Booking Spreads Its AI Bets Across Five Projects as Expedia Narrows to One

Booking Holdings is running five separate consumer AI projects at once, including two stealth internal startups, while Expedia has killed its all-in-one assistant and rebuilt around cooperating specialist agents. The two largest online travel companies have now picked visibly different answers to the same question, and the results will show up in distribution economics before they show up anywhere else.

The detail came out of executive comments at investor conferences last week and was reported by Skift on 14 September. Three Booking brands are building trip-planning tools: Priceline, Agoda and Booking.com. Alongside them sit two arm's-length ventures run as greenfield startups, Lola.com, led by Kayak co-founders Steve Hafner and Paul English, and an unnamed West Coast venture. Chief executive Glenn Fogel has told both to operate independently of the core business.

Priceline's Penny is the furthest along. Recently rebuilt on an underlying agentic system, it is the one Booking executives keep citing on earnings calls. "We're definitely getting good CSAT (Customer Satisfaction Score) numbers," Fogel said at the Citi 2026 Global TMT Conference, adding that engagement and conversion were both up. He also called it a "very small sample," which is the qualifier that matters.

Agoda is next in line. "Agoda will launch something similar in the near future," chief financial officer Ewout Steenbergen said at a Goldman Sachs conference. Agoda has already started a limited public beta of an AI assistant for selected users of its mobile app, built to turn a loose travel idea into a bookable trip.

Booking.com itself is being handled more carefully. Skift reports the company is working the funnel upward from the booking step rather than rebuilding the top of it, on the reasoning that its existing conversion path is too well optimised to disturb. That is a revealing piece of caution from the platform with the most to lose, and it suggests operators should not expect the Booking.com search and booking experience to change shape quickly.

Expedia went the other way. It scrapped Romie, its attempt at a single chatbot that handled the whole trip, after concluding the end-to-end model was not practical, and moved to what it calls a multi-agentic architecture of specialised agents that hand work to each other. It bought Berlin trip-planning startup Layla in June to speed that up.

What this means for accommodation providers is less about the assistants themselves and more about what feeds them. Every one of these tools needs structured, current content about a property to make a recommendation it can stand behind. Rate, availability and photo feeds were built for a search results page, not for an agent answering a question about whether a property suits a family with a toddler. Properties whose content is thin in the channel manager will be the ones these systems cannot confidently recommend, and there is no ranking dashboard that will tell an operator that is happening.

Neither company has published usage or revenue figures for any of these products. Skift's advice to investors is to wait two to three earnings cycles before judging which model works. Operators have a shorter horizon than that, because Agoda's assistant is already in beta and whatever it says about a property is being said now.