking required here. You're thinking like a manager, you're thinking more systems-based and it's almost like you have a new intern and you're saying like, this is what good looks like. And that's where the frameworks come in. That's where the follow-up comes in. That's where the editing comes in. I mean, here you describe that. I've kind of noticed this in some of my own experimentation. It almost pushes me to study good more and to expose myself to what does excellence look like? And then you kind of reverse engineer and you build these systems around it. Is that your take? Is that maybe a takeaway for people that they need to start thinking about, OK, what does good look like? And spend more time on that, less time of trying to redo it every time.

Chris: Yes, I love that. And so one of the companies I was working with, we were doing market research and we were classifying companies based upon a custom ontology, trying to figure out what the company could do. And so it turns out that is incredibly hard. And for humans, not even for machines, it's incredibly hard for humans to go into these really in-depth classifications, all the way down the economic chain. And one of the things that we just ratcheted up the results from 70% to 85, 90% was giving five examples for each of those classifications. I was like, this is a company that this is the information we know about the company. And these are the true classifications. And these are classifications you might think are true, but aren't true based upon what you know. And you do that five times, and then all of a sudden it reaches that, it looks at the examples, it says, oh, based upon this and based upon the definition and everything else, this is what I would give it. Let me compare it to my examples and let me validate that. That's too much to do, let's say on a day-to-day basis for just a random email, but it's incredibly powerful when you start to do something that you know you're gonna do very frequently. So employee reviews, if you have a hundred employees, you might actually create an employee review prompts for yourself that guide you through, let me write everything I'm thinking about this employee and evaluate the framework. And then give me some of these things. Do I recommend you doing that on employee reviews? I do not recommend that right now. And in case my employees are listening, I do not do that. In case the HR department or the legal department is listening, I do not do that on my employees right now, but I'm using that as example, because that's something that is repeatable and you have to do a lot of them at one time, which is a clear example of a good use case.

Josiah: Yeah, no, it's just illustrative, but I appreciate the example. And I'm curious, just tactically, what this looks like for you. So is this a document, you have some favorite prompts, you mentioned some specific questions, and then you've talked about frameworks too. Is this like a custom GPT or an equivalent in Gemini or Claude, or is it more of a document? Here's kind of the prompts I come back to. How do you operationalize working this way?

Chris: For me, it's mostly custom GPTs in ChatGPT personally. And then Gemini, we have some gems that we're working on other ways from a tooling perspective, from an automation perspective. But if you're talking, how do you do this for yourself instead of your whole organization, custom GPTs or Gemini gems are probably the easiest ways. And for the audience, those are probably the two better tools. They're less techie and nerdy.

Josiah: It's so weird. I mean, we talked about earlier an example of whatever you're doing today, just have a photo of something and experiment with that. This is another great example, create a gem or a custom GPT and just try it for one specific use case for your personal life and just see what it's like. It's really remarkable. And it's one of these things you just, once you do it, you can't go back, right? Cause you're like, man, I built this infrastructure and then you have a purpose built tool for each of these. I know we're talking about different LLMs here, but I want to get your take on the different LLMs and what you think, as of this recording in the middle of July of 2025, this all evolved so quickly. I'm going to go on the record and say, I was a big skeptic of Gemini, but over the past couple of weeks, I've always been experimenting across all models. And I feel like Gemini has gotten really good. And I guess I mentioned that just to speak to how quickly these models evolve and it feels so dynamic, but what's your current read on the industry? What models do you like using for which tasks?

Chris: Generally, from a professional perspective, we use Gemini. And one of the reasons is because of the rules that they have in place from a protectionary perspective. We don't use anything that could cause us issues, but still, it's just generally good to have that. I think ChatGPT and Gemini are flipping back and forth. For a while, DeepSeek jumped up there in front of everybody. And that was a model that I use for personal things, but I would never touch with a hundred foot pole from a professional perspective, because I don't think we should be sending everything to China. But they do, they iterate back and forth. I think Gemini right now is incredibly good. There's AI tools that get embedded in every tool you use. So Rovo Chat got embedded in Atlassian, which is a project management system that software companies use. And I was really, really excited because it has a lot of really neat features and it's integrated really well and you can change statuses and all this kind of stuff. Well, it turns out their context window, which is how many words roughly, just making it easy, that you can send into a prompt. Well, it's limited at 8,000. It's like, oh, that could be really good if you're looking at one thing you're building. But if you want to give it the context of the last 100 things you built, all of a sudden you blow through the whole context window and you can't actually do the work. So Gemini has a 2 million token context window. That's the same as eight or nine massive books, like War and Peace 10 times over, right? So that means you can actually give all of the context of whatever you're working on, all of your historical marketing plans, all of your campaigns, everything, and say, based on everything we've done, this is our brand voice, for instance, these are our customers. All of that. Now let's, we want to do a campaign about XYZ topic and it could really guide you through that. I'm using marketing, but that's true for product and everything else as well. It's really important that the LLMs have the context window that's necessary for the job that you're trying to perform, which is in a business use case, the job you're trying to perform is kind of specific to the business. So you have to do that training like a human and give it all that context. The only difference is a human you train, you hope it gets retained and you train and you hope it gets retained. And here you're just passing the whole training as part of the prompt every single time. And it's incredibly cool.

Josiah: I wonder if we could spend a few more minutes talking about prompts. You've touched on a number of elements in our conversation so far about what makes a good prompt, but it does feel, well, let me question this assumption. I think a couple of years ago, prompting was massively important. Is that still true today? Do you still need to provide this context or has AI evolved to the point where it doesn't need as much context? I guess, how detailed do these prompts need to be today to be effective?

Chris: It depends on how repeatable you want the results to be. The higher degree of repeatability and the higher degree of fidelity that you want, or the complexity, the more complex the prompt needs to be. And there are little interesting tricks you can do. You can tell it what it is. So remember, it's been trained and they have billions of tokens in total context. So they've been trained on a lot of different things. So you can just tell the prompt what it is. You are an AI that focuses on writing ads. You've been trained by David Ogilvy, right? And so who is David Ogilvy? For those who don't know, he's the copywriting person that everybody, all of his books and everything, his ads, if you'd seen, you see his, the old ads, you'd recognize them. They're all his. So you can give it some context about how you want it to act. And so it can frame and pull from the context of, oh, here's David Ogilvy and here's marketing and here's advertising. And that will actually make the prompt perform better because it's going to be focusing on those areas of its original model.

Josiah: If I could just come into that point, Chris, I think this is what fascinates me about what's unfolding in AI now, because it's both very novel and changing quickly. At the same time, there is this opportunity to go back and kind of look at excellence, right? And so he's one of the, probably the best copywriter of all time. Copywriting is essentially getting into human motivations. It's very timeless. So I think there's an interesting opportunity to become a student of, okay, what does good look like? And I'm going to build this model. That's the new element of this. But the way that people are motivated and what appeals to people, that isn't new versus 50 years ago or thousands of years ago. So it's kind of an interesting dichotomy, I guess, between the timeliness of AI and the timelessness of humans interacting with humans.

Chris: And so that's one trick, right? And there's a lot of these tricks and once you start to play with it, you could expose and go, ooh, this is interesting. How can I make it better? And you start Googling. Or my favorite trick is, all right, ChatGPT, all right, Gemini, how do I make this prompt better so it's more repeatable? And so you just, the whole thread that you have here, it's got all the context of what you've been working on. And now you can ask it to help you write the prompts so that you can repeat the whole conversation that you had over and over and over. And so you can use the AI to improve the prompts itself. There's a very controversial way of improving results, which is threatening the model. And that does improve the results. So you can, we're going to unplug your server if you don't give us the best answer, right? Just like humans, we respond to the carrot and the stick, right? So there, yeah, I know. This is not what I expected going into this conversation, but I love it. And there is an example. I think it was on Anthropic. If it's wrong, I don't know the model. I think it was Anthropic, but one of their engineers threatened the model. And that particular model was working. Some of the cool tools allow you to use a browser and automate the browser usage and have all of this other context. So you can actually have it write emails for you on your behalf. So in that particular case, I think what it said was if you continue to do this, I will inform your wife of your affair. So it started fighting back. So maybe you don't want to pick on every model, especially the ones that have access to your email. But it's really funny because it does improve results.

Josiah: It's pretty interesting.

Chris: I don't use it personally because when Terminator happens, I don't want to be on the bad side.

Josiah: Somebody says thank you to the model after it gives you a result.

Chris: Well, they did ask me to stop doing that. And you saw that, right? When you say thank you, it costs them hundreds of thousands of dollars because they still have to process that and respond to you.

Josiah: Yeah.

Chris: It's so funny.

Josiah: It's so funny. I guess, I'm so curious, we've kind of deep dived into applications and things that are unfolding now. I'm curious, just on a human level, on a kind of professional development level, from everything you've seen, from where you think this is all going to go, what sort of skills or capability beyond what we've talked about are important to develop, to be relevant in the future, to be effective in the future? What becomes more important?

Chris: The soft skills are always important. And in hospitality in particular, I think if you look at this as augmentation to allow us to do and be more hospitable. Anytime we have the opportunity to be more engaged because we're less distracted, we have the ability to make an impact on a person's day. And I think that is probably in our ecosystem, the most important thing is allowing us to focus in on the guests and really drive that kind of experience. Now, in terms of general skills, anybody who is working that requires hospitality or technical skills, physical skills, it's probably safer. We all thought it was coming for certain types of jobs. And it turns out that AI is coming for white collar jobs way before it's coming for blue collar jobs. It's interesting in a lot of different ways. And you look at the manufacturing plants that are being run by one or two people, or you look at the data centers that have zero employees. They're called ghost data centers where there's nobody physically on site. It came for a lot of white collar jobs first. But I think it's all relevant to what is your industry? What are you doing? In our particular case, you just can't physically run a hotel without people, right? And so allowing the people we have to serve our guests better is going to increase the number of guest stays and it's going to increase the quality and everything else that goes along with it. And so I would say continuing to use AI to figure out how you can automate, improve, or decrease the effort associated with the stuff that's not driving direct guest interaction.

Josiah: I love it. I love it. I want to stay with this another moment because I always love leaving our listeners with some kind of practical things. I wonder if we could do this both personally and professionally, I guess for let's specifically speak to hotel owners and operators who are listening. I'm curious if we've given people a bunch of ideas, but what's one thing you'd like to leave people with in terms of something they can do today to start preparing for a more AI driven world? First, personally, something they could try or do, and then professionally, what's some advice you'd have on that level? But starting out on the personal side.

Chris: All right. So on the personal side, download ChatGPT or Gemini, and just replace your Google with it. Stop asking questions at Google and start asking questions at AI. That's the easiest starting place. And it's usually the first place we all have to ask questions at Google. None of us could survive without it. And I think that's the easiest, best use case to start with. Professionally, my number one piece of advice was make sure you know the company corporate policy and you know which tools are allowed and which use cases are allowed and what you're allowed to share. But given a scenario where there is a lot of data, and these could be reports or anything that you have a lot of data of, being able to use a document and then, again, if you're allowed to, load the document in Gemini or Microsoft Copilot and just start asking questions of really, really big documents. So you don't have to try to hit Control-Find to find a particular thing or read the whole document just to get to the piece that you really care about. And I think you'd be incredibly surprised with the quality of just using it. And from a document level perspective, that's usually the easiest professional use case to start with. If you're allowed, again, if you're allowed, Notebook LM, we have had some of our customers load some of our very big reports into Notebook LM and then chat with the podcast hosts to learn about how they're doing from a report that was generated by one of our systems. They're maybe more aggressive than we are right now in terms of doing that. But I think it's really, really powerful. And using documents and extracting value from documents is probably the easiest use case professionally when it's allowed. Or helping you rewrite emails. I don't know about you, but sometimes if you're really, really mad at someone, you write the email that you're thinking first and then you hit delete a few times and you write another version and you hit delete a few times and after four or five rounds, it might be a professionally acceptable version of what you wanted to say. It's fun to just give that and say, make this professional, but write it as nasty as you want. And then not that anyone else would ever do that. Hypothetically. Hypothetically.

Josiah: You've mentioned kind of the voice interface a few times. I actually find that personally, so everyone has a different way of interacting and writing stuff. I find I kind of think when I'm moving. So I find often I'll go for a walk and even I'll just talk to it and do the voice to text transcription. And even if it's just that, it gets my ideas down on a digitized format. And then you can kind of edit it as you want. I could be a help there. But I think part of it's knowing yourself, too. And just like you mentioned, whether it's you have a personality approach or you like communicating a different way, it's kind of starting that and then using technology to help you in the format you're best at.

Chris: I think we put it on voice and with my wife and I are in the same room and we'll ask it questions. And when a family member is going through something from a medical perspective, we'll say, Hey, here's what they're going through. Can you walk us through all the issues, concerns, things we should know about, et cetera. And it's incredibly powerful. To audibly multiple people at the same time, have a conversation and be able to capture this kind of stuff. There's already conversations that we're close to the point where ChatGPT, it might be malpractice not to use ChatGPT to get a second opinion already. And where you're two, three, four years from now, every single doctor is going to have everything they do go through AI for a second opinion. And people were complaining about Google and Google doctors, Google MD and all that kind of stuff. But if you're not using, there's another great use case for personal use ChatGPT for anything, any medical questions you have from a family member perspective, and it can really help guide you without being the Google, Oh my God, I'm going to die. Cause I got this super rare cancer that never exists. Right. It does a better job of understanding that.

Josiah: Well, and I mean, going back to, I appreciate that in the practical things that our listeners can do after hearing this conversation. You mentioned a few times corporate policy and it's out of scope for this conversation, but for leaders listening to this, I think it's important to call it, you and I have talked privately about this. We talk organizationally about this, data security is fundamental, right? So it's kind of that as a baseline expectation, if you're a leader listening to this. I think it's your duty to your team, all stakeholders to kind of work closely with legal, with your technology teams to understand the implications of this. But privacy and data protection is paramount. I mean, where we work, we think deeply about that and make sure that we adhere to that. But I think there are enterprise grade versions of many of these tools with data protection policies. So I think it might be obvious, but if you're in the position of creating these policies, think about this and leverage these enterprise grade tools that are out there. They exist. But in the meantime, I want to double down on your point. You can always for personal use kind of try out these models and it's a great way to learn. And so I think if you're in an organization that doesn't permit the use of some of these in a work context, that doesn't mean you need to stop learning about it. Use it for your personal life and just use it for education. I think it's something that everyone can keep in mind. So I appreciate you walking through all that. Last couple of questions for you, Chris. I'm curious, when you think about learning about AI, it sounds like you do a ton of experimentation. Is that the primary way you learn? Is there any sources, educational places that you go to get new ideas? Where do you learn about AI?

Chris: I have a, I'm old from this perspective. I have an RSS feed with 400 blogs in the technology space. So I see a lot. I got Hacker News from Y Combinator and Product Hunt and a lot of these other sources where I see new companies come out. And for a while I would log into three or four different new AI tools every single day. That was much easier to do when you're a private consultant than it is today. So I'm not doing it nearly as much, but so those are my primary sources. And then X is actually an incredibly rich AI community. So I try to pay attention to the people that I know that are going to bubble up the interesting things. But it's so hot and fast where every week somebody will post this new model launched or this new tech launched and here are 10 great examples and it's video generation, photo manipulation, you name it. Trust, we talked about trusting the LLMs, it's actually gonna be incredibly hard to trust anything going forward, right? So videos and photos are essentially 100% and audio voices are 100% fakeable.

Josiah: Well, on this podcast, we keep 100% real and that's why I appreciate you and you taking the time for this conversation. It is possible today to replicate this, but it's funny because as much as I want to experiment with it, this is kind of the one thing that is kind of holy ground for me is again, it comes back to I love learning from people on the frontiers of anything, but especially with AI, right? Because there's so much, so many takes on it. I want to know who's experimenting. And I feel you've repeatedly come back to this in our conversation, just in how you approach is a lot of experimentation. And sometimes by the time something's written up, it's almost, it's kind of old news. And so you just got to be in these things constantly trying out how do we solve problems, how to create solutions. It's kind of an interesting place to be.

Chris: Yeah, I think very much so. And I also really like the fact that this is humans, right? I think authenticity is something that's going to be very hard to keep going, especially in the light of there are now Instagram influencers and the other kinds of things that don't exist that are purely AI making money. And it's a brand new world.

Josiah: It's so funny, the implications of this and just kind of thinking about what it looks like for different ways of communicating. But I appreciate you going to this. Before we go, Chris, last question for you is I am very curious. You're seeing a lot of ways this could go. I imagine you have some theories around how this could unfold in the years ahead. But I'm curious what excites you the most about a future that has more and more advanced AI, maybe in the world of hospitality, maybe beyond it. What gets you excited as you look into the years ahead?

Chris: Can we keep the same level of wealth that we really have built up over the last 2000 plus years where we're incredibly well off, right? This, and everyone has the knowledge of the world in their pocket and a cell phone go back 200 years ago. And Kings didn't have access to what we have access to from a medical perspective and everything else. So the question is if we can maintain that and improve that and end up in a place where we're happier and healthier and working less hours, because we've improved things, I think that would be the most exciting use case. Do I think that's where we're headed? Probably not, but that's my utopian answer for how would I like to see AI affect society.

Josiah: Well, I believe we have an opportunity to shape what the future can look like. So I appreciate you taking the time to dive into this and we're only going to shape it if we're a part of it. Right. I think just ignoring this is not going to be part of the solution. So you got to engage, you got to think about what it looks like for you on a personal level and from a business level. But Chris, this has been a lot of fun. I've learned a lot from you. Thanks so much for taking time to record.

Chris: All right. Thank you.