AI Is Breaking Travel’s Search Economics.

On Monday’s Good Morning Hospitality, A Skift Podcast, Brandreth Canaley, Michael Goldin, and Jamie Lane break down what AI is actually doing to travel economics and operations right now.

The conversation opens with a story that stops you in your tracks: one traveler used Claude to run 881,076 searches on Etihad Airways to find one flight. The infrastructure cost of that search exceeded the commission.

From there, the team digs into Brian Chesky‘s move to build an AI lab outside of Airbnb and what it signals about the platform’s AI strategy, before closing with a look at what real AI adoption looks like at scale through Evolve‘s playbook of deflecting 60% of guest inquiries without human intervention.

This episode is presented by ⁠⁠Cloudbeds⁠⁠, Bilt⁠⁠, and StayFi.

Watch This Episode

Good morning.

Happy Monday.

Best day of the week. We’re back in full strength. This is good.

We’re getting in a little summer routine here.

I know, before we all start traveling for summer.

Brandy’s tied to her desk because it’s June, and she is sworn off travel in June.

Yeah, I actually, because I, well, swore off travel in June, I had like a great south shore, like solo adventure day on Saturday, and ate like so much great seafood, got like a ride on my friend’s boat for a couple hours, and I did that all because I

didn’t leave Boston. So.

Vacation, so you did travel.

I mean, I drew, I’m counting, that was like a little day trip. I drove like an hour.

Okay.

I slept in my own bed.

It’s always nice.

Yeah, I had a nice weekend with the kids. Went out to dinner on Friday. Actually, we had a baby shower.

We had a nice dinner on Saturday night. So it was kids went to the grandparents. It was a good weekend.

Vacation for you too.

Yeah.

Lovely.

Well, my wife was out of town much of the weekend, so it was boys’ weekend. Let the chaos ensue, you know? So I survived.

No broken bones, no nothing like that. So a good, not so relaxing, but a good weekend.

Wife is back though for this week?

Yeah, she was back at night. She was just gone all day. So day trips like Brandy was doing.

But back and then she’ll be out of town again this weekend. But tomorrow, going to see Messi against Argentina. So not Argentina against Iceland.

Amy is going to Iceland soon. So I invited him down, but he doesn’t like Albert. So yeah.

Anyone starting to get World Cup fever?

Okay, I had this, I want to be more excited about it.

I don’t know. It’s just like not catching. I don’t know.

And I’m in a city where it’s happening, maybe when it actually starts. But part of me, it feels like an old man attitude of like, I don’t want to go in all those crowds, go like down into like downtown Boston.

But I think I need to find some cool locations that are doing stuff maybe out in my neighborhood and not in the central watch parties.

Yeah. So this is a good tee up into the first article. But so I’m watching, trying to find tickets.

I’m like, why am I like hitting these sites every day? So I vibe coded a scrapier to go and look at ticket prices for each match across Ticketmaster and StubHub. So Brandy, if you can throw that up.

So this is how tickets are trending, and at least over the weekend, not trending down yet, but $228 for the cheapest match. Czech Republic for South Africa, Congo versus Uzbekistan. Still not cheap, but not crazy.

Under $1,000 to get the whole family in the thing. But it is just an example of how easy it is now to spin up scrapers that can just run however many times you want across all the different ways you’d want to filter, analyze tickets.

It’s pretty incredible how these things can work these days.

They looked pretty flat. I didn’t see a whole lot of dynamic pricing in there, Jamie.

No, this is, I’m hoping they just start collapsing. So for those listening, it is just a bunch of flat lines.

Not the prettiest graph, not the most exciting graph, but hopefully we’ll see once you get closer to the game.

5:28

AI Travel Search Costs

Well, before we get to our first story, the main tag of that, the first story is that someone ran a search on Etihad, and it came back with almost 900,000 searches.

So we’ll get into how scraping is changing, how we all search and what the companies are doing to deal with that. But before we get into the first article, we want to shout out our sponsor, Cloudbeds.

They unify your operation, distribution, guest experience and revenue marketing in one place, powered by Signals, Hospitality’s first foundation AI model.

So whether you’re trying to drive more direct bookings, which we all are, I guess, depending on the episode that we’re talking about, cut training time or finally get your data working for you, Cloudbeds is built for what’s next.

Learn more at cloudbeds.com/gmh, and you can find that link in the show notes.

So with all these searches, not Jamie’s scraper for World Cup tickets, but for people who are doing really crazy travel searches, the cost of all of those is becoming part of the conversation. So who is paying for the power to look at?

That’s one flight that he did almost 900,000 searches for, or not he, the agent did.

So what is your take on how you see, the whole industry is grappling with not only just the change in behavior, but now there’s an infrastructure and cost component that is kind of taking people by surprise.

So Michael, can you explain just like what this infrastructure looks like sort of behind the searches?

Nope, I sure can’t. I do know that the more structured your data is, the easier it is to search and query.

Now, because the training that AI has, it’s going over thousands, hundreds of thousands of searches, which is not how a human typically searches. We have our go-to places, and I think this is where it will evolve over time with AI.

It’ll get to know us and know what we like and know what we book. But until then, it’s just absolutely buckshot. It is everywhere.

Birdshot, not buckshot.

There’s someone out there right now who’s listening to this, just like screaming in their car.

Jason.

Yeah, Jason.

But the reality is like how these agreements get set up for, and these search companies or these OTAs that are then hitting the GDS to be able to serve all these airline rates, build out the connected trip.

It’s sort of predicated in the prices underneath it are that someone’s going to search three, five, ten different combinations of trip length and location to get a sense of prices. And then they’re going to go about and book one of those.

And that the underlying costs of three, five, ten different searches is reasonable for what the GDSs are going to charge and ultimately what the OTA is going to get in terms of the commission. But the article…

25 years of data. Yeah. It shows their conversion rates.

They got top of funnel. They got all the way down.

And just so we know, those costs are two to $10. So pretty manageable per ticket.

Right. But then if someone has an agent doing those searches for you, that 10 searches… I think the article…

And it does a good job of pulling out this LinkedIn post that is just throwing out like I’m someone going ridiculous of 800,000 searches, but I’m a typical agent, if you asked it to go and search these five different nations, these five different

destinations, these different date combinations to help me find the right search. And we’re looking at hundreds of searches now, or maybe even thousands. And the economics of running an OTA and hitting the GDS to serve that up don’t work.

And so the article goes into I think some ways they’re trying to combat it. The one is like catching all the information.

Like if you search it, someone searches it once, they can sort of save it and serve it back and not have to hit the GDS to be able to pull that back.

And the more people that are searching and the more sort of results that you can sort of have stare stored and serve back like they’re not getting charged for every single one of those. But that is infrastructure costs on their side.

But and there might have to be some big changes on either the cost of what the GDS is charge or that and these sites are going to try to start blocking these scrapers that and that they’re just not going to allow on all this search traffic to come

through their site. through their site.

10:37

Legacy Tech AI Clash

Well right now this is a great example of the clash of like legacy infrastructure and technology that really hasn’t had to adapt in a very long time with a new product that is getting more sophisticated every day.

And I think that, I mean, I certainly don’t think about it when I go to Google and I like look up flights. I’m not thinking about the technology and like the rate systems that airlines are using to load their rates on, right?

Like I’m not at all thinking about that. But that technology from the article, it sounds like that hasn’t been updated or advanced in a very long time.

So, they’re playing like a kind of a silly defense where they’re like, how do we like prevent this from happening instead of how do we like radically change our infrastructure, which is, I mean, I’m not saying that’s simple at all, but like, it’s

like trying to, I don’t know, trying to fight this tsunami with like a small like sandcastle sandbucket trying to, you know what I mean? Like it’s just not going to happen.

Yeah, doing that or building a fort around yourself is not going to work either. If you’re going to block all AI transactions, you’re going to lose.

Saber and Amadeus have, Saber did some caching, Amadeus did some machine learning that filtered 175 million unproductive transactions a day. So absolutely massive volume.

So we’re in this weird transition phase, where both AI, AI on both sides is going to get smarter. On Amadeus side, the machine learning looks to be really productive.

We don’t know what that is in a percentage, but 175 million transactions blocked today is going to be impactful, whether it’s 1% or 80%.

But AI is also going to know Brandy and Jamie and Michael better to where it doesn’t have to go look at Spirit flights or Frontier flights or anything besides Delta for flights for Michael or Jamie as well.

So, you know, we’re just in this weird phase. And yes, you don’t want to be caught holding the bag, losing money on every transaction for the next, I don’t know, two years while this tech transition happens.

But I don’t know if there’s not an easy fix, unless your tech team is really good and can spin up some some band-aids real fast.

Well, and I think it’s interesting that the source of for this, or the start of this article is that that was a guy searching for one flight, right? And so you think about, I mean, now because you can just kind of like talk at any of these agents.

And so you have people that are trying to search like, oh, I could go here or I could go here and I could go this day or I go this airline. Like, I can see the potential for just like an absolutely astronomical volume of search queries.

And that’s just someone like, that could be like someone just like had a couple glasses of wine and is like, lalalalala. Like I’m just trying to like look at stuff with maybe no even intention to book.

So yeah, it’s a it’s I don’t envy the role of like the CTOs and CIOs that are trying to figure out how to manage not only the new technology and how you adapt to it, but then there’s so much security stuff that goes around all of this.

Like those two positions right now, stressful.

Well, and Paul had a really good comment. Legacy versus Gentic infrastructure is going to be a really big story moving forward.

Now, on one hand, and we’ve talked about this on the show before, legacy systems have all of the data to make the AI models better.

But if they’re built on the infrastructure that’s 25 years old, and they’ve just got all this tech debt stacked up through years, they’re going to be looking at in the room talking about, do we just scrap this all and rebuild?

Or is there another way around it? Jamie, what’s your take on Legacy versus Gentic?

Yeah. It’s been at the forefront of us thinking about, we’re building this new dynamic pricing tool and how our old infrastructure was. And that’s, and AirDNA was started in 2014.

It just couldn’t support what we wanted to do with a new Gentic tool. It was better to just scrap everything we had, start entirely from scratch, to support this new infrastructure and new product.

So I think, and John’s comment is, I think, relevant. Like there are startups that seem to be building businesses on current token costs. The investors are not investing and keep losing billions.

These startup models will collapse overnight. If you’re building this new tech on top of old infrastructure, I was just saying, the math doesn’t math. As the fun infographic on Skift.

I think there’s ways to develop new infrastructure off of the old data. The old data sets and old data and models still have everything you need. You’re now building for a world where you’ve got to serve millions or billions of searches.

It just takes evolving the infrastructure to be able to support it.

All it’s trying to do, Jamie, is to make you happy. So it’s searching as many areas as it possibly can to make you happy. There’s a quote in the Skift article from Gurav Roy.

It says, AI-driven interactions are often more contextual and commercially relevant than traditional search patterns.

If you have your data clean, structured, AI-legible, it’s going to help you win, especially in the short and intermediate term, more times than not, as long as you’re answering the question and making AI let Jamie be happy.

But also, there’s the downstream applications of this volume of things that are built off of search volume.

Like, what are all these pricing tools now that you search as an indicator, where if you’ve got all these searches that are now not actually an indication of booking, but are just an agentic tool and you’re adding a whole lot of noise to any signal

Well, also, I feel like the whole kind of all of the internet analytics, especially like any marketing department uses, like I think all these KPIs are just going to get like blown up.

We’re going to have to like scrap them and start again because…

It’s going to be AI conversion rates, right? It’s like, who cares how many people are going to see your site? It’s going to be…

Are you pleasing the robots?

Well, I even noticed like my own personal behavior of… Instead of going to Google to ask something, I’m just talking with Claude. Like I barely…

And I was like, wow, that’s just… It’s so interesting to see how your own personal behavior changes. But before we get into our next AI topic, it’s an AI heavy day today, folks.

I want to shout out our sponsor, Bilt. They’re helping restaurants and hotel F&B teams better understand guests and create more personalize experiences that drive repeat visits.

If you care about loyalty and guest experience, as you know, all of you do, it’s absolutely worth checking out. Head to joinbilt.com/gmh and you can find that link in the show notes.

18:43

Brian Chesky AI

So this next topic, Brian Chesky building an AI lab, is we don’t really have a ton of information on this yet. So this is just like a call out, something to be paying attention to. But this is not happening in Airbnb.

This seems to be a side project. And so it’s interesting that this is a side project and not within Airbnb. So do you guys have any initial thoughts on that?

Yeah, I talked to a buddy of mine who’s a VC in San Francisco last week, and he’s like, it’s wild out here.

It is the wild west. It’s an absolute gold rush. There’s so much money flowing around everywhere.

That’s in my read, that’s what this is. I don’t, I wouldn’t anticipate it has a ton to do directly with Airbnb. I’m sure he’ll have ways to tie it back in.

And he’s also publicly good friends with Sam Altman. So I’m sure Sam’s in his ear of like, oh, you should do something like this. Or I wish we had done X, Y and Z.

Brian is a design forward guy. Nothing in AI right now is very design forward. So I would expect to see something blending those two.

But I don’t know how much agents care about looks and design, but maybe they do.

You know, he is always talking about, he wants to be for like the end consumer, the individual host, the person with their shared room. So maybe this will be for the user interface, give something back to the people, something to look at.

Also, we were talking, he just needs something to spend his money on.

It’s worth what, $10 billion now?

Yeah.

But it’s not going to be infrastructure because you need at least $10 billion to get an AI infrastructure project stood up.

Yeah. He’s a guy with very strong opinions on how design should work, and how user interaction should work in this age of AI. He obviously has an outlook through Airbnb, of which to impact a single app and a single user flow.

But if he has broader vision on how consumers, how companies should be leveraging AI in their tools, I think it’s bold and a big bet that he can create a lab that can sort of impact a much more broader ecosystem.

So I do think he has a lot of power within Airbnb to show other companies the way through. And here’s how we did it.

And how many companies out there, how many, just think of every single website that exists in the travel space now, like that is essentially copied Airbnb over the years in terms of their design, their UI.

But maybe this is a way to go and try to monetize it a bit more.

Yeah, so we’ll just keep a look out. I mean, we’re always seeing what interesting things he’s up to. And yeah, we’ll see.

And also, if he wants to spend his money on something else, he can take me out for dinner. That’s totally fine. I’m open to it.

Just hopefully he never listens to any of my brandy rants about this podcast over the last five years. And before we get to our third and final AI article of the day, I want to give a shout out to a new sponsor, StayFi.

StayFi helps short-term rental operators build direct guest relationships through branded Wi-Fi portals, email marketing, and guest data. We actually were big users of StayFi at Romy, so I love Arthur and his team.

They turn one-time bookers into repeat guests, so getting those elusive guests that come through with the masked emails, that’s key for growing your BookDirect data.

Start growing your direct booking revenue today at stayfi.com/goodmorninghospitality, and you can also find that link in the show notes.

22:54

Evolve AI Deflection

There was the big Skift Data and AI conference last week, lots of really fun nuggets that were coming out of that.

One of them was the interview with Evolve, and the crazy stat coming out of that was that they have deflected 60 percent of guest inquiries with AI, and that’s a huge number, and we’re always talking about how effective are some of these agents at

really solving the problems. But 60 percent is a pretty significant number, and being able to also focus that human interaction, that need for high-level judgment, that’s where they’re focusing their human attention.

I think this is a really great example of how this can be successful.

Yeah, I think they’ve taken it step-by-step and rolled out AI in listings, and guest communications, and pricing. They’ve run all the way up and down.

What scares me a little bit is their business model in general was basically getting the guest to the property, and then local on-site team, or a maintenance person, a cleaning person has to go in and do the local stuff.

They’re 20 steps ahead of anybody else in this space, but this is a fairly disruptible business model now, more than it was five years ago with the rise of AI.

A lot of these things exist in point solution cases, but Evolve has done a good job of rolling those in already, and I’m sure their headcount or productivity has improved, one or the other, or both.

I’m curious, I’d love to chat with some of them on what their plan is to keep competitors at bay, because I do anticipate they’ll start having more and more competition in this segment in particular.

Well, yeah. Oh, sorry, go ahead, Jerry.

I would say, you think about their business model, it’s like, it’s marketing, it’s communications, it’s revenue management. Like, all three of those things should be able, they leverage huge amounts of AI to automate aspects of it.

And then it just becomes, for a host, do you want to hire Evolve to do all of it for you? Do you want to piecemeal it through the different point solutions out there? Do you want to have your PMS try to do it all for you?

And each of those sort of charging for it. So, and it could just be that they capture the people that just want to all in one solution, just sign me up, do it for me, don’t and do a great job.

And AI helps them do a significantly better job because the ability to automate it and do it cheaper.

Or that their ability to charge as much for it just gets compressed because of all the other tools that can do it and just as well, but at a fraction of the cost of what Evolve charges.

Well, and that part of I think we’re also in this era where everyone’s like, you can just vibe code your own, blah, blah, blah, blah. And it’s just like, I’m not on that train. Not everybody is as comfortable with all of this technology.

I think that there’s going to be the need for the Evolves, or the people that are going to do all of these solutions for you.

I do agree though, that cost is going to be like, the commissions are probably going to have to come down at some point, because there’s just going to be increased competition among maybe even bigger PMs, or PMs that are smaller right now, but they

have really tech-savvy people on their team. I think the competition is going to start to come from more upstart groups that have maybe these more tech-savvy teams.

But one interesting thing that I noticed from the article was about owners and damage claims and things like that, and that’s one part of the business, like super unsexy is insurance and damage claims, things like that.

And that’s a part, like that’s something, the huge component that automating would be fantastic. Like if you’re people, it’s such a time suck of figuring all of that out. So that’s like any operator out there with whatever AI tools you’re looking at.

If you’re trying to think, what can I do right now that might really help the business and free up some time, figure out if you’re PMS or something you vibe code or something can help you deal with damage claims and insurance, because it is, that’s a

Yeah, fair enough.

27:39

Agent Experience Shift

One of, and this probably isn’t a surprise, and this ties in with everything else we’ve talked about today.

One of my favorite quotes from the interview is, so my big prediction over the next 12 to 24 months is that we switch from a user experience to an agent experience. Beat that drum, beat that horse dead at this point.

But we’re already talking about it. We’re seeing 175 million deflections from the Sabre a month or a day. That was a day.

Obviously, this is going to happen. Now, can the companies adapt? Can we as an industry adapt fast enough to capitalize and to make sure we’re profitable on it?

We got a question come in from Gabriel, and I just pulled the data real quick because I love the stats.

I love the real-time data poll, Jamie.

Yeah.

If Devolve is deflecting 60 percent of inquiries with AI, what does that do to the guest experience benchmark? Are scores up, down, or flat? That is the real test.

I just pulled point in their communication score and their overall score across Airbnb for the past 12 months, past 24 months.

Their communication score past two years is up, past year is flat, and their overall score is actually up pretty significantly the past two years.

So the fact that they are able to deflect 60 percent and at least maintain if not increase their overall scores is, and I think a pretty good benchmark. There is still an at around an industry average, like overall average for Evolve is a 4.71.

So that’s not great, but that’s pretty typical for a large property management company.

Unless you tell your AI to be grumpy that day, you don’t have human emotions on the experience side to weigh. So it makes sense why reviews scores shouldn’t change a whole lot.

Well, and also think about so many of those comments are probably like, is there Wi-Fi? How far is the pool? All that kind of stuff.

So when you then you have someone who’s irate or they’re super lost, they need someone to call them and get whatever like then you have those human agents that are probably now have more time to actually communicate with them.

And then the AI is kind of getting rid of the riffraff or like here’s our FAQs, whatever. So I think that that is a bonus. But it’s a great point that we should be monitoring these.

You know, any operator should, if you implement any of these kind of agenic tools, you should be monitoring those communication scores because the customer will tell you pretty quickly if it’s working or not. That’s real, almost real time feedback.

Yeah.

Amazing.

Well, to put a bow on the Evolve stuff, sounds like they’re taking, we haven’t heard a whole lot about them in the industry for a little while, for whatever reason, at least I haven’t.

But sounds like they’re taking all of the right steps that they need to be taking and they’re getting out ahead of this trend.

Again, the companies that have the most data have the biggest advantage so long as they have good competent CTOs in place that are putting in steps to be ready for the future or ready for today, what we’re already seeing.

So, fun episode, probably more technical than I’m good at.

Yeah, someone was asking, the question started to get technical and I was like, ooh. But we are also learning in real time more about all of this. So, it’s exciting.

My emotions about it oscillate week to week. Do you guys have anything fun coming up this week?

Got the soccer game.

Oh, yeah.

World Cup starts, kickoff June 11th, Mexico, South Africa. It’s exciting.

Yeah. I’m going on a one-night trip to Sarasota to go down to Prime, which I’m like, it’s a real boomerang trip, but I’m trying to commit to my, that’s a work trip, Michael, so that doesn’t count.

I figured, no, it does not count. You got to go where work demands.

Yes, exactly.

We’ll save travels.

Thank you. Excited to hear about the soccer games this week. It’ll be fun.

Yeah, it should be interesting to see how a bunch of people in Auburn, Alabama react to soccer because that’s not a sport that we have in this city.

So I’ll report back.

Amazing. Well, thanks to all of our sponsors and all of our listeners. Amazing comments today.

We love when the audience participates with us and asks us some interesting question. Shout out to Jamie for pulling real-time data right there at the end. So hope all of you guys have a great week and we will see you next Monday.