How Booking.com builds hybrid AI agents for traveler support

A modular, layered design that improves accuracy without relying on a single large AI system — with useful insights for hotels adopting AI

Booking.com has developed agentic AI systems in a careful, step-by-step way instead of adopting one large, expensive model. The company mixes small, travel-specific AI models for fast answers with larger models when deeper reasoning or accuracy is needed.

According to the company, this modular setup has already doubled accuracy in booking-related retrieval, ranking and service interactions, and it allows more customer conversations to be automated without losing quality.

Booking.com is also using AI to personalize search experiences at scale while being careful about privacy and memory, so customers receive useful help without feeling monitored.

Key takeaways

• Smarter service automation: Booking.com doubled topic detection accuracy, which automates more routine inquiries and frees human agents to handle complex, guest-critical situations.
• Better guest personalization: Travelers can describe what they want in free text (e.g., “room with a hot tub”), and the platform instantly applies personalized filters instead of forcing customers to search manually.
• Higher loyalty through service: Faster and more accurate handling of customer issues supports greater guest satisfaction and long-term loyalty — directly benefiting conversion and repeat bookings.
• Privacy and consent first: Booking.com is cautious about storing long-term guest preferences, ensuring personalization does not feel intrusive or undermine trust.
• Fast where speed matters: Simple recommendation or search tasks use small, fast models, ensuring travelers don’t wait for answers, while slower, more accurate models are used only when precision is critical.
• Avoiding costly technology dead-ends: Booking.com keeps its AI agents flexible and avoids architectural “one-way doors,” so it can improve or replace parts of the system without rebuilding everything.
• Practical lesson for hotels: Start small with AI use cases, use ready-made APIs where possible, automate obvious pain points first, and only build custom tools when scale, quality or brand experience truly require it.

Get the full story at VentureBeat