EP 351 – Can AI Finally Help Us Have a Better Travel Experience? – Adrienne Enggist – Booking.com

Read the best-effort transcript below (This technology is still not as good as they say it is…):

Michael Waitze 0:00
Hi. This is Michael Waitze, and welcome back to the Asia Tech Podcast. We are joined today by Adrienne Enggist, a Senior Director of Product, Customer Experience and Platforms at booking.com it is so great to have you here. Adrienne, thank you so much for doing the show. Let’s give our listeners a little bit of your background for some context.

Adrienne Enggist 0:36
Sure, I’ve been with booking for 14 years, so I was the second product manager ever hired there, and prior to that, I’ve been in the travel industry since 1998 so I got my start. I know I got my start working for Iceland Air, which was only 12 airplanes at the time, at a working installing the Amadeus online booking engine for them. So 98 they were like one of the first airlines to pilot distribution over the internet, and I’ve been working on that pretty much ever since, across a wide range of things.

Michael Waitze 1:09
Got it in the travel industry for a long time. I love it. I love it. Let’s talk about something that we almost have to talk about for every industry, artificial intelligence. Is this going to make the travel experience better? And if it is, how?

Adrienne Enggist 1:25
So I Yes, I think, I mean, 1000 times, yes, planning travel is so complicated. Yes, it is. It can be so stressful at the same time. Planning travel is absolutely a joyful thing to do, and we have been doing, I mean, for as long as we’ve been doing research at booking, there’s been a very clear divide between the things that people relish doing when they’re planning travel and the things that are just extremely painful and generally, the stress levels mount when you start to see that you need to coordinate uncharted territory, so making things like flights match up, making sure that you pick up, you’ve arranged your rental car pickup on the day that you actually plan to pick it up, something I’m notoriously bad at doing. By the way, showing up a day late or a day early is my gift. And I think that the the cool thing about AI is that this is all like unstructured data that is not in any way taxonomized In such a way that we could really easily make those connections simple enough for the customer. So we’ve had this concept forever of, like, this idea of a self healing trip. So we could, we have all these predictors. We have so much data at booking. We know what goes wrong. We know what can go wrong. We know what makes things go wrong. And we should be able to make those predictions about like, we see something in the distance, and we contact you and tell you, Look, this is a this is a potential bump in the road coming. Would you like us to fix it? Or would you like to fix it? The that is so hard to do at scale, it’s so impossible to do at scale, so you need to, I need to almost be an expert. And so like that. When we got rid of travel agents, we moved everything online and let you have this power at your fingertips. We lost that capability for that little safety net. So now I think it’s coming back. I think agentic relationships and the ability to put structure around unstructured data is really the key to all of it. Okay, talk

Michael Waitze 3:15
to me about this, because you’ve hit on something that I’ve been thinking about for the longest period of time. You know, I go into a restaurant and I want to get served by a waiter. That waiter is like a professional person. I remember once, and I don’t know where you’re from in the United States, I don’t remember. But before I left Sagan, Virginia, okay, awesome. So at least from the East Coast, before I left for Tokyo, when I was 24 years old, I said, I’m treating myself to a really nice dinner. And I took my girlfriend to Montrose, which is a very fancy restaurant in New York, and the guys that were the waiters there, and it was mostly guys, because it was 1990 that was their life like that. They were not doing it. They weren’t trying to become actors. You know what I mean. They were professional waiters. They knew their wine inside and out. They knew the ingredients in the food. They knew everything. And you could almost walk in and just go, just tell me what I should have tonight and what I should drink with it. And you were kind of done. But I feel like, I feel like, even in the restaurant industry now, they’re trying to automate, like, the waiter away. And I feel like I’ve become my own travel agent. And you’re right, because you said this earlier. It’s so complicated, and it stresses me out, because I’m like, If I mess this up, it’s almost impossible to get a refund and do all that other stuff. Like, how do you then use the AI to make all of this better, so that I don’t have that stress? And you’re right, it’s all unstructured data, but just tell me, walk me through, like, how, where in the process, where I go. I want to stay at this hotel in Singapore. I’m on this flight. Do you know what I mean? Remind me that I have to do my visa thing to get to Singapore, all these little things that stress you out. How do we remove them, using that unstructured data and applying AI to it?

Adrienne Enggist 4:51
Well, okay, so for example, I’m trying not to go too deep into the plumbing of travel. It’s good. Back to your restaurant analogy, you don’t need to know how the food gets made. You just need to be able to hand over a bit of information, or be able to have some sort of context so that that waiter can make your meal relevant to you. So I think that that indeed is unstructured data, like the things that motivate you, the things that make you happy, isn’t necessarily something that you could fill out a form and then have something come back, and then, you know, the idea of the way that travel has coped with the removal of a travel agent is by giving us access to the tools that equivalently, equivalently take the world’s largest spreadsheet of information, so availability, all of the prices, all of the policies I’m talking let’s talk about accommodations alone, availability, prices, policies, room types, qualities of rooms, the qualities of the property, the qualities of the surroundings of the property, all of that is we try to put structure around that, and then we give you basically the ability to create a pivot table that has been for the last 20 years, what we’ve given to customers. And people go, great, cool. I can use the calendar picker, and we will do 1000s of experiments on just making that calendar picker easier, easier. We will, you know, I’ll drop down for my kids ages, a drop down for this and that after a while, you know, you don’t want to give over too much more structure anymore. And then we spit out a pivot table, a list of things that we feel are relevant for you. And then you can sort and filter. Oh, my God, it’s so boring, but that’s what you can do with structured data, because

Michael Waitze 6:29
you’re talking to somebody who lived his entire finance life in Excel using pivot tables to do this. In my head, I’m going through the machinations. And I think a lot of people that listen to this will understand that it’s not boring at all. Actually travels. Super exciting. Sorry, go ahead, right.

Adrienne Enggist 6:43
But imagine now though, and the things that we were always trying to do, like, especially, it became really obvious during the pandemic, when we’re talking about like, adding health and safety qualities to properties. To get those health and safety qualities, we had to often the easiest way to do it was allow unstructured data, so free text each property could come through and describe the enormous amount of work they were doing, because putting structure around it was going to be too hard, and collecting it was too hard for 2 million properties and so on, being able to parse what’s the content of that unstructured data and things like customer reviews. Another we try to put structure around it. You know, if you do a booking with booking.com we ask you to get the positives and negatives. People stick negatives in the positives and vice versa, you get little nuggets of like travel information, super tips, like, Hey, did you know this thing? This one hotel in Melbourne is across the street from the only free parking lot in Melbourne. Like, these are real, real situations that we can’t highlight in structured data very well. So then you come in with a model that you can then that is trained off of enormous amounts of information, generally. And then you can fine tune or augment with unstructured data that is very specific to our platform. And then things start to get super magical, because then you can create these non linear journeys through this information that is not pivot table, right? And then you’re sort of curating. You’re not making lists and sorting and filtering, you’re curating collections of super relevant things. And it’s like, Oh, I’ve over curated it down to two properties. But maybe, okay, I’m going to remove this one little requirement that then expands your your bubble, then of options. And I think that’s when, you know, then you’re really, that’s the fun parts of travel planning,

Michael Waitze 8:21
yeah. So I’m gonna, I’m gonna expose my opinions on this a little bit like, I always want to stay at a five star hotel. I always want to stay at the best hotel that I can afford, in the place that I’m going, because I feel like that’s going to be, that’s, it’s not necessarily the place where I spend the most time, yeah, but it’s going to be the place where I get clean, where I get changed, where I use the pool, all this kind of stuff that it’s really important to me that everything be high quality. And this gets back to the thing you were just talking about the reviews, right? Whether it’s structured or unstructured, is there a way? And let me finish the whole question first, before you see where I’m going, right? But is there a way to take the AI to make sure that, like the gamification that happens sometimes in the review space is removed, and the equivalency, to me is a lot of insurance companies in the claim space, right, which is the flip side of this, try to use artificial intelligence to understand, is this fraud? And I’m not saying that the reviews are fraud, but is it like, not true, right? Because one of the problems travelers have is they’ll book a hotel, is just because we’re talking about hotels, right? And they’ll get there and it won’t be what they expected, but now they’re kind of locked in, because you’re your own travel agent. You’ve paid a company that’s not paying the whole Do you know what I mean? Like, how do you use the AI to fix that? Because that’s a big issue, I think, for travelers.

Adrienne Enggist 9:35
Tell me about the gamification, though. What do you mean by that? Well, say more,

Michael Waitze 9:39
yeah. So like, the photos can be adjusted plenty of times. You’ll get to hotel and you’ll be like, yeah, the room doesn’t look like what the photo looked like. Oh, yeah. And artificial intelligence, through machine learning and vision training, should be able to disintermediate some of that, it would seem to me, and that will solve part of the problem of when I get there. Get what I expect. Again, going back to the restaurant analogy, if you tell me strip loin, right, and I get flank steak, I’m gonna be pretty mad,

Adrienne Enggist 10:09
but flank steak is delicious anyway, not a good example, because I love, I would I love flank steak

Michael Waitze 10:16
my dad was in the business, so we had flank steak so many nights of the week. But sorry, go ahead. Yeah.

Adrienne Enggist 10:21
So, okay, yeah, I worked on user generated content and set up the reviews program at booking. So I’m pretty I’ve done lots of research with users, and it’s really everything we do at booking comes down to matchmaking, right? So one of the things that like we say, okay, we love five star hotels, and then you would have this feeling that everyone, every like a hotel, would say, Well, I have no, no chance of competing, because I’m only a two star. And one of the things that we learn from user generated content is sometimes a two star is exactly the thing that the customer needs. And so you can tell a good review, you can have a perfect 10 point review in a two star property. If that person who came in and saw that property’s description and the photos of it and it met what they were expecting, what they’re hoping for, for that particular price point, you can get amazing reviews. You don’t need to be five star to get a 10 point review, right? And that was that’s like, the trick there is that the gamification, I don’t know, the misrepresenting of one’s property. Let’s say so I generate my photos, or I’m using the the Photoshop them, yeah, yeah. Or the spec photos before the hotel was even, like, I’ve seen those before, where the background of the picture out the window is the same thing in every photo. Like, come on, this is not real like, show us the real thing, because if you show the real thing, you’re going to get people who are happy with the thing that with the thing that they get. So it’s really a matchmaking play. Same thing about like, you know, location is an interesting thing, because the location score really depends on what you wanted from that property, if the location is near or far to the thing I wanted to do. So I think that that, to me, is AI can help with also looking at one of the things that we know that, or I we’ve observed in many, many research sessions, users who are looking at the reviews for a property that they have added to their curated list, they’re looking at all of the reviews and looking for deal breakers and looking for commonalities to prove to themselves that these people care about the Same things I care about. So, like, I’m right there with you. I always want to stay in five star as well, because there’s a certain thing about needing that. I need that sort of, I’m the sort of person who needs that safety net in order to be able to go out and do things that are very not five star activities. So, you know, I will eat in a hawker center every night, but I’m coming back to my five star in Singapore. You know,

Michael Waitze 12:42
you’ve justified my entire travel experience. I’ll take every other risk. I’ll take every other risk that I’m traveling. I’ll jump out of a plane without a parachute as long as I land in my five star hotel. Sorry, yeah, yeah,

Adrienne Enggist 12:54
exactly. It’s just this nice place of like, okay, but I know that I can go back there and everything will be fine. And, you know, I can have a nice shower and get a good rest. So I think that that’s part of the same thing that we’ll be able to use AI for and machines, especially large language models, is to be able to do that same sort of, you know, tell me, if look through the customer reviews on every property in Singapore, and tell me the ones that seem to sound like the same kind of thing that I care about and that can be done at scale, which is something that we were doing manually ourselves by every night, going through every property and reading everything.

Michael Waitze 13:25
So that’s a really good point, actually. And look, when I first started at Morgan Stanley in 1987 I had a 10 key on my desk. You probably don’t even know what a 10 key is. You’re too young to know what that is. It’s one of those little calculators that has the paper coming out the back. Oh, yeah, of course, yes, yes. That’s called a 10 key. And mostly accountancy, is that I was in the back office when I first started. I ended up on a trading desk. But as the technology changed, and as we had real time data and more access to data, and then more ability to do machine learning and then do some historical analysis, it changed the way we did everything. You must have seen the same thing, right? If you were there when Amadeus was being implemented at Iceland airlines, you were really there at the beginning. And to me, that’s so cool, right? Because you saw the impact that Amadeus had on just how flexible you could be with booking, load management, all these other things you could never do beforehand. And I’m curious if your view on product development, product testing and stuff like that has changed in the time that you’ve come and now that you have this really powerful tool, the AI, the large language models, the machine learning has it changed the way you look at product development, and particularly this customer centric product development, if you include even all the data stuff that you have now,

Adrienne Enggist 14:37
oh, man, so much. So I’m so interested in

Michael Waitze 14:40
this, like, I don’t even have that written down anywhere. It’s just It’s just what I’m thinking about, what you’re talking about. I’m just like, you’re there. You must know. I really want to know. Yeah. So

Adrienne Enggist 14:51
if you would ask me two years ago what I would have told myself 10 years ago when I started on this, I would have said, think, number one. Best thing you can do for yourself is data engineering, making sure that you can don’t your machine learning data scientists are not spending 50% of their time trying to sort out the data, because if you have a very agile product development team who’s creating new data sources for themselves to then create models off of and then read and write to those models, you’re creating tiny pockets of relevance inside the bigger front end experience for the customer. But what happens then is you’re essentially shipping the org chart to your customer. If I am getting a predictor in one part of of the customer journey, or I’m getting a recommendation in one part of the customer journey, and that recommendation is based off of a few customer data points, behavior data points that I have somehow injected through interacting and then that team builds a model off of that behavior and makes a recommendation for, say, a destination, and then you move two steps down the journey, and a different team who is not using those same customer data points right, makes a New predictor, makes a different decides, you know, paints a slightly different picture of me. I start to feel that that disjointed journey, right? So I would say, you know, or taxonomization of data and data engineering like the number one thing. I wish I could go back and fix siloed

Michael Waitze 16:15
data, essentially, at some level, right? Siloed data, absolute

Adrienne Enggist 16:19
nightmare. So, but now, now the thing that I’m kind of obsessed over is the idea that we are right on the edge of a of a of a of the we need to have the ability for AI to control the user interface. So we talk a lot about generative AI, but we also need to think about generative UI now that can, in the most extreme forms, be a case where you’re literally create, writing code on the fly to be able to change the customer journey, so according to a set of standards, however, so when would

Michael Waitze 16:52
that? When would that want to happen? When would you want that to happen?

Adrienne Enggist 16:55
Well, I, I mean, I wouldn’t, right now, I think that’s, that’s hyperbole, but just like, Yeah,

Michael Waitze 16:59
but just as an example, because I’m just running through my mind of like, when would that be a good idea? You must know, I just, I’m just saying I don’t know when that would be a good idea.

Adrienne Enggist 17:08
Well, let’s back away from that idea first before we go there, right? So instead, what we need is more of like a composable UI, a dynamic front end, so machines not only need to be able to read the customer data, but we also need to have a brain behind the page, so that, if I am having I think that this whole chat bot thing is going to run its course pretty quickly. We’ve been, it’s been it’s in the trough of disillusionment. Expert systems have been and gone from the trough of disillusionment, we need to start to go into this slope of productivity, and the slope is going to be when the chat bot breaks out of the walls of chat, because, honestly, it’s the equivalent of trying to put travel over the telephone with somebody, or, you know, it’s it’s just not, it’s not realistic. We need to break from the walls of the chat and actually make use of the pixel space before us and be able to have because it’s so important that you could be able to compose a result for you based on a conversation you’re having with an AI. So then you get the best of both worlds. And then that conversation can be a multi, multi modal as well, where you could show a picture of something. You say, I want this vibe, you know, and I put something very space age in, or a picture of David Bowie, and it goes, we got you, you know. Or my favorite thing, my favorite thing, my favorite reference point, is always Blade Runner, because I’m, like, on this eternal quest to recreate Blade Runner. Anyway you get what I’m saying. So then it go, yeah, yeah. You want a Blade Runner chip. I’m going to help you create this Blade Runner chip. And then we, we are working on at booking composable UI, because they think screen composition is going to be the thing that we need to be able to have things on the shelf that that are understood by the machines that it can, like, use to create the curated search results, or, you know, just we need to have something that allows you to have consistency and familiar, familiarity that you would get from like a more standard, the same, the same way We, you know, create pivot tables now using standardized componentry, but to be able to then structure those on the fly. There’s

Michael Waitze 19:06
so much to unpack here. Tell me why chat bots are bad, and and then, and then, and then. I had only because I had a discussion a couple of days ago, and I published an episode about this from with a guy from a company called di D. And what they do is they create AI generated avatars that sit on top of large language models, so it feels like you’re talking to a human, and that kind of technology is going to get better and better and better. Do you think that the and I’m honestly asking this question, because I fence as to whether that’s something I really want to do, right? Because I hate chatbots, because I just feel like I just want to talk to a lady who’s going to help me. Please, just let me talk to her, because I can talk to her really quickly. And I’ll be like, I have to sit on the aisle. I have to, I don’t care where in the plane it is, and I need, I can’t be above the 12th floor. And last time I stayed on 12, you know, a or whatever it is. And I love that room. If it’s available, just get it. Me kind of thing, and I just don’t know if that’s ever going to happen with avatars. I’m just curious what your opinion is, right? So you have this human type interaction sitting on top of a large language model that mimics the behavior of a human, but it’s actually like a chatbot. I’m

Adrienne Enggist 20:19
trying to think so the uncanny valley of having an avatar set on top of something that let me double back a little bit. So the reason, the reason why travel agents fell out of fashion was because when you would have a conversation with a travel travel agent, often you were limited by whatever they had in their personal experience. So they would be destination experts, but not global experts, or they would have brochures. So you would have the you would essentially have to make a decision off a very limited amount of

Michael Waitze 20:49
visual it’s a good point stimulus. It’s a really good point so then

Adrienne Enggist 20:52
we swapped that personal understanding out for being able to have the to get as close as you possibly can to the visual representation we got bombarded with data, which is, you know, if you’re that sort of maximalist, amazing. But I know that sometimes I get very overwhelmed. I’m not necessarily. I can be much more essentialist in that. I I like that. If you put me in a video store and give me five things to choose from, back in the day, I would be the one choosing from that, because, frankly, I get overwhelmed right now. What you’re describing of like, I don’t want to be over the 12th floor. I don’t want to be I only want an aisle seat. A human being is going to have there’s causal inference in there. Your human being’s going to pick up on, okay, we’ve got some claustrophobia, we’ve got some height issues, a human beings going to then extrapolate that out into all their other recommendations and store that causal inference back in their memories. For you, a chat bot, at least, the ones that we’ve been dealing with over the last few years are terrible. Deep Nether networks are not great at causal inference. It’s getting better. The reasoning capability is getting better. They still get it wrong. And so I think that that’s probably the big limiter to me, is that that that sort of inference that you get from human contact, the micro data that gets picked up from those conversations that you have, those are those micro data are really critically important to helping you make a decision and helping you feel confident that you’re understood. Yeah. Avatar, no. Avatar, you can be talking to, you know,

Michael Waitze 22:22
yeah, like I told you, I’m on the fence. I I’d much rather talk to a human even, even in the sense that, like, they’re going to make these inferences, it’s so hard to like these, these neural languages are so hard to replicate what’s happening in the brain. Yet, I think at some point we’ll get there, but not, not in our lifetimes. I don’t think,

Adrienne Enggist 22:40
I mean, the thing is, though, like, it could be that this AI tipping point could be the thing that makes that human interaction scalable, because now that human can go back to starting to have interactions with you, because they don’t need to necessarily also manage and balance all of the other things in the background. So I think there’s a productivity play there that we could bring back the concept of a travel agent, even things like, you know, customer service help, often customer service help can’t also carry the context of all of the travel agent or all of the things that you needed when you made your decisions to make bookings. Imagine that all of that inference could be combined together with a human giving self a customer service help, and then they have like, summarization capabilities in the background. They pick up on these threads. You don’t have to tell the same story 1000 times, really basic stuff, which is massively painful if you’re, especially if you’re at a stressed out moment, and you’re like, I just need this to be fixed now.

Michael Waitze 23:35
Yes, there are equivalencies in what I used to do when I sat on a trading desk, right when I first joined a portfolio trading desk, which means trading 1000s of stocks at the same time, we literally had to do so much prep work to get it to be right. And there was a limited number of customers trades that I could handle, because I could only pay attention to like 17 things at one time, and an error was expensive. Could be millions of dollars. Even a simple error could be millions of dollars. So what we did was we built this thing called algorithmic trading, so we could give all this information upfront to an algorithm that said, Oh, I know what to do. And then all we had to do was pay attention to the salient pieces of every individual trade. And we went from being able to do literally, like $400 million of trading a day, to $4 billion of trading a day, and massively lower costs and still make more money. So to me, this in keeping the human, I mean, element in there was really important to that too, because the sell side, the buy side trip would be like, what’s going on on with my Toyota, because it was the only thing that mattered to him, right? And that’s what you’re suggesting the travel agent space. I would make the case to you that we took the travel agent out because we wanted to book our stuff at three o’clock in the morning and they weren’t in the office.

Adrienne Enggist 24:49
Then, I mean, there’s that would be a useful case for an avatar. Then I suppose I

Michael Waitze 24:52
don’t want the avatar. I’m just saying as a little bit of a humorous sign. I think that that’s probably why that happened. Because after war, like during war. Work like, here’s the thing, as you had two wage earners happening, the other person, whoever that was in your family, could not book The could not book the trip. Somebody had to go after work, and that after work meant that the travel agents had to be available from like seven o’clock at night until midnight, and that was never going to happen. So I think that’s part of the reason

Adrienne Enggist 25:17
why. Yeah, I think that’s and also the like, the some of you, sometimes you just don’t need it a human for a not terribly complex trip, right? And so I think that’s another, I mean, we definitely observe for years and years we were prioritizing, you know, single destination, one property, two people trips, and then you lose this entire, I guess, long tail, but extremely valid group people who are doing multi like journeys and would like to have a more complex, deeper experience. I think there’s a lot of value to be added by humans there to give that that help. But I think also the idea of, like, the sort of So, so many things we do are already algorithmic, like the way that we have to operate in our lives, like, if you’re developing software with multiple teams, this is, this is truly an algorithm where we’re constantly doing trade offs of priorities. We’re doing strategic interventions to up weight or down weight things at certain times. And I think that that’s one of the ways, as long as we can remember that it can be quite simple to keep that chaos at bay, because otherwise we end up, you know, the entropy sets in, and we just make, make huge mistakes. When?

Michael Waitze 26:25
How do you, how do you keep on top of all this stuff, right? So, like, again, when you first started, the big change was, we’re going to install Amadeus. And I would suggest to you that the biggest change that happened after that didn’t happen for another five or six years. Yeah, right. But I’m only, I’m only extrapolating from my own experience at Morgan, Stanley and Goldman Sachs. Stanley and Goldman Sachs. Like, first we had it, first we had a computer on the desk, and that was like, Oh, my God, our lives have changed. But five years later, then all those computers got networked, and we were like, Oh, I get it. Now I can send a file to London, and doesn’t take any time. And then we got real time data. But then just stuff started changing so fast, it was like almost impossible to keep up with it. How do you keep up with it now? Because now it’s changing faster than it’s ever been changing, right? Like today’s LLM is tomorrow’s like space laser. Do you know

Adrienne Enggist 27:11
what I mean? It gets it’s, it gets a little overwhelming. I would say it’s, it’s less it’s a bit frustrating sometimes, because we still have pockets. It’s not we have pockets of folks who are really struggling to adjust the mindset, even that we are moving from like, there haven’t gotten past the idea of heuristics that have probability built in in decision making, for example, or the concept of, like, when you talk about chat, there’s a very dismissive and it’s like, Wait, you’re thinking of the expert system, bad experience chat, or completely, you know, like, we need to, or have, I’ll talk to somebody who’s talking about horizon planning, and they’re going two, three years horizon three, three years from now, three years from now, our processing power will have, uh, exponentially multiplied, not even close. Things are, it’s not, it’s Yeah, so we need to think in terms of months instead of years. We need to change how we do planning. What is possible that we can run a POC on something and and in the course of, you know, eight weeks build something that would be, it would have been completely mind blowing in terms of trip planning. And then we just kind of go like, okay, the only thing that we need to put the brakes on for is to be constantly working in tandem with regulation, understanding the regulation that’s coming and to understand the spirit of regulation that’s coming so that we can make sure that we protect the interest of the customer, around their data, around their privacy and so forth. That is actually one of the things that allows us to sort of put the brakes on a little. I would love this.

Michael Waitze 28:44
So I live in a tourist location, and I hadn’t thought about this a lot until you and I started talking. But if we’re going to use artificial intelligence, right and all this data that we have, it would be super cool if, like somebody in Germany or in somebody in Boston who has no idea like what the weather’s like in Thailand, books a trip in August or September, when it’s just gonna rain every single day, right? In their mind, Thailand, it’s just like the land of islands and sun, but it’s gonna rain every single day. And I feel bad when I see tourists on the train. They’re like, they’re in shorts and sunglasses. It’s pouring outside, but it’d be great if went during the booking process, they were someone who just had, like, Thailand’s nice most of the year. But during this period of time, it could rain for seven hours a day. Do you know what I mean? Like, that’s a great use of it. No.

Adrienne Enggist 29:30
And the inference, the knowledge of that, is coming from maybe the prices are cheaper. So it’s like, I’m getting this great deal on going to Thailand during the during the rainy season, not understanding that you know the context of where you’re going is really critically important. There’s so many things like that, I mean, and honestly, like, there’s the planning months out, you have different needs and requirements for your planning on that discovery journey. Some of the stuff we’re building at booking we started in that discovery like months out, part or. Early, early stuff, because that is where the like llms are pretty good at aggregating that data offline, pulling that onto the planning experience, and then we can start to make it connect back into real viable stuff, right, which was harder for us to do for a really long time. So we just let people discover outside and then bring them in. But then like that context is when you need to know about this rainy season business, right? Because then it doesn’t feel super expensive when you’re sitting on the bus and sunglasses and it’s pouring and you’re wondering, like, is this going to stop? That’s like a seriously devastating feeling. And that’s when being around travel agent is

Michael Waitze 30:37
particularly if it’s your honeymoon or something like big event, right? Or you saved up all year for a big trip to Thailand. How much time do you spend thinking about when you’re doing this product creation, right? About personalization, like, the way I want to travel and the way, like my brother wants to travel, are completely different. I’m guessing, first of all, because he’s got three kids and I have one, but, but even so, like, how much time do you spend doing that? And again, how can all this new technology, all this new data, help make that a real possibility, not just like some kind of fantasy possibility?

Adrienne Enggist 31:07
So we personalization. I’m actually starting to become allergic to that word.

Michael Waitze 31:13
Okay, that’s good. It’s a good answer.

Adrienne Enggist 31:15
I’ll tell you why personalization means something different to everybody. So, yeah, exactly. It’s a personal. It’s personal. It’s like, the word loyalty, you know, yeah, when I first joined booking, they said, you know, go. First weekend, they said, go figure out loyalty. You might as well just say, go figure out purple, right? Yeah, no. Like, what’s meaningless. So personalization means something different to everybody, personalized, but also it can be, you know. So then you get feedback of like, well, it’s too creepy if you know too much, or if it’s if you get it wrong, it feels really so I prefer to think in terms of relevance. So you should so there’s degrees of relevance. There is, please don’t show me anything profoundly stupid when I’m trying to use your extremely huge volume. So again, the world’s largest spreadsheet of places to stay, things to do, rental cars, flights. It is very it would be very easy for us to show you a lot of irrelevant stuff. Yeah, so please don’t show me anything stupid. Is like the very, very lowest bar, harder than it seems. Um, again, if you’re your own personal travel agent and you’re trying to to to work out where are the things that are most relevant for you, we can improve upon the relevance by taking or inferring from the things that you’ve already done and how you felt about them, or inferring about things that you have put on a short list, we can infer and then make more relevant recommendations for things to do, and then we can look at the halo of things that are usually accompanying those. So if we know that you’re going to New York City, to stay in Manhattan or whatever, it’s unlikely we will suggest you rent a car. Most people do not rent cars, right? It’s just an obvious thing. But there’s a lot of very non obvious things that if you’re going to Thailand in the middle of the rainy season, we might suggest that you bring along. I don’t know I have, I can’t, I can’t freestyle on that one. But there’s probably from my going point, though, but you get the point. Yeah, yeah.

Michael Waitze 33:11
But I get the point

Adrienne Enggist 33:12
So there’s a halo of things we can infer. So I think that like the hardest thing again, for doing matchmaking when you run the world’s largest travel operator, travel OTA, is, is that when you, when you it’s like, Yay, 2 million properties. It’s like, oh, 2 million properties. Yeah, yeah. It’s too much. It’s just too much.

Michael Waitze 33:31
So let me tell you, that’s what it’s to me. It’s

Adrienne Enggist 33:32
about relevance. Let

Michael Waitze 33:34
me tell you a funny story about that. I was sitting on stage during an E commerce presentation that somebody was giving. I was, I was moderating a panel, and I won’t say the name of the company, but the guy who was running their product thing, I said to him, Look, there’s, there are, like, how do you handle this product relevance, right? You know who I am. You know what I’ve bought before. How do you handle this product relevance, discovery, I said, on the platform, so hard, right? How can you make it better? And actually, I remember asking him that day, like, if it’s raining, why aren’t you recommending to me an umbrella? Question I asked him that day, maybe I’ve got a fascination with rain that I didn’t understand until this conversation caught in the rain a few times. But he said to me, discovery is not a problem. He said, we have two to 3 million products on the platform. And even the audience went like this, exactly like, Do you know what I mean?

Adrienne Enggist 34:23
Yeah, that is weird. Anyway.

Michael Waitze 34:25
So I love the fact that you push back on this personalization thing, because I think it’s one of these buzzwords that everybody talks about. I’m much more like this idea of relevance is the thing that you’re recommending just based on the data that I’ve shared with you, relevant, which I guess brings up another question for me, right? You talked about this creepy thing. Do you want to spend any time talking about self sovereign data? You seem very, very tuned into like, what’s happening in the tech world. So self sovereign data is this idea of like, I own my own data, and I can decide what to share with you and what not to share with. Based on what I think you need for me to make that decision. Do you, and this is a blockchain or a distributed ledger technology thing, do you think about that at all?

Adrienne Enggist 35:08
Not, I mean not, not in my day to day, we don’t refer to it in the terms that you’re describing. But there is as part of EU regulation, which is a thing that I’m pretty intimate with these days, right? The right to be forgotten, the inference of personally identifiable information that can come from usage data, for example, even the ability to say you we want to introduce natural language search, the ability to do free text input on that you can share things that are personally identifiable into any free text input that we give you. So then we need to make sure that it will disappear clearly and we don’t store it in any way that can be tied back to you. There’s a lot of things to be careful of, and I personally would like that to also be my possible for me, right? Because I’m not super keen on my data floating around. I know that there’s a certain amount that is just you want. Nothing you can do about it. Yeah, you can do about it this, right? If you

Michael Waitze 36:06
don’t want people to know where you live, then you should live like in the middle of the woods, because someday they’re gonna see you walking out of your house or walking outside your building. It’s just fact. Anyway, yeah,

Adrienne Enggist 36:14
we just have to, yeah, get used to it, right? But so the thing that I, sort of, I’ve been thinking a lot about lately is the idea of agent to agent or agentic relationships. So when we, probably in the near future, will have personal llms on our devices. We already carry this tracking device. It knows everything about us. It can understand and infer things. It will infer more. We’ll give it explicit preferences. There’ll be implicit things baked in as well. We’ll need to, I think that’s where we’ll be able to censor this data that we want to be available. And then the agent can have instructions built in about what is shareable and what is not and what is used to make more relevant recommendations. And it’s that agent to agent relationship where you will have your personal model will be a very the domain is you. You are the domain. It’s a domain expert of you. And then it will need to communicate with other domain experts, like a booking.com LLM or travel GPT kind of thing, and then be able to and that’s where these, these curation moments will happen, and then you get to be an audience to that interaction, where then you get to have the sort of fun parts of vetting the images. And does that image look real to you? This image is definitely AI generated. I don’t trust this property. Let’s move on. You know, that kind of thing I

Michael Waitze 37:34
got it.

Adrienne Enggist 37:35
That’s the rosy future I’m hoping for.

Michael Waitze 37:37
So I want to ask you one more thing. I know I’ve said that before, but this conversation is too interesting to me, but this, I promise will be the last thing. I mean, I promised before, too. But anyway, my presumption before I ask the question is that you love to travel like otherwise, I don’t know how you could do this job. Or again, it would be like being a chef and hating food. It just wouldn’t make any sense to me. Is there like in your day to day life, there’s always some travel related thing, even when you’re not at work, right? Even if you’re just getting on the subway to go somewhere, there’s movement involved you’re traveling. Is there any way for you to separate yourself, or does everything you see in the world look to you like, oh, that’s going to impact the way somebody books that thing. Do you know what I mean? Because, because, at some level, whether it’s walking on the sidewalk or eating at a restaurant or hailing a taxi or just doing stuff you do in your day to day life, that’s something that somebody does when they travel. And I wonder if that the things that you learn from that actually seep into the stuff you do at work.

Adrienne Enggist 38:36
Yeah, totally, absolutely. So a thing that I live in a highly tourist destination. I live in Amsterdam, yeah, yep, it’s, it’s one of the, it’s one of the places that’s actually suffering a bit from tourism. I’m a part of the problem. But, well, because I’m, I’m one of the people who’s making sure that people can, you know, explore and discover the world and all that. So I think it’s great, but we have to be super careful. We think about like, it’s a question I’ve been asking for years, is like, what? What is the difference between me going out and doing something exactly the same way as a tourist? You staying in the five star next door and he’s coming from my house. We’re sharing the same transportation, right? We need the same services. Why would I distinguish as a travel provider? Really, we’re talking about the ability to experience hospitality. I mean, I think, what is the, what are the limits of the things that my my company, and that I’m responsible for? I think it’s about hospitality, yeah? And I think it’s about embracing cultural experiences, even within your own or mixed cultures, that kind of thing. So, yeah, it is absolutely everywhere. I think about, you know what we do to move crowds between locations. So if you have ever been to Amsterdam, you know that some of the museums can become overwhelmed. Yeah, and you really want to see the Van Gogh, or you really want to go to the Stade lake, or the Rijksmuseum, or all of these famous things we. Could do so much more in terms of like algorithmic traffic control to help make sure that people had the best possible experience we can make if, again, if we we have permissions to understand where you are at any given moment, like the FOMO of traveling, like the idea I the one of the things I love to travel, but I always tell myself, well, I’ll just do it next time when I come back, because if I don’t, I will get paralyzed, right, with FOMO, and I get I when, when I’m, like, living in a place like Amsterdam, where I didn’t go to the Drax museum for the first decade that I lived there, like, because I had no FOMO, right? So, you know, I think that that’s Yeah, and I think we’re, I think the more that we have this always on connection to to to what’s going on around us, and to have that we’re finally to the moment where, if I’m walking through Tokyo, or actually, this actually happened. We were in Osaka, I wanted to go to this one particular denim store. Could not find it. It was in an industrial estate somewhere, and I just was devastated. And then I found out it was literally one block away from me, and if I had had my little companion with me to go, it’s just there. Yeah, turn right, you know. So that’s the kind

Michael Waitze 41:05
of thing. So again, you remind me this, and then I promise I’ll let you go. The first time I was in Vietnam was in 1991 or 1992 and I remember, when we finally got to Hanoi, there was this one little restaurant, I think was actually called the piano bar. And all we had back then was the Lonely Planet God, right? There were no GPSs. And I remember when we found the restaurant, there were like three of us, three guys, like two other guys and I were traveling together. We high fived each other because it was so exciting just to find that thing that you were looking for right, because you’re right. The next time I was in Vietnam was 25 years later, so I don’t even know if that restaurant’s still there, but I’m just saying I get that. Just saying I get that thing. There’s a little bit of fun in this serendipity as well, anyway. And then you can

Adrienne Enggist 41:47
find the little edge cases too, you know, like, it’s not like the Vedic guide back in the, you know, the 2010s or whatever. The 1910s everybody had to go through the long list of things. You know, it’s, it’s, or the Lonely Planet, or, I think there was, there’s a European version that my husband used when he did his gap here in the early 90s. Like, for sure, it’s totally different now. Like the things are fleeting, and we can discover all sorts of cool things and then make them part of the Zeitgeist. Okay,

Michael Waitze 42:15
I’m gonna let you go. Otherwise we could go on forever. Adrienne Enggist, a Senior Director of Product, Customer Experience and Platforms at booking.com. You’re awesome. Really awesome. I knew this was going to be easy, but it was easier than even I had expected. Thank you so much for doing this. I really appreciate it

Adrienne Enggist 42:27
Thank you so much. I really appreciate letting me come on and talk with you.