AI Lessons From China: Fliggy’s Path to the Omni-Intelligent Travel Agent
Summary
Dr. Alex Chen, CTO of Fliggy (Alibaba's online travel platform), delivered a data-rich presentation at ITB Berlin outlining Fliggy's transformation from a traditional OTA into what he calls an 'omni-intelligent travel agent' powered by a multi-agent AI architecture. Fliggy currently serves over 4 million daily active users, has 500 million cumulative members, and covers more than 25 non-stop flight routes and 90% of international hotel groups. Over 60% of active users were born after 1990, making them AI-native consumers whose behavior differs meaningfully from traditional search-based users.
The centrepiece of the presentation was Fliggy's integration with Qwen (Alibaba's flagship large language model), which serves as an agentic AI gateway across Alibaba's entire ecosystem — including Taobao (e-commerce), Alipay (payments), Amap (mapping), DingTalk (enterprise), Lazada, and AliExpress. During the 2025 Chinese Lunar New Year, this AI-native gateway processed 12 million orders in just six days — a striking demonstration of transactional AI at scale. A separate Chinese New Year experiment saw over 100,000 flights and tickets booked natively within a chat interface, with one-in-four detail-page viewers converting — on par with or exceeding traditional search-based conversion.
Dr. Chen presented three live demos showing Qwen's multi-agent travel capabilities: (1) a flight search from Beijing to Berlin that surfaces the sole non-stop option and three layover alternatives ranked by value and travel time; (2) a hotel search using natural-language input ('hotel with Eiffel Tower view for family with children') that assembled a specialised AI team — hotel advisor, loyalty specialist, service expert — recognising the user's AOR Platinum membership and delivering three bookable options with pricing comparisons; and (3) an in-trip destination guide for Berlin triggered by a voice prompt.
He proposed a framework of 'Three Es' to organise Fliggy's AI strategy: (1) AI for Efficiency — 85% of Fliggy developers use AI co-pilots; 6 million hotel listings auto-matched daily; 1.2 million images optimised by vision models; 4 million overseas room descriptions auto-translated; AI customer service now handles 19% of all chat and phone inquiries with a resolution rate exceeding that of human agents; (2) AI for Experience — hyper-personalised, multi-turn, multi-agent customer journeys; (3) AI for Supply — open APIs enabling hotels, airlines, and offline agencies to surface their inventory through the AI marketplace.
Dr. Chen then outlined six strategic perspectives for the global travel industry: (1) AI brings high-value, high-retention new traffic via content planning, research, and cross-category inspiration — traffic that barely existed in search-based UX; (2) leisure and inspirational travel products are the biggest beneficiaries, citing a bundled 'ice hotel + animal kingdom + breakfast' SKU that would never surface in keyword search but converts naturally in chat; (3) AI enables a genuine one-stop all-inclusive travel experience — referencing Booking.com CMO data showing 30%+ of orders and 10%+ of GMV are already multi-category; (4) AI breaks the 'large-scale, low-cost, personalised' trilemma by allowing users to express complex needs in 20–30 tokens across multi-turn conversations; (5) supply chain owners and service providers are 'halo assets' (borrowing Goldman Sachs terminology: heavy assets with low obsolescence) and should focus on being AI-agent-friendly rather than B2B/B2C-centric; (6) competitive advantage will come from private data, local/niche knowledge, long-tail expertise, and digitised service DNA — even Gemini 2.5 Flash achieves only a 23% satisfaction rate on specific travel tasks because it lacks local data.
On the broader industry outlook, Dr. Chen stated the transactional AI shift is currently in single digits (5–6% of bookings via AI chatbot) but predicted a rapid, potentially monthly inflection once product-market fit is found — and warned the industry has 'less than one year' to prepare. He noted AI-guided conversion rates are already 6–9% higher than search-based equivalents, with AI-native orders in niche categories reaching 20% conversion. He closed with a call-to-action for all value-chain participants — hotels, airlines, OTAs, tech providers — to reposition around supply/agent roles or risk disintermediation.
Key Takeaways
- •During the 2025 Chinese Lunar New Year, Fliggy's Qwen AI gateway processed 12 million orders in six days — entirely AI-natively — demonstrating that transactional AI at scale is already live in China, not theoretical.
- •A separate Chinese New Year chat experiment generated over 100,000 flight and ticket bookings natively in-chat, with a 1-in-4 detail-page conversion rate, on par with or exceeding traditional search-based conversion benchmarks.
- •Fliggy's AI customer service now fully handles 19% of all chat and phone inquiries, and the resolution rate for these AI-handled cases already surpasses that of human agents — a significant operational milestone.
- •85% of Fliggy's developers use AI co-pilots for coding and development. On the supply side, 6 million hotel listings are auto-matched daily by AI agents, 1.2 million images are optimised by vision/VR models, and 4 million overseas room descriptions are auto-translated — illustrating how AI for internal efficiency precedes AI for consumer experience.
- •Dr. Chen stated global transactional AI adoption is currently 'single digits — probably 5 to 6%' of bookings, but predicted the shift will accelerate on a 'monthly or daily basis' once product-market fit is established, and warned the industry has 'less than one year' to prepare.
- •AI-guided conversion rates on Fliggy are 6–9% higher than search-based equivalents and still growing. AI-native orders in niche/SKU categories (e.g., train tickets, leisure spot tickets) are reaching 20% conversion — beating traditional search-based apps.
- •Even frontier models like Gemini 2.5 Flash achieve only a 23% satisfaction rate on specific travel tasks because they lack local, proprietary, and long-tail data — underlining Dr. Chen's argument that private data and niche knowledge are the true moats in AI travel.
- •The 'Eiffel Tower hotel' demo illustrated a core AI advantage: a family request to book 'a hotel where we can see the Eiffel Tower day and night' cannot be fulfilled by keyword search (no such field exists), but AI agents scour UGC and descriptive data to match it surgically, including recognising the user's AOR Platinum status to prioritise relevant loyalty options.
Transcript
Who who of you guys had the honor or the opportunity to to book or search for flights on a Chinese OTAA? Are that that Oh, yeah. Okay. Of course. Yes. Okay. We have a few Chinese audience here, but I assume some of you are curious how this is going on, especially when OTAA turns into an omni channel or omni intelligent travel agent. That's what Fleiggy is going used to be one of the leading OTAAS by the way globally and we have the great honor to have here the chief technology officer of Frigy w...
