Solution
Booking.com integrated OpenAI's GPT models into its existing marketplace APIs and data systems to make travel discovery conversational rather than filter-led. The first implementation, AI Trip Planner, translated natural-language prompts into structured travel parameters such as destinations, dates, and availability, then combined proprietary property, pricing, and cancellation data with unstructured inputs like reviews and listing descriptions to generate destination ideas and itineraries. The same foundation expanded into Smart Filters powered by GPT-4o mini, Property Q&A grounded in user-generated content and property descriptions, AI Review Summaries that condense large review sets, and Help Me Reply for partner messaging. Because the models were wired into existing APIs and infrastructure, Booking.com could hack, prototype, and launch the first trip-planning experience in 10 weeks while continuing to iterate on deeper personalization and support automation.
Data flow
Traveler prompts enter AI Trip Planner, where GPT models map conversational intent to structured fields such as dates, locations, and availability. The models then combine Booking.com's proprietary pricing and inventory data with reviews, images, listing details, and property descriptions to generate itineraries, filtered search results, property answers, review summaries, and partner replies through the relevant Booking.com experience.
Solution architecture
5 components · 2 layers - Compute
- GPT models Interpret natural-language intent and generate itinerary, discovery, and recommendation outputs from Booking.com's data.
- GPT-4o mini Powers Smart Filters and related natural-language understanding tasks for search refinement and review analysis.
- Serving
- AI Trip Planner Handles open-ended destination discovery and itinerary building from conversational traveler prompts.
- Property Q&A Answers property-specific traveler questions using user-generated content and property descriptions.
- Help Me Reply Generates automated partner replies and templates for common guest communications.
- AI Review Summaries condense large review sets into key themes
- OpenAI models were integrated through Booking.com's existing APIs and data infrastructure
- Property Q&A is fine-tuned on user-generated content and property descriptions
- Smart Filters uses GPT-4o mini to interpret natural-language search intent
- The AI Trip Planner combines structured data with unstructured reviews and descriptions
"OpenAI's models were integrated through Booking.com's existing APIs and data infrastructure, allowing teams to rapidly test and iterate on new features."
"The first prototype, capable of destination discovery and itinerary building, was launched in just 10 weeks."
"The team integrated OpenAI's GPT models with Booking.com's proprietary data on properties, pricing, and availability."
"Uses GPT-4o mini to understand natural language prompts like 'sunset views' or 'great gym'."
"We'd spent years fine-tuning our structured data, like pricing, availability, cancellation policies. But now we could layer in unstructured data, like user reviews, natural language descriptions, and generate curated suggestions based on both."
