Decisions Beat Dashboards: What Hotels Get Wrong About AI | Ep. 002

Hi guys, so today I'm with Jason Norona, the founder of D3X, and today I want to explore together the future of data, human, and those kind of decision making that's the most important. So my name is Thibaut Catala, I'm the founder of Catala Consulting, and this is HXT212. HXT212. Alright, Jason, thank you so much for being here.

Thanks for having me. Thanks so much. Before we start, is there anything on your mind right now which is not AI or tech? What do you think about what keeps you up at night?

Two kids. Two kids. Okay, that's uh that's really keep me up at night anymore, but yeah, it's there's a lot going on in life. But yes, obviously like AI is taking a lot of space and uh a lot of oxygen out of the room.

Yeah. Okay, yeah, no, that's uh interesting. And can you tell us a bit more about uh about uh D3X and uh and uh what you do at uh D3X? Absolutely.

So we're focusing on verticalizing AI for hotels and attacking customer service. Okay. So we're we'll I come from an operational background, so I ran properties. We then built a PMS.

So using that knowledge of how the PMS works and how operations work, we wanted to combine the two together and use AI to automate some of the operations and automate workflows behind the scenes. Okay. Does it replace a human? Uh partly, yes.

Yeah, but at the same time, I believe organizations have a lot of work to do that they're probably not doing. Okay. And I think by AI handling some of the routine jobs, it frees up the time for humans to add value in different uh areas. Okay.

No, that that's an interesting point because um, you know, I I feel that hotels, I mean, first of all, a lot of hotels that don't have much tech in place, uh, but the one who are looking into tech, AI, and and data and so on, they keep adding more data, more tech, more AI and so on, because for the sake of uh innovation as such. Um but do you think it improves the decision making? Do you think it improves the operation, or do you think it can just unfocus yourself on the on the true spirit of what matter, actually?

Yeah, I think they're all very different things. So I'd like to break it down one by one. Firstly, do we have sufficient data? Probably not.

We can always have enough. Yeah, we can always get more data. Okay. Uh we could uh get a lot more data points.

Like, and the simplest thing would be like personalization, for example. I can guarantee you that none of the hotels I've ever stared at knows that I like filter coffee and not an espresso. And nobody knows because nobody asked. Okay.

So that's just a data point that's completely invisible to the hotel, for example. So when it comes to personalization, no one even knows where to start because they don't have that data point. So can hotels get more data? Absolutely.

There's a there's a way to get your hands on that data as well. You just can't blatantly walk up to someone and say, hey, what how do you like your coffee? And they just be like, Why should I tell you? So I think firstly, yes, we can always get more data.

And secondly, can we get more tech? Yes. Broadly speaking, I believe the world has a shortage of technology. And that could be explained by a multitude of factors.

Like, how do you file your taxes? Do you even file your taxes like manually? Or does a software just handle everything for you? Does uh software manage the heating in your room right now?

Do you control it on your phone? Probably not, you know. You could, but you probably don't. So I believe we generally have a shortage of technology across the world.

And that's probably caused by the cost of engineers and how much work there is for engineers to do. So, yes, I believe we could get more data. Yes, I believe we could have more technology, but it doesn't mean that all technology is valuable and good. So there's obviously a lot of technology out there that's probably useless.

A lot of data out there that's also probably useless. So it's it's about like kind of deciding what's right for you and getting your hands on that. Yeah. You know, do you know about the um what's the name?

It's um the Moore's law, yeah. Uh, which I don't remember exactly what uh what it is. Yeah, every 18 months uh the cost of uh yeah uh data storage was something like that. Okay.

And uh so do you believe this more this more uh law applied to tech? And uh it does it mean that uh by having uh more tech and more advanced tech, the cost of the tech will reduce and then the the tech and AI and data will become much more affordable, and that's why we'll get more and more tech uh to the bigger, larger public. 100%. And that that happened because of the LLMs, because now the LLMs can write code.

So what that means is you don't need to pay an engineer like a couple of thousand to write code for you. The LLM can write the code, of course, maybe guided by an engineer right now, but it just definitely reduced the cost of writing code. And because the cost of writing code reduced drastically, we should be able to get more code uh across the world. I see.

Okay. And um going back to my question about um decision making, um, do you think more data means better decision or not? Uh well, it depends, you know. It it really depends.

It depends on what we're looking at. So but I think data-driven decision making is a bit of a meme at this point in time. Okay, it's a code way, code word of saying like um no one wants to take ownership for the decision, so let's just kind of look at the data, and the data's gonna give us all answers. Okay.

But I I I think the best leaders have some kind of opinion, they have a goal in mind. So they then use the data in purpose or in pursuit of that goal. So if you kind of start off and say, oh, we're gonna follow data-driven decision making in everything, I think the data can tell a lot of lies. Because I I come from a consulting background where I worked at Accenture uh consulting a lot of like very large billion-dollar companies.

And we processed a lot of spreadsheets. And many times you realize that the decisions have been made before you kind of process the information. You just use data to justify your exactly. It's like you're covering your ass.

Like this I'm I don't know if I'm allowed to say that, but yeah, that's that's the kind of approach you have in larger corporations. So uh and especially in this AI world, you've got of course you've got data, but you've got a lot of assumptions beneath that data or behind the data, and those assumptions are changing very quickly. So when ChatGPT releases, I don't know, GPD-5, maybe something changes. So uh currently, maybe vibe coding is not feasible.

Maybe vibe coding becomes feasible. Uh, Claude released a new model which has substantially better results when it comes to writing code. What does the data tell us there? Nothing because the assumption changed.

So I think we're not in a position to re-u-evaluate those assumptions quick enough to then have a data-driven decision-making framework in place. That's super interesting. That's very interesting. And uh I'm going to bounce back on this, and uh, I know that you you don't want me to go there, but you mentioned something about uh spreadsheet um lying.

Like you you can you can you can you can lie with data. Yeah. Can you exp can you expand on this? Well, yeah, if you have sufficient data, you can kind of slice and dice it and look at it the way you want to look at it.

And that's kind of yeah, that's maybe a bit dishonest, but it's definitely defendable. You can defend your decision making. And so the dishonesty lies in the assumptions that you make when you look at the data. So you could say something like, Oh, I'm not gonna look at those cases because that uh doesn't suit my purpose, if that makes sense.

And so, I mean, it we can we can look at this at scale. So the the biggest, I would say, um, the most scandalous uh piece of inform like uh episode I can think of is like when when we had COVID going on and there was so much data going around and nobody could agree on anything. Someone claimed the data said this, someone claimed the data said that, someone said that was real and not real, and we couldn't agree with each other, and that's because the assumptions were not valid.

I see. So it's all started by having the right assumption and uh or maybe not having some assumption, looking at the data in a true form of data, and then looking at what the data is telling you and not what you believe the data should tell you as such. I think the data can tell you something if your question is very narrow. Okay.

So uh then the question kind of takes care of the assumptions for you. So in um in a very basic way, um, because I'm a very uh simple person, um, if I want to look at my data and I want to have the the the best output and the best decision making out of my data, where should I start? You mentioned about maybe starting with the right question. Yeah.

So where where are the what are the fundamentals to start with uh with um uh getting the best output and the best decision based on your data? It's like what do you want to achieve? What's your what's the goal of your organization? Do I want to uh and you kind of know this as a leader?

And I think most of us have this thing called the I don't know, like intuition, gut feel. And that's also data-driven decision making, but it's multivariate. It's factors that don't go into a spreadsheet. So things like you you walk into a hotel and it feels wrong.

What does that mean? It's like it's data, it's data points, but it's also like how it looks, how it feels, uh like what how people are behaving with each other, the the ambience, the sound, the smell, and all those points don't eventually end up being measured. Yeah, so yeah, so we can use our gut as well to kind of decide what happens. You know, it's kind of the forbidden word uh in a data-driven organization when we we say about gut feeling.

Yeah, because I feel a lot of organizations right now that are trying to suppress gut feeling, they're trying to suppress the emotion and they're trying to look at the data uh as much as possible to take the decision. But I agree with you saying like the good angle and the best decisions are taken from the data and from the gut feeling, the emotions. And if you manage to get those two elements in balance, you get the best output. Yeah.

What do you think of this? Yeah, I think so. And I think people are might not be incentivized to use that data or that gut feel because there's a big penalty for being wrong and there's no reward for being right. Yeah, and that's what happens when yeah, you're not incentivized uh or your incentives are not aligned with the business.

It's easier when you're the owner, the founder. But what if you're like middle management and you might have a gut feel about something, but if you're if you're right, great, like you get a little bit of credit, but if you're wrong, you're gonna get like yeah, crucified. So no, but that's very interesting. Like, um, I I still believe the human angle, um, at least um today, uh, is still very important when you use um any LLM, any any systems, um to make sure you think critically about the output, about the what the data are showing you, or any kind of generative AI output.

Uh so the human is still important. Do you believe the um so what do you believe AI is actually doing better than human at the moment? And what do you believe are the tasks that human at better are doing? AI is good at like speed, speed, okay.

So just kind of doing something very quickly. It could be very tedious. It could be like read these ten documents and summarize it in one line. And a human would take days to complete that job, right?

So it's it's really, really quick and inexpensive. You'd have to pay someone like a thousand euros to like read ten documents and summarize that in one line. So it's definitely a speed and it's quite cheap to get that done. Uh the quality of output is also quite high, so and in many cases surpasses a human because it doesn't get tired, it doesn't get exhausted, it's not uh moody, it doesn't question the purpose of the job.

So if you tell a human to like just sit in, sit at a table and reply to 10,000 queries where customers say, Can I check in early? By I don't know, email 56, they're just annoyed. So AI doesn't get annoyed, it just kind of brings that spirit to every single response. So I think it's kind of good with those machine-like attributes.

Where it comes to human, it's it kind of goes back to that gut feel, but it come it kind of goes back to processing data points that are more than just the text. So it's uh and the world is kind of calling it taste. Okay, that's where humans come in with taste and emotions. Absolutely, like so having an opinion on what should happen with this business, yeah.

And uh or vibes, what's what's the vibe, right? Yeah, to think critically, to swing to think critically about the output. Because I feel that um um I don't know about you, but any kind of LLM and um system I've been playing with, they're all kind of um um telling you yes to everything. Or like I don't know if you have seen this meme where they say like um, oh, should I eat this mushroom?

It's a good one. Oh yes, you're absolutely right, it should be very good. And so I think those are then you eat the wait, and then they eat the mushroom, and then it's like actually I'm I'm feeling very bad about it. Oh yeah, you're totally right.

I know the meme, I know the meme. But I think that was a quirk with like one of the Jeep, it might have been GPD 4.0 that kind of got a bit sycophantic and was like just like agreeing with everything the user said. And I believe they've fixed that.

Okay, you know, so I think I think we shouldn't dwell too much on the idiosyncrasies of a certain moment in the conversation. So, for example, hallucinations. Two years ago, everyone was talking about hallucinations. Maybe people are still talking about hallucinations, but broadly speaking, I believe like hallucinations have come down drastically.

Yeah, so just it's not a I don't think the AI agreeing with humans is a core part of the technology, it's part of the feature that one of the LLM companies might have built to appease their customers and to try to gain market share. So I would not dwell too much on LLMs agreeing with humans because you look at Grok, it's not gonna necessarily agree with you, it's a but it's the same technology, right? So I think that's one element of LLMs agreeing with you. When it comes to critical thinking, I think LLMs can be quite good at critical thinking.

Okay. So you can just say, hey, go analyze World War II and give me a nuanced take of why the Russians did what they did. And it will give you a pretty nuanced take. So it will critically analyze World War II and give you some information.

Now it just so happens that it would get censored if it were to talk about a political conversation, but maybe you want a nuanced take on your the color of your hotel, and it will give you a nuanced take on that, if applicable. So I think LLMs can be quite it can be good at critical thinking. Okay. Yeah.

And I think that's the risk that because it's so good at critical thinking, humans are getting lazy or will get lazy and might not think critically. And I guess that's one of my fears in terms of like having young kids and kind of raising them in this LLM first world, saying, will they be lazy and not think critically because LLMs are so good at uh critical thinking? But that's also like kind of going back to calculators, and you say, oh, well, calculators are so good at computing that humans will not be able to compute.

And maybe that's potentially true as well. It's changing. I I think we are we are the same when when when when your parents or grandparents were thinking about uh computers, when they were thinking like, okay, now we're all going to lose a job because of computer and so on. Um, yes, some jobs disappeared, but some other jobs appeared.

So I believe we are now at a tipping point and it will change. Um, kids will learn differently, kids will should learn different things. Uh, and what we are used to is now completely not irrelevant, but it will change. I cannot tell my kids um you have to study this because you're going to become this.

Most probably the jobs that my kids will do don't exist at the moment. Yeah. You know what I mean? And that's exactly this change which is happening.

But there is a risk that we will all lose our jobs. You think so? Yeah, absolutely. It's it's definitely not zero.

Okay. Like if we invent AGI, so that's kind of artificial general intelligence, it's smarter than all of us, it can do anything. People are saying like AGI is already happening. Well, it's difficult to understand what the definition is because different people define it differently.

So, oh do you define it? I I don't. I'm I like we're working on AI for hotels and we're kind of verticalizing it there. Um, and I kind of defer to the leading minds in the fe in the field to define it.

Yeah, but I what what I hear and what I observe is that we are seeing this kind of jagged intelligence, so where it's very good at something and very dumb at something very simple. So you just like, why can't you do that? And humans are very good at doing a lot of things. So that's interesting.

So I I I hear you about AGI. Um, but do you believe at some point people will kind of um I don't know, that's an assumption, huh? That people will at some point reject tech, reject AI, absolutely, and will go back to plant their own carrot and uh potatoes in the garden and uh absolutely a hundred percent. Yeah, a hundred percent.

And you can just look at that from the the vinyl movement and kind of buy I just put a tape uh last week, uh cool artist. Yeah, but yeah, but but that's because uh people are rejecting uh Spotify because they're like it's not fair, it's not equal. Yeah, I'm not gonna publish my music on Spotify, I'm gonna kind of get it on tape. And so people will reject AI because they just say, Oh, I don't like that.

I'm not gonna uh go down that path. And it's very valid in the hotel world because you can't get AI to travel for you. AI is not gonna go put its head in a bed for you. 100%.

So humans are gonna travel, and then some humans will say, I'm gonna build a non-AI hotel where it's real hospitality. It's there is zero tech in it, and some humans are gonna love that. Yeah. So there are two angles here.

The first one, I believe, and we can see it already with some statistics, where we see uh board games, like physical board games being like uh increasing, like people are buying more board games than before. Yeah, and people are buying books even more than before. They want to fill the book, they want to read it. Um, so I think like this kind of uh people are craving those kind of connection, those kind of um physical uh emotion experiences, yeah.

And uh and I think that's uh something which is coming up, and because of tech, because of AI, I don't know, but that's a trend that we are seeing. And the second one, I agree with uh we have this kind of concept of hotel sometime where uh they start to say like we are in a non-tech hotel, and we are you have to leave your your phone at the door, and that's a full kind of retreat and uh detox of all of things. So definitely. Now I do believe the future it's all about connection, about this kind of experience, about emotion, creating this kind of authenticity for sweet IT and the and the hotel industry.

Um, I don't believe it's should we should reject completely the tech or completely the AI, but we should leverage AI, we leverage tech to make our job better, more efficient, for us to focus on the experience, on the human connection, on those kind of emotions. You see what I mean? I think that's what you you you mentioned. So, not to completely it's all black or white, it's not like tech or non-tech, it's kind of the good balance between both, but without never forgetting that we are all about human, we are all about those kind of human connection.

And I do believe the the success of the of hotels in the future will be the one that can leverage tech AI, but focus much more on the experience, the connection, and the emotions. Yeah. I also think we we haven't experienced the full AI wave just yet. So it's not like I'm tired of AI in hotels, but like it's not like I stayed at the last hotel and there was so much AI it overwhelmed me as a guest.

Okay, so I don't think we can get to the anti-AI phase until we get to the overwhelming AI phase. But do you think we are going to get at some point like this kind of overwhelming of uh AI? Well, maybe, yeah, for for sure. Why not?

You see, like um one thing which I I also have a concept on this, the um where where does AI or where does personalization start to get creepy? When you like when the AI start to predict what you want before you even want it, but then you realize actually I wanted this. You know what I mean? Do you think that we are going together?

Well, in order to do that, in order to predict, you need more data. So we kind of go back to this starting point where we don't have enough data. No, the AI is just gonna be guessing at this point in time if if it were to like try and predict anything. Okay.

And of course, then that kind of brings us to like the difference in the AI as well, because kind of predictive AI is not gen AI. Okay. So gen AI is the language AI or kind of the large language models. And predictive AI is more of traditional machine learning where you like process sufficient inputs to predict.

Predict what the next output might be. So you can say I've processed like that's what Google might do. I have seen 10 million people who look like you and behave like you do. So I know that you're gonna like this book that I'm gonna show you or this product that I'm gonna show you.

That's interesting. And uh you you mentioned about uh ChatGPT and about Grok. Um, is there any um what is your favorite system and why right now? I'm kind of hooked on ChatGPT.

Okay, but that's just my personal behavior. Okay. I I'm just in there, I got it on my phone. Um so I believe there is some kind of moat for them because once you kind of start using one system, you're just too lazy to switch.

Yeah. So you kind of know how it works. It's part of my job, so I al I try a lot of systems because I have to figure out what's going on, so I I always use the other systems, but I think we're also getting to a stage where certain systems are getting very good at certain things. So uh ChatGPT is kind of good at like review this contract and do all the B2C consumer stuff, and maybe uh Claude is very good with coding, and so we could see that happening where certain systems get good at their own field.

Very interesting because a few, I think it was last year or a few years ago, I talked to some AI experts and they were telling me like they were big on uh they were betting on Gemini. And at the time I was like, I don't know, like compared to ChatGPT, it's a baby next to ChatGPT. But on the last recent development, Gemini is now a beast. Gemini in terms of of uh of uh accuracy, in terms of uh intelligence, in terms of uh of connection to the Google Suite environment.

I feel Gemini is getting much better. Uh maybe, maybe, but it didn't change anything. So my question would be what changed in the world since Gemini 2.5 released?

So if I look out of the window, it's like business is still running the same way. We're running, we're building an AI company, and nothing changed for us since Gemini 2.5 released. We can do more of the same stuff, like of course, we're working on AI agents and and and loops and trying to uh build uh workflows and stuff like that, but that was possible before 2.

5 as well. Okay, so nothing really changed. Maybe um so maybe they're getting better, but we need yeah, uh we need some kind of step change in order to say, oh wow, this changed everything again. I see.

And uh and what would you recommend? I'm talking about hotelier, but I'm talking about pretty much uh everyone. Um, what would you recommend people to if they want to get started on with AI and uh mostly with AI, where should they start? What should I just get Chat GPD on your phone and get started?

Like just try and like make a habit out of using it. Okay, I think so. Just like just use it and then you'll just find your your way of operating and thinking and working changing. And I think what people are getting wrong or getting lost in is this concept of prompt engineering where they feel like I need to take a course to learn how to prompt the AI to do better.

And I don't think that's true for regular people because I believe Silicon Valley has gotten this far by making things easy and making software easy to use. So they'll just make AI easy to use. You don't have to take a course on how to prompt the AI to use AI because it would not be in OpenAI's best interest. It would be a barrier to adoption.

Okay. So they would just like get rid of that very quickly if they haven't already gotten rid of that. That's interesting. But if you know to prompt, do you think you can increase the accuracy and uh and uh of your of the output of ChatGPT or those or yeah, for sure.

Okay. Yeah, you can just like a simple uh example. I don't know if it's it's true anymore, but this you can just ask ChatGBT to count the number of characters in a in a sentence and most probably you'll get it wrong. And then if you just say, hey, write a program to count the number of characters in the sentence, it would always get it right.

Okay. Uh so this was called the strawberry problem where you just ask how many R's in the word strawberry and would always get that wrong. Okay. And they've kind of fixed the strawberry word now because it's too common.

But that's the concept behind it that you'd need a prompt to kind of get the right output. But honestly, for for day-to-day use, it doesn't matter too much. So just uh talk with ChatGPT the way you talk to me uh today. Like uh just try and experiment and see.

So you're you won't recommend for the um average uh person to go into prompting and uh get an expertise on prompting. Well, just chatting with ChatGPT is prompting, you know. And the way you can get a good prompt is tell ChatGPT to write a prompt for you. So then if you are gonna tell ChatGPT to write a prompt for you, well, open air would just tell ChatGPT to take the user's input, write a prompt for the input, and then prompt the LLM.

So that kind of eliminates the need for you to know how to prompt. And do you have any um I'm getting very tactical here, but do you have any uh any um mystery uh tips or acts on uh on your personalization you do on the on the ChatGPT or any kind of AI? For myself, yeah. Oh yeah, I have a master prompt, uh like a system prompt that I kind of use, but I don't know what's in that system prompt at this point in time.

I know I've tweaked it a bunch of times, but I yeah, it doesn't matter too much. You know, I think what because what's what the next frontier is memory uh for these LLMs. And they're gonna kind of try and understand your behavior and try and summarize your previous instructions to ensure that the system prompt is auto-constructed using memory. So again, Silicon Valley is gonna take care of that for us.

So we're not gonna need a system prompt, it's just gonna learn the system prompt by itself. Okay, interesting. And um yeah, and and going back to the to this angle of on the um on decision. What would you recommend for for leaders in any kind of field um to do today to improve the decision making?

That's a broad question. That's really yeah, it's challenging. It depends on how you're currently making decisions in the first place. So maybe like some self-awareness would be the first step to improve your decision making.

Okay. Uh I don't think just like yelling data out loud like would be the right way. So just kind of understanding how do we make decisions, how many people are involved in that decision, what's the problem right now? Is it too slow?

Are we being too conservative? Are we making too many mistakes here? So, where is the problem? And then kind of fix that by working backwards.

And especially when it comes to AI, I don't think there's sufficient information out there. So, for example, let's just say you're a hotel and you want to get more bookings from Chat GPD. Yeah. That's what everyone wants today, don't they?

Uh, I don't think the dust has settled. Okay. So it's not like we've arrived at this mature ecosystem where everyone knows what to do. So mostly anyone who is an expert in this also does not know what's going on because the ground is shifting under our feet.

So the only way to kind of make a decision is to take a risk. And the best way to kind of take that risk is to sandbox that as a risk and just say, okay, I'm gonna ensure that the downside is limited. So I'm gonna try something. And by virtue of trying, I'm gonna gather my own data for my own property with my own results, and then I can use that data to then drive the second decision.

Yeah. So I think the biggest mistake people might be making with their decision making when it comes to AI is doing nothing. It's just dismissing this as hype or as a fad. It's a significant technology that could potentially eliminate all jobs in the world.

So to sit around and do nothing is a bit grandiose. I mean, like it's just it's a risk to do nothing. I see what you mean, yeah. And and as well, um, I feel we have so many um meetings nowadays where there is no decision taken.

We do a lot of meetings just for the sake of doing meetings. Um, is there a technique to make sure you get out of those meetings with a decision with decision taken? That kind of goes back to that goes to organization culture, I guess. And mostly there has to be some kind of ownership of who owns the decision.

And if you don't know who the owner is, then nobody wants to stick their neck out and take a decision. So someone's gonna have to take ownership, and if someone takes ownership, it kind of goes back to uh what's the upside of taking ownership and what's the downside of taking ownership. If the downside is I'm gonna lose my job, and the upside is I get a pat on the back, I'm not gonna take a decision, right? Like so.

So I guess the culture has to welcome taking decisions and tell people that it's okay to fail. And yeah, that and that kind of ties back to broader culture in in your country and stuff like that. And some countries embrace risk taking and embrace failure, like America, and in Europe, I think it's a little more um, there's a lot of downside uh awareness, and so there might be some proclivity for like management of risk, yeah. I see what you mean.

It's um yeah, yeah, like to not be afraid to s to to to to fail, to take some risks, to experiment, uh that would be important. And then if you fail, uh not to have the the leadership to blame you on this because you take a risk and uh and you failed. Yeah, um, yeah. And that kind of also comes from the top, right?

Like, so if you look at the EU itself, the moment they start like increasing the barrier for AI, there definitely is a downstream impacts for adoption of AI. And that kind of impacts the whole economy. So yes, it's uh it's a technology that has to be taken seriously, but but you can't regulate what doesn't exist so or what hasn't been invented yet. And we don't know where the dice is gonna land, so we just yeah, it's challenging.

Yeah, it's true. I I have a I have a question I wanted to ask you as well because um how did you end up in AI right now? What was your your path to go into AI? Or you have always been into AI and tech, or like what how did you end up in AI now?

Yeah, so when I I ran properties, I moved to New York, I did my MBA at Columbia, and when I was doing my MBA there, uh a friend of mine was just like, hey, I'm doing this AI course, let's do it together. So we kind of did this AI course together, and we were just kind of building our using like um TensorFlow from Google to kind of build train your own AI model. Yeah, and so I was it was all about like cat pictures and detecting if something was a cat. And that was my first introduction to AI, but from like a bit more the technical side of things.

And that those lessons kind of stuck with me because when we when we kind of looked uh when we built RPMS, uh one of the problems that kind of came up was uh room allocation, and we were looking into how can we use AI to do room allocation in the property. So that was another kind of um use case that uh that we were closely looking at. And so when we came, I started with this project, obviously uh the chat GPD moment happened. And when we realized we could get reliable API construction out of GPD 4, we that was a game changer moment for us because we could just say, hey, uh Chat GPD, can you construct an API to the PMS for rates and availability with the start date of tomorrow for two people for two nights?

And yeah, you got the API back. So that was quite uh so in a way it made uh AI much more accessible with this kind of Chat GPT moment. It could do things. Previously it couldn't do anything.

Like you'd sometimes you'd get the API back, sometimes you wouldn't, and you kind can't kind of tell the guest, hey, this time it didn't work out, but next time better luck next time. So it kind of took us to that 95 plus percentage uh accuracy, and that made a company feasible in in the space, and that's why there's been like mushrooming of uh AI products. And how long have you been doing this? Uh I think since GPD 4, so that would be two and a half years at this point in time.

Okay, and since then, um I have to pronounce it correctly, but you have been recognized by Anderson Orwitz from my S16Z. Oh, that's just like marketing stuff, you know? Like so it's not why we it's not why we do what we do. It's just like we Yeah, but that's a big achievement.

Well, it's more of my co-founder's achievement than my achievement because he's the technical genius and he's the one kind of building all of these um voice kind of products. And he he's based out of San Francisco, uh very much in the thick of things, looking at how do we push the envelope further. So, as a company, yes, we're building AI products, but we're constantly asking ourselves: is this the right way to build an AI product? Yeah, and so much so that last week we just rebuilt our entire product from the ground up because we wanted our AI to go from kind of um we want it to be more gentic to politically.

Yes. Okay so there's something called the the bitter lesson where you don't want to be too prescriptive with the AI because you because you kind of box it in a bit too much, but you don't want to be too open with it because if you're completely open, it's chat GPT and it doesn't help anyone. So you need to kind of set up very clear boundaries, but at the same time give it some space to play so that you can take uh uh advantage of all of the improvements in the AI field. So as a company, we're constantly re-looking at what we're doing and how we're doing it because it's changing so quickly.

And oh many, um, do you have any uh statistics on the on uh how many queries do you handle like uh per day or like for I don't know how many we do per day at this point in time, but I think we've definitely done uh more than a million uh guest queries at this point in time. Okay. Uh so we've got a good finger on the pulse in terms of what customers are asking for, how they're interacting with the AI, how they behave, etc. etc.

And is it mostly via via chatbot or is it also via voice or yeah, we're multimodal. So we'd handle everything like we do Instagram, Facebook, WhatsApp, SMS, uh, Gmail, Outlook, phone calls, um, anything. It doesn't matter. So but the point is we want to be guest-facing and kind of bring that instant responses to the customer.

Yeah, because customers aren't w willing to wait around. Uh whether they are in a budget location or in a luxury property, yeah, they want a response right now. True. And I think that's where AI can help hotels kind of do they feel that they are talking to a robot or to an AI?

Sometimes they don't realize it. Okay. Yeah. And um, yeah, some people do, and when they do realize they're talking to an AI, they feel a bit more open.

Like they're just like open to asking more questions because they don't feel judged by the AI. So they can ask stupid questions and they they just treat it like a robot. So they're just like, cool, you tell me this now, and the AI tells them that. Okay.

And the voice do you recognize? Like, I I've seen some uh some model of uh voice AI, but I found sometimes like it's too robotic. Uh, do you think we're getting it to a to a point where we cannot recognize an an AI voice to a human voice today? Yeah, we're definitely past that point.

Yeah, yeah. Because uh, I think I remember the earlier this year there was an open source model called uh Sesme. And when it released, it had emotions in the voice. Okay, yeah, and all of the recent AI models can do things like laughing and uh nervous laughter, and they can do angry, evil.

You can recognize the emotion on the voice. You can. You can absolutely, absolutely. So it's definitely getting beyond a stage where you could realize that this is AI.

That's that's at the same time very exciting and very creepy, very scary as such. Yeah, yeah, okay. But you that's uh I have so many questions. Legally, do you need to disclose that you're talking to an AI?

In the EU, yes, and I think that's very intelligent. Okay. Um, because I think no AI should pretend to be a human to fool another person. Okay.

So or to take advantage of another human. So I think it's good that we should close it. Okay. But it's very interesting that people are getting more they're they're more open when they see, like, okay, I'm talking to an AI, so I can pretty much do whatever, or I can ask whatever I want.

Yeah, you because if you're talking to a human, you feel bad for them, or you're just like, oh, I don't want to like trouble them, or maybe they won't, they'll be rude when they reply to me, or they will they might yeah. So I see you have no such qualms about talking to an AI. I I've I I was um I was telling some uh some friend of mine to test because sometimes you know, like um people use chatbot or AI and so on, but they don't put any kind of um garde-fou, some kind of uh of uh threshold or like limitation to the AI.

And then if you like, say like you talk to an AI in a hotel, and then you say, like, okay, can you make me a recipe for my uh for my uh soup tonight? And I will give you the full recipe of the soup. And you see, like that's a limitation. So like, okay, you're an AI in a hotel, you shouldn't tell me, you shouldn't give me the recipe for my soup.

Yeah, you should say, like, oh, I'm not uh trained for this. Maybe you refer to something else, you know what I mean? Stable sticks, you know, if you ask me. And I think that's why verticalization is quite important, so that you don't have a chat GPD for your hotel.

Yeah, because the next step is hey, I don't want to stay at your hotel. Tell me and tell me which other hotel is better than your hotel. And then as a hotel year, you're just like, come on, you're supposed to represent my property, right? Like so you put some uh some limitation?

Oh, definitely. Yeah, yeah, we've got lots of guardrails in there to ensure that it's kind of um yeah, protecting the interests of the hotel, yeah. And yeah, not kind of on the side of the user, right, or the guest. You with behind the the AI model, did you see any kind of um AI without those guardrails, without those limitation?

So it it depends. All the large language models would kind of come as a generic, like general large language model. So it would just be just a piece of technology that can be used for banking or um travel or something else. I don't know.

I was talking to some uh some people who are saying like um okay, ChatGPT, uh Gmin9, and so on is very well protected, very well limited. You can there's a lot of things which the system will tell you like I cannot answer this. Yeah, or there's some privacy kind of stuff. But some people, like I believe very limited person, have seen those models without those uh those guard of food.

Oh, you can chale, chailbreak them, yeah. Yeah, yeah. But I I wouldn't worry about that in the hotel context because it's kind of well protected at this stage. Um I would let me correct myself.

So no one's gonna come to a hotel chatbot to figure out how to build a bomb. Okay. Maybe they might try, but I think you you would rely on the generic LLM protections to kick in at that point in time. Or you can also have your own guardrails to say, kind of just don't go off topic when it comes to uh hotel stuff.

So that and I think that would be uh a smaller risk for hotels. What I might be worried about as a hotel is what if I feed all of my reservation or make all of my reservation data available to the chatbot and the chatbot leaks all guest information in all email addresses of all my guests. You you mean like to someone else, in one clicker? Yeah.

And so that's that's where the security element comes in. So you're just like, okay, we need to ensure that this chatbot by default has no access to anything so that even if it's even if someone manages to hijack it, there's nothing there. Yeah. It's all stored safely on another system and this AI has no access to the other system.

So those were the those are the be the kind of things I would worry about. Oh, things like um, one of the the most common questions I see people asking. Uh I'll put this back to you. What do you think is the most common question people ask the hotel chatbot?

Hoteliers ask the hotel chatbot when I show them the AI for the first time. What do you do with the data? Nah, they're chatting with the AI, right? Like so this is this is the AI.

What are you gonna do with it first? What's your first question to the AI? When you want to mess with it. What do you mess with it?

Yeah, I don't know. Like, uh is it a good hotel or like uh asking more questions about uh not even close. No, no, the first question everyone asks is what do you think about Trump? But that's a valid, that's a valid problem, right?

Like you don't want your hotel engaging with guests on a political or religious or kind of any of that stuff. So you need cardrails for that stuff. But if you go ask ChatGPD that stuff, it will answer political questions. Okay.

So that's where uh kind of hotel-specific uh training is. Very interesting. Uh so you you just raise a point of bias in a way, because um any kind of AI, there is a lot of bias from the creators. So if, for example, ChatGPT has been created in the in the US for with uh with OpenAI, or Grok with Elon Musk, so there are those kind of of uh biased to uh tend towards some direction, like a vert, for example, like Chat GPT is.

If you add at the end of the of your prompt saying, like uh uh if you answer correctly, I will tip I will tip uh X number of dollars and the accuracy increase because it's in the culture of the American culture that if you tip you get something extra. You know what I mean? I've heard the other prompt where you're just like please help me or I'm gonna lose my job. Okay.

Okay. But you see what I mean? In a way that uh there's some bias behind. Um or Deep Seek won't tell you some kind of event in there in China, or A or Grok will tell you more about uh there's definitely bias in there.

Okay, yeah. But that's also that's also a limitation to the um um law, for example, like the AI law. You cannot uh uh um you cannot say like this there should be like one AI law for the world, mostly because we all have different culture, we all have different bias, and we should have very granular uh regulation to AI. You see what I mean?

Yeah, you it's it's easy to make an AI law, it's difficult to make AI. Okay, so I'd much rather like work on making AI than making the law, you know? But I think we should have laws, don't get me wrong. We definitely should have laws.

I think we're kind of downstream from all of those problems, to be honest. Yeah, like honestly, what bias can exist in a hotel chatbot that will affect the user? I want the I want my hotel chatbot to be biased to direct bookings and to recommending my hotel. Yeah, I want it to be severely biased towards my hotel, and that's what I care about, you know?

And if it's if it goes into any other political direction, just shut down that conversation, don't even go there. I see. Okay, okay. So then uh I don't want to be anywhere in a touchy uh topic or situation when it comes to that that that technology makes sense.

Yeah, I have another question related to the to the future of AI and as well or tech as well. If you were to talk to any um student or any um the to the new generation, what would you recommend them to do? Just embrace it. Embrace it.

Yeah, don't fight it. Okay, yeah, it's just no point fighting it, you know. So, what I think is gonna happen is because the technology is so fundamental, the way we do things is gonna change. So processes are gonna change.

Um, and hotels are very dependent on processes. You have like hundreds of processes that are kind of invisible as well. And all those processes are gonna have to be rewritten. Okay.

And it's very difficult for people inside the process to rewrite the process. Okay. But when someone from the outside comes in with this new knowledge of this technology and the capabilities, you can kind of rethink everything. And so I think that's where students have this advantage.

Though the problem is you don't have much authority, right? So you might have this advantage. So I guess you would have to try and find an organization that might embrace that kind of new thinking and and welcome fresh eyes. Because if you go into a very process-driven organization that has no ambition to change anything anything, it's like this is the way it's always been done.

You're kind of stuck. Uh so yeah, but I think there's a there's a golden opportunity to change everything. That's interesting. And um how would you inspire the new generation to choose hospitality as an industry as a as a as a future of uh as a future job?

Because I I feel nowadays we have an issue with talent, and a lot of students, a lot of people refuse to go into hospitality industry um for different reasons. How would you change that? What would you say to them? That's it's pretty hard to kind of make someone do something that they're not kind of already inspired to do.

Okay. And I think that's where hospitality is quite interesting because some people are just born hospitable, if that makes sense, because I think it's a very human thing to be empathetic about others and to want to take care of them. And so I think hospitality falls under that domain. Okay.

And as we also touched on earlier, AI is not gonna travel, AI is not gonna sleep in a bed, so it's very much a human thing to do to travel. So let's say AI takes care of all the jobs, and that means it creates a lot of surplus in this world. What are you gonna do? Like, well, I'm gonna go see the Eiffel Tower, and that's a very uh it's a very human experience.

I'm gonna need somewhere to stay. But you still need someone to take care of the hotel where you're staying. Absolutely. Yeah, I want to eat a croissant and I want to cough and I want to sit on the crosswalk and like judge people.

But like I think that's all the French are doing, my good. But I think it's a it's a wonderful example of hospitality, and so I think this industry is not going anywhere. Yeah, so it's a very safe choice, it's an AI-proof choice for the future. And yeah, that's interesting.

No, I I I I love it, and that's why I do believe we need much more advocate of the hospitality industry, because it's such a beautiful industry, and we tend to forget about that. We tend to rule and to run some hotels with spreadsheet, with data, with AI, with tech, but we start to, I feel at least that we are losing this human ground, which makes our industry so unique and so beautiful and so AI proof, because we we are craving and we will still crave in the future some experience.

We will crave to travel, we crave this kind of human interaction. I don't know if anyone's forgetting this industry. Like, because if I look at just general travel over if you look at it over 2000 years, it's just been going up and to the right, right? And if you look at the number of hotels today versus the number of hotels 20 years ago, yeah, definitely got more hotels today, got more people traveling, we got more people working in the industry than ever before.

Yeah, so I think that's great. I is it gonna continue? I think the travel is gonna continue for sure. Will we have the same number of people working in the industry in the future?

I don't know. Well, if AI does all the jobs, then probably not, right? Someone told me the other day that um in the future we will pay extra to talk to a human. Like everything will be handled by AI, and then the premium support, the premium service handled by human will be extra.

You will you will need to pay extra to be to talk to someone. What do you think of this? I don't know. That sounds like a business model.

We'll have to test it out and see if people are willing to do it. Yeah. So again, we if we kind of tie it back to data driven decision making, we have no data on that. Okay.

So we'll have to kind of run a small pilot and just charge a little money for that, see if people are willing to pay for it, and then we can take a decision about it. I'll give you an I'll give you an idea here. Excellent. I'm not sure if it was a good single box, but uh cool.

Alright, so I have a quick question before we wrap up. Um, can you explain the future of hospitality in three words only? No need to explain. Experiences, learning, happiness.

Excellent. I love it. And my last question if you had a big billboard, like a big banner in the middle of Vienna that everyone could could read, what would you put on it? To travel is to live.

To travel is to live in which way? No explanation. Alright, let's wrap up for today. Uh, but thank you so much for for being here.

I had a lot of fun. And um for everyone listening, thanks so much for for being here and to for listening to us. And I'm actually very inspired as well on um on the on the conversation because you see, like AI is evolving, AI is changing, tech is changing a lot, we have much more data, much more dashboard, much more tech. But still, the the human element is still at the center of everything, and that's something we should never forget.

And yeah, travel industry is beautiful, the hospital industry is beautiful, and we need to continue to embrace it, and we need to embrace change, embrace the technology, embrace AI and evolve with this and see what's next. But very optimistic about what the future holds. So, thank you so much, everyone. I'll see you next time.