Atiyu

The $263bn question: Will OTAs be AI's first big victim?

‘Plan my next holiday to Tokyo’, has rapidly become the ‘hello, world’ of AI Agent demos. It seems like every lab and consumer facing AI startup wants to capture this market. So what happens if they succeed? When AI agents can seamlessly handle travel planning and booking, what happens to Online Travel Agencies (OTAs) like Booking Holdings ($177bn market cap), Expedia ($36bn), and Trip.com ($50bn) that aggregate hotel capacity and sell it on to consumers for a cut? Will decades of network effects and aggregation power simply evaporate as users trust their AI assistants instead of clicking through Booking.com? Or will they continue their slow march towards total travel dominance?

The market’s response so far has been complete indifference. OTA stocks have barely moved at each new agent announcement. In fact, the big 3 OTA stocks are up an average of 130% since 2022. Analyst reports mention AI as an ‘opportunity’ rather than an existential threat. Either analysts see something the tech world doesn’t, or they’re suffering from the same blindness that afflicted newspapers in 2005, taxis in 2010, and cable companies in 2015.

Why OTAs capture so much value

Before evaluating the threat from agents, it is worth discussing exactly what OTAs are and why they capture so much value. An OTA is a platform that holds contracts with hotels, airlines, and rental car companies, allowing consumers to book directly on their site. In exchange, OTAs take a commission on each booking. OTAs control the consumer facing side of the vast travel tech stack acting as marketplaces solving coordination problems between the hundreds of thousands of travel industry suppliers (hotels, cruises, trains, flights etc.) and billions of consumers. By solving this problem and controlling the end customer experience, they have strong negotiating leverage with hotels, allowing them to charge sky high take rates of up to 30% (avg 15-20%). OTAs have built large moats through network effects: the more inventory they aggregate, the more consumers they attract which forces more supply onto the platform etc. etc. Once this flywheel gets going, it’s hard to see a natural stopping point.

Over the last two decades, OTAs have benefitted from the dual tailwinds of a growing global middle class able to afford travel and strong network effects to grow revenues an average of 15-20% YoY. Today, c.40% of all global travel spend is booked through one of these OTA travel marketplaces, and with average take rates of 15-20%. This market power translates directly to the bottom line with businesses like Booking Holdings generating c.30-35% EBITDA margins, serving essentially as a tax on all global travel.

Agents offer a better experience

However, despite the relentless march of consolidation that the big OTAs have been on for the last few years, a dark cloud looms on the horizon: AI agents that help consumers book their trips have the potential to completely disrupt the OTA business model.

Agents offer a better consumer experience than OTAs and capture intent earlier in the funnel. Rather than forcing users to search across multiple sites, compare options manually, and piece together an itinerary themselves, agents collapse inspiration, planning, and booking into a single conversation.

But can they actually deliver today? Partially. I tested mindtrip.ai by showing it a trip I saw on Instagram, entering my budget and loyalty points balances. Within minutes, it returned flights, hotels, and a day-by-day itinerary. What would have taken hours of research across Tripadvisor and Google Flights took five minutes. The recommendations weren’t perfect, but they were good enough to be useful.

This is the key point: agents don’t need to be flawless to disrupt OTAs. They need to be good enough that users prefer them. The bar is low. OTAs offer a frustrating experience: endless scrolling, no personalization, no understanding of your constraints. Agents can factor in loyalty status, seat preferences, budgets, and trip context in ways OTAs structurally cannot. They can monitor prices after booking and rebook if rates drop. As the underlying models improve, the gap will only widen.

With this kind of experience, it seems almost inevitable that agents will become the primary consumer interface for travel, displacing OTAs.

How OTAs get squeezed

This has multiple consequences for OTAs:

First, they’ll lose their leverage over hotels. Today, OTAs can charge hotels 15-30% commissions because hotels have no choice but to list with them. If a hotel isn’t on Booking.com, it becomes invisible to nearly half of all travelers. That’s enormous power. But when consumers start their search with an AI agent instead of an OTA, hotels are no longer hostage to Booking’s distribution. The billions OTAs have spent on brand marketing (Booking alone spent $10.4 billion in FY24) will matter far less if consumers never visit their sites.

Second, agents will commoditize OTAs by putting them in direct price competition. Today, most AI travel agents still complete bookings through OTA infrastructure. They function like a price-comparison layer: the agent searches across Booking, Expedia, and Hotels.com, then shows you which one is cheapest. As more consumers book this way, OTAs will have to compete on price rather than brand. Commissions will compress across the industry.

You might ask: isn’t this just metasearch? Platforms like Google Hotels and Kayak already compare OTA prices and put them in competition. True, but that traffic is brutal for OTAs. Bookings that come through metasearch generate roughly 0% operating margin for Booking.com. So why hasn’t metasearch killed the OTAs? Because the experience is worse. Metasearch is clunky. It sends you bouncing between sites to actually book. That’s why it captures only 10-15% of bookings.

AI agents are different. They don’t offer a worse experience than OTAs. They offer a better one. That means the share of bookings flowing through this low-margin channel won’t be capped at 10-15%. It could be dramatically higher.

Finally, there is the potential for Online Travel Agencies to be disaggregated entirely. Under the status quo, the big OTAs have invested billions of dollars and many years of effort in building integrations across the deeply fragmented hotel market, allowing them to maximize the size of their inventory. In a world in which the AI travel agent market consolidates, there is strong incentive for both the agent company and the hotels to cut the middle man out entirely, and for agent companies to build integrations with hotel property management systems directly. This enables them to offer higher payouts to hotels and simultaneously offer cheaper deals to consumers, further crowding out OTAs.

Enter: Directbooker, a new startup backed by the former CEO of Tripadvisor and the former head of Google Travel, that intends to develop new AI-native supply side aggregation of inventories to help hotels bypass OTAs. The startup is looking to aggregate availability, rates, and inventory (ARI) information, and help hotels directly pipe this data to AI companies (including mainstream AI products like ChatGPT and travel specific agents) using an MCP (Model Context Protocol) endpoint. This allows hotels to bypass OTA fees and provide more details to agents about the specifics of a hotel (e.g. whether or not it has a hot tub) than traditional OTAs allow.

If this model scales, it represents a structural unbundling. AI agents will capture intent, supplier-aligned aggregators provide the pipes, and OTAs get squeezed from both directions.

Why OTAs won’t win the agent race

This is not to say that the big OTAs are going to go down without a fight. Firstly, while independent travel planners have been growing significantly in popularity, the big OTAs are also building competing products. On the consumer facing side, the big OTAs like Booking and Expedia have started to build out their own AI based offerings. Expedia now offers a product that will plan trips based on instagram reels. Booking.com has been steadily rolling out AI features and has signed partnerships with OpenAI to build its own AI trip planner. Additionally, the OTAs have the financial heft to outspend any startups in the space.

So why will they lose nonetheless? First, OTAs face a classic innovator’s dilemma where any potential AI based offerings cannibalize their core business and potentially compresses their margins. Second, these businesses have been built through decades of negotiating supply agreements and building connectors to hotel software, rather than by building delightful consumer experiences, which is likely the axis of competition for AI trip planners. Third, distribution is shifting. OTAs built their moat by spending billions on Google ads to capture search traffic. If consumers start planning trips inside ChatGPT or Claude, that spend becomes worthless. Put simply, incumbents with resources often lose to startups when the new product requires cannibalizing the old one.

However, the monetisation model of agent products is still not a settled question. There is a world in which agents monetise via advertisements, in which case the OTAs can use their substantial free cash flows to cement their position and resist commoditisation. However, whether or not consumers will be willing to accept and use agent products that serve targeted ads remains to be seen. In fact, I would wager that consumer products in this space monetized by ads are unlikely to win, as consumers are probably less willing to make purchasing decisions if they know that the agent is suggesting travel options based on ads, rather than on what might be the best fit.

Conclusion

In conclusion, the AI Agent vs. OTA platform story is simple. OTAs are middlemen that are able to charge large fees because of the gigantic coordination problem they solve with software. But in the age of intelligence, they are going to get squeezed on both sides of the network. Travel focused startups and core LLM platforms alike are starting to capture the consumer funnel for research and booking, and hotels now have options to go directly to these AI platforms. Both sides of the network have always had a strong economic incentive to cut out the OTA, and with AI they now have the means to do so. It seems almost certain that OTAs will not be able to compound earnings over the next decade in the same way that they were able to for the last one.

While I think OTAs are likely one of the first to fall, similar stories are likely to play out across a number of large established software categories where legacy middlemen have built previously insurmountable moats on network effects. AI-native apps and infrastructure will create superior matching experiences tailored to customer preferences, and generate a platform shift that allows suppliers to bypass incumbent gatekeepers. Some categories that might be next to fall are:

In each case, incumbents look difficult to unseat today, but billions of dollars of enterprise value will be created in the platform shift that unseats these middlemen. If you are building a business of this shape, in travel or otherwise, I’d love to hear from you!

PS: Thanks to Trevor for reading early drafts of this.