S16.3 · Online Travel & Distribution
A sub-segment of Online Travel & Distribution. Market structure, and how AI is reshaping it.
Leisure OTAs are the multi-category online travel agencies that take a consumer booking directly and earn a merchant margin or agency commission — Booking.com, Expedia, Hotels.com, Priceline, Trip.com, MakeMyTrip, Despegar, eDreams ODIGEO, Agoda and Traveloka among them — across flights, hotels, packages and ground transport. Referral-only sites that hand the transaction off elsewhere are metasearch, a separate category; short-term-rental marketplaces, hotel brand and operations, airline operations, GDS infrastructure, franchised agent networks, packaged-holiday distribution and corporate managed travel all sit outside this boundary. The business model, stripped down, is renting demand from Google and Meta at enormous scale and clipping a take rate on what converts. The newest threat aims at exactly that: AI search and agentic checkout intercepting travel intent before it reaches an OTA's own search-and-compare function. One operator, Booking Holdings, has already priced the risk with a disclosed writedown. Its trading multiple has not moved on the thesis.
The global travel market is $1.72T (2025), projected to reach $1.8T by 2027 (Phocuswright, 2025). Online bookings are on pace to reach roughly $1.2T by end-2026 (Phocuswright, 2025), implying online penetration near two-thirds of total travel spend — a derived range, since no source states the ratio directly.
At the top, concentration is extreme: Booking, Expedia, Airbnb and Trip.com took 96% of the sector's $58B revenue pool in 2024 (Skift/Phocuswright, 2025). Strip out Airbnb, whose revenue belongs to the short-term-rental category, and the pure leisure-OTA pool is smaller and even more concentrated among Booking, Expedia and Trip.com, plus regional champions MakeMyTrip, eDreams, Despegar and Traveloka. The leaders outgrow the market they sit in: Booking Holdings' revenue reached $23.7B in 2024, up 11%, on gross bookings of $165.6B, up 10% (Booking Holdings, 2024 10-K/earnings release); Trip.com Group's revenue reached RMB62.5B ($8.9B) in 2025, up 17% (Trip.com, 2025 20-F).
Rented demand is the segment's defining fact, and the marketing line proves it. Booking Holdings spent 31.7% of revenue — $6.8B, up 13% — on marketing in 2023, equal to just 4.5% of gross bookings (Booking Holdings, 2023): traffic bought from Google and Meta at enormous scale, even as per-booking efficiency improves. Blended hotel commission runs 15-25% (Cloudbeds; Mews, 2026); air commission is near zero, replaced by service fees. Booking's gross-bookings-to-revenue ratio — roughly 7x, revenue equal to about 14% of bookings — is the segment's take-rate proxy. The merchant-of-record model is gaining share via Booking's payments platform, though processed volume is undisclosed; direct-booking share for chain hotels is directionally rising on loyalty pressure, with no reliable aggregate figure.
Three conditions frame the AI question here, and the first is whether the OTA holds a regulated risk position. It does, but only partially: Booking's payments platform carries genuine chargeback, fraud and FX exposure approaching a regulated position, IATA accreditation bonds air agencies, and the EU Package Travel Directive and UK ATOL impose insolvency-protection bonding — but only on the "dynamic package" slice of bookings, not standalone hotel- or flight-only transactions. Outside that slice, this is compliance cost, not a moat. Second, the OTA owns the transaction surface cleanly, with no inventory risk. Third — and this is the crux — supply fragmentation cuts differently by geography. US hotel supply is roughly 70% chain-affiliated (10minhotel, 2025), a mature, consolidating base; China is about 40% chain-affiliated with a larger independent base; Europe remains majority independent, with chain consolidation advancing everywhere via conversions and soft brands. Commission erosion should therefore concentrate in mature, chain-heavy markets — the US especially — with more runway where supply stays fragmented: Europe, emerging Asia, Latin America. Air confirms the pattern: consolidated airline supply already sells at near-zero commission.
Keep the two AI effects separate, because they are not the same trade. Cost-side automation of customer service and merchandising labor is an opex gain that never touches revenue, and it is already underway. The consequential question is whether an AI agent can intercept travel intent before the traveler ever reaches an OTA's own search-and-compare function — unresolved, on a two-to-five-year horizon. OpenAI abandoned in-chat Instant Checkout for complex categories including travel in March 2026, retreating to search-and-discovery, with Booking and Expedia running apps inside ChatGPT on unchanged commission economics (Skift, 2026) — a reprieve, not a resolution. Hotel chains are meanwhile building direct-agent presence: Marriott is bookable via Google's AI Mode, and Accor's ALL app runs inside ChatGPT (Hospitality Net, 2026). Three uncoordinated commercial models now coexist — OTA apps inside a chat agent, an aggregator-toll model, and proprietary hotel servers on the Model Context Protocol — with no settled commission structure, and OpenAI/Stripe's Agentic Commerce Protocol is itself a new intermediation layer that did not exist eighteen months ago. A possible new fee pool is being live-tested through these integrations now.
How are operators themselves pricing the risk? The clearest evidence came from Booking Holdings, which booked a $457m impairment against Kayak's trade name and goodwill on 29 October 2025, naming the cause explicitly: "free traffic that everyone in the business has been getting is now paid traffic," driven by Google's shift to AI Overviews. Trivago and TripAdvisor recorded parallel writedowns of $52.6m and $36m on the same mechanism. Yet none of the three parents' public trading multiples moved specifically on the disintermediation thesis; what movement there was tracked growth deceleration, interest rates and DMA compliance costs rather than a priced AI discount. When OpenAI walked back in-chat checkout, the reaction was a relief rally, not a lasting re-rating — the signature of unpriced risk, not assessed risk. On the most liquid comp set in online travel, operators are marking AI risk on their own books before public multiples reflect it.