10 key takeaways from a presentation at WiT Japan by Amy Wei, Senior Product Director, Trip.com Group
1. The shift is already here and accelerating fast.
Trip.com launched its AI assistant TripGenie in February 2023, just 2.5 months after ChatGPT. Three years of real user data tell a clear story: AI-assisted bookings are up significantly year on year, real-time tool usage has grown 3,100% year on year, and the share of users asking booking-related questions (as opposed to general questions) has shifted from under 50% to over 60%. Adoption is not coming.
2. People are using AI to reduce uncertainty, not to get more options.
The primary job AI is doing in travel is not inspiration or discovery – it is confidence-building. Users want to understand what they are actually buying: fare conditions, baggage rules, cancellation policies, lounge access. Over two-thirds of pre-booking flight queries are about lounges and conditions, not prices. A good AI does not win by providing more options. It wins by helping users decide with less uncertainty.
3. AI has moved from the planning stage into the trip itself.
The assumption that AI is primarily a planning tool is being overtaken by reality. Real-time tools – menu translation, local navigation, medical assistance, live translation – are seeing growth. Travellers are using AI in micro-moments throughout the journey: where to eat in this neighbourhood right now, how to read a handwritten Japanese menu, what to do when something goes wrong mid-trip. The AI companion, not just the AI planner, is emerging.
4. Grounded answers beat fluent ones every time.
The biggest lesson from three years of TripGenie: accuracy matters more than eloquence. An answer that confidently states the wrong cancellation policy is worse than no answer at all. In travel, hallucination is not a minor inconvenience – it costs real money and destroys real trips. AI must be connected to real inventory, real prices and real-time policy data. Fluency without grounding erodes trust rather than building it.
5. Decision compression is more valuable than information delivery.
The hotel comparison feature that reduced the average browsing session from multiple nights to fewer than two is not primarily a conversion tool, it is a decision compression tool. The metric that matters is not how much information AI delivers but how quickly it helps a user reach a confident decision. Less friction in the decision process, not more data, is the value proposition.
6. Agentic AI was triggered by a conversion data point, not a strategy decision.
The move to agentic AI – AI that takes actions, not just answers questions – was not a dramatic strategic pivot. It was driven by a specific data observation: users who asked eSIM-related questions had a 6.8% conversion rate within 48 hours. They already knew what they wanted. The gap was execution, not information. When you can see that clearly in the data, building an agent to close that gap is the obvious next step.
7. More automation does not mean less accountability.
This was Wei’s most emphatic principle. As agentic AI completes bookings within a conversation, the question of who is responsible when something goes wrong becomes urgent. Her answer: accountability does not change, it just needs to be made more explicit. The AI must surface product details, restrictions, prices and potential consequences clearly before payment. The interface changes; the responsibility does not. “More automation does not mean less accountability. We just make it clearer.”
8. China’s AI advantage is ecosystem depth, not model power.
What makes China’s AI travel innovation move faster is not superior models — it is the completeness of the surrounding ecosystem. Integrated payments, mobility, local services, food delivery, all operating in the same digital environment, allow AI to function not just as an information layer but as a transaction and service layer. The feedback loop is faster because the entire journey — not just the booking — is digital and connected.
9. Legacy policy content must be restructured for AI to work.
The lesson Trip.com learned is that human-readable policy text does not translate cleanly into AI-executable logic. Long-form articles and rules work for human customer service agents who can infer context. AI needs structured workflows: if A happens, then B is next. Restructuring legacy content into decision trees and process graphs is unglamorous but essential work for any travel company deploying AI seriously.
10. TripGenie will never travel for you and that is the point.
The one thing AI will never do is have the experience on your behalf. It can solve the logistics, compress the decisions, translate the menu, find the eSIM, rebook the disrupted flight. What it cannot do is feel the moment of being somewhere extraordinary. Wei said, “AI is very good at solving the logistic issue. But it will never help you experience the moment that you are there.” The technology disappears so that the experience can emerge, which is, ultimately, why travel still matters.
