Why the Most Confident AI Hotel Pricing Recommendations Are the Least Trustworthy

Glenn: [00:00:00] Hey, everybody. It’s your hospitality, friend Glenn. I’ve got Kartik Yellepeddi with me. He is with Duetto. Very excited to talk to you, man. This is gonna be exciting. Our first conversation over here. I want to talk to you a little about what the theme of everything seems to be. All we do is talk about AI, ai, ai. But yes, we’re gonna talk a little bit of a of a turn because everybody’s been talking about how to implement it. Nobody’s been talking about how to smartly use it and get things actually done with it. So you’re relatively new to hospitality. You’re relatively new to Duetto. How are you thinking about this particular issue and what is your outside experience? Help inform you to help people in our incredible industry?

Kartik: [00:00:41] Yeah, no, I think that’s a good question. The core thing to realize is that in order to be able to make sure we are optimizing for profit of an hotel, there are many, many decisions that have to be made by a hotel sales, marketing, distribution, pricing is just one such decision that’s being made. And in order to be able to bring more AI into this, we have to start from the ground up. What I mean by that, there’s actually I look at it as three levels of AI. The first level is the scientific foundation, the core pricing and forecasting algorithms. Right. Whether you are pricing groups, whether you’re pricing transient bookings, whether pricing ancillaries should all use more or should use machine learning. And these new set of AI tools that are available, that is the foundation. The next level is how you how AI could be used to see how a revenue manager use is interacting with that intelligence at level one, layer one. And that could be done through AI and lens to to be able to make sure that the right are there is there is enough understanding using these AI, using conventional techniques to be able to guide the intelligence that is at layer one is really layer two. And layer three is essentially more organizational, not a specific user, but what we call as performance engineering, which is essentially how do you make sure that you build an AI that then guides the humans Who are in different departments, whether it’s a sales person or a revenue manager person. So, and, and those are the three layers of intelligence that we have to build. And I’m very excited about it because that is that is exactly what I’m here to do. And I joined this company. And we are way along the path of doing the most

Glenn: [00:02:30] Basic because one of the problems that I’m having with AI is that communicating with it isn’t as obvious as the way we’ve interacted with computers in the past. Right? How are you trying to simplify it and making it easier for users who aren’t as smart in computers as you are? And or maybe as dumb as I am with computers? Yeah.

Kartik: [00:02:54] I mean, that’s the responsibility of the AI too. So I would say that the algorithms that are being built now should not be just making recommendations, but they should be capable of explaining why they are making these recommendations. Why and where are they? How confident are they about the answer? How much are they hallucinating? So these things become a property of the intelligence. Intelligence is not just about giving an answer and being very confident about it. No, no, because.

Glenn: [00:03:21] They’re more confident they are the least, the less I believe it. Yes, exactly.

Kartik: [00:03:24] At that point. So they have to earn their trust. Yeah. And and they also have to guide you in the right place because again, as I said, millions of decisions being made by a hotelier every day across all these different departments. How do you make sure the human attention is directed to the right place where the AI is least confident, is where the human should jump in and try to help the AI. Yeah. So it is also the responsibility of the AI to, one, find the needle in the haystack and then help you explain why it is struggling so that you can help the AI.

Glenn: [00:03:54] So what I think I’m getting out of this is it’s not all me. It’s not all the AI. We’ve kind of got to kind of meet in the middle and figure out how to communicate and understand each other a little bit better.

Kartik: [00:04:04] Absolutely. And, and that judgment is what revenue managers are really good at. They just need tools to be able to do that.

Glenn: [00:04:11] Right. So how does all of this then translate into products and helping hoteliers out there find more success?

Kartik: [00:04:20] Yeah, it translates into, as I said I strongly believe revenue management as a discipline is heavily underinvested across the entire industry, travel industry, and I come from the airline industry, and I can speak for that. After 20 years spending there, we are not spending enough money on revenue management. Right. And I think that is foundational to be able to spend more time and spend more, invest more into the into the fundamental science, which then translates into algorithms. Then you also have to change, move away from workflows and dashboards. So more conversations and more alerts so that it becomes more natural the way the, the human interacts with.

Glenn: [00:05:00] So it’s got to be more like the computer in Star Trek, right? No, I mean, I’m like only half teasing because the way they interacted with it was conversational. Exactly. And it worked. And I kind of feel like we’re sort of getting there now, but there’s still a lot of that. Like you were saying earlier, the hallucinating issues, the unearned confidence issues and that sort of thing.

Kartik: [00:05:21] And the way you build these products also have to change the product management and engineering as a discipline is changing as a result of this. You need because it’s not deterministic anymore. It is all probabilistic. So you need to make sure that you have the right evals. You’re building the trust, you’re building the confidence in the product, and so on and so forth. There’s a there’s a lot of things that have to change in the product.

Glenn: [00:05:42] And how are you thinking about this in terms of your, your commitment to hot stats? For example?

Kartik: [00:05:47] Yes, in hot stats, I think the biggest opportunity is obviously one big opportunity is making sure AI is. There is a more agentic, I would say a way of interacting and doing the art and science of revenue management. The other aspect is what kind of expanding the decision surface. We have spent way too much time optimizing one lever, which is price, right? We have to optimize restrictions. We have to optimize group prices. We have to optimize ancillary prices. We have to optimize the profit labor for the labor spend and all that. So there are just so many actions, so many decisions that have to be made. There is just a lot of focus on one. So one of the things that I am very passionate about to do at Duetto is how do I make this company not just data rich, but decision rich so the decisions are being made across the entire gamut of ecosystem and commercial department instead of just one level over over leveraging one particular level.

Glenn: [00:06:47] Yeah, I love it. Anything else you want to add? That was a whole lot. We’re gonna have to rewatch this one five times, guys. Yeah.

Kartik: [00:06:53] No, I think as I said. So watch out for the space again. The, the key point here is that we are in this together. And there is a lot of opportunity for the entire industry to benefit from. But the key point is to be open to new technologies that I think is going to be the biggest challenge, that we are all going to face new ways of doing things. And yeah, but, but I believe that hospitality in this industry is ripe for that kind of innovation.

Glenn: [00:07:20] Yeah, it definitely, it definitely is. We, we need, we need your help over here. And I can tell he can help from the what I’ve been paying for airline prices these days. So he knows he knows what he’s doing over here. Thank you so much for being here. Please do me a favor. Check out our friends over@joanna.ai. We’ll see you soon. Have an amazing day. Love you guys. Bye.