Booking.com: Designing a Generative AI Trip Planner Around Traveler Preferences

You are a product manager at Booking.com. How would you apply generative AI to build a trip plan tailored to each traveler, using what they tell you and what you know about their preferences to choose the destinations, the activities and the places to stay? Explain which travelers you would build for first, what the experience looks like, how the AI produces a plan that stays accurate, and how you would know the product is working. ```hint Bookable beats plausible A language model can write an appealing plan in seconds, but a plan only helps Booking.com and the traveler if it can be booked and still holds up on arrival. Decide where in the planning journey a generated plan adds the most value, and what live data each part of it must rest on. ``` ### Clarifying Questions - Is success measured by more bookings per trip (adding activities, car rentals or taxis to a stay), by winning travelers who have not chosen a destination yet, or by repeat use and loyalty? - Which travelers come first: people still deciding where to go, people planning a trip with several stops, or people who already booked a stay and want to fill their days? - May the plan include free or third-party activities, or only what can be booked on Booking.com? - Which surfaces are in scope: search results, a conversational assistant in the app, or the trip page after booking? ### What a Strong Answer Covers - A business goal for the planner, and why generative AI serves it better than adding filters or improving ranking alone - A chosen traveler segment and planning stage, with concrete pain points ranked by reach and severity - An itinerary experience that ties destinations, stays and activities together, lets the traveler edit and lock parts of the plan, and leads straight to booking - Grounding of generated content in live availability, prices, opening hours and travel times, with a check that catches impossible plans - Success metrics tied to incremental bookings and plan quality, with guardrails for wrong facts, privacy, and conflicts between commission and fit ### Follow-up Questions - The plan sends a family to an attraction that turns out to be closed on the scheduled day. How do you prevent this, and how should the product recover when it happens anyway? - Should a partner that pays a higher commission rank above a better-matched option in the plan? How would you decide? - Travelers who open the planner were likely to book anyway. How would you show that the planner caused additional bookings?