Ctrip CTO Gan Quan on Building a Data‑Driven Personalized Recommendation System

In the era of consumption upgrade, time has become a new battlefield and currency; Ctrip focuses on personalized services to optimize user time rather than creating addictive experiences.

Since 2015, Ctrip has used cross‑screen data processing, real‑time APIs, and data‑model training platforms to predict user behavior and deliver personalized pre‑travel recommendations via web pages and EDM.

Personalization now extends beyond pre‑travel suggestions to real‑time ads and product pushes across all terminals, such as displaying destination‑related ads after a search or suggesting similar hotels after an abandoned booking.

CTO Gan Quan emphasizes that personalized recommendations should be intelligent and convenient, allowing users to find needed products with fewer clicks and less search effort.

From 2015 onward, Ctrip has continuously expanded its big‑data platform, integrating internal business line data and third‑party sources to build a rich tourism knowledge base and data center.

Algorithmically, Ctrip has broadened its deep‑learning toolbox, iterating models based on user and data performance, and has connected over 60 business lines and external advertising channels.

The recommendation engine enables cross‑domain suggestions, such as recommending hotels, transportation, and activities after a flight search, using multi‑dimensional data to generate relevant offers.

AB testing across various channels (home page slots, search, discount flight listings) helps identify the most effective recommendation patterns for different user segments.

A small data team of fewer than 20 people built the personalization platform, which contributed an additional 120 million CNY in revenue in 2016 with a high ROI, and further boosted specific business lines by up to 6% with minimal engineering effort.

Challenges include integrating heterogeneous external data, normalizing it for internal use, and maintaining a second‑level data loop that instantly reflects user actions across the system.

Ctrip’s “Discovery” channel combines user interest points, historical data, and contextual factors (price, season, weather, etc.) to deliver personalized destination recommendations and comprehensive travel plans, aiming to shorten the average decision cycle of 20‑40 days.

Technical and business teams collaborate through shared metrics and a big‑data committee, fostering alignment and rapid iteration to continuously improve the recommendation system.

Gan Quan concludes that success in personalized recommendation requires robust data accumulation, adaptable system architecture, effective algorithms, and real‑world business validation.