Booking.com Used GPT to Crack the Discovery Problem in 10 Weeks

Booking.com CTO Rob Francis hit on the core problem with a perfect example: you can’t filter for “heart-shaped beds or Elvis impersonators” when you’re hunting for a cheesy romantic getaway. That’s the discovery gap the travel giant had been trying to bridge for years, and it’s what made Adrienne Enggist, Senior Director of Product Marketplace, feel a “tingle” when ChatGPT launched in 2022. She compared it to the dawn of broadband — a fundamental shift in how people could engage with travel. The company’s existing machine learning was great at the “last mile” of getting you from search to booking, but it was useless when you didn’t know exactly what you wanted.

The speed of execution here is genuinely unusual for a company of this scale. VP of Product Marketplace Joe Futty described running a hackathon with OpenAI’s API and shipping the AI Trip Planner prototype in just 10 weeks. That’s a blistering pace for integrating large language models with a proprietary data moat that includes decades of structured pricing, availability, and cancellation policies. The real magic was layering that hard data with unstructured information — user reviews, natural language descriptions — to generate suggestions that feel curated rather than just filtered. A traveler asking “Where should I go for a romantic weekend in Europe?” now gets an answer that understands intent, not just keywords.

The early metrics, though still maturing, point to a stickier product. Users are spending more time on the platform exploring personalized itineraries, Smart Filters have slashed search friction, and the Property Q&A feature has already reduced customer support contacts by serving accurate, in-app answers. Review summarization is also speeding up the decision-to-book timeline. These aren’t flashy, futuristic demos; they’re practical integrations that directly address the anxiety and paralysis that often accompany trip planning.

What’s most interesting, however, is the downstream effect Enggist highlighted about overtourism. Booking.com’s data shows the top 15 European destinations are heavily over-touristed. By using AI to surface hundreds of equally compelling but less trampled destinations, the platform isn’t just solving a user experience problem — it’s quietly nudging the economics of the entire travel industry toward a more distributed model. That’s a second-order effect no rule-based system could ever have produced.