Understanding Ctrip Flight Ticket Tracking System (UBT) and Its Key Metrics

Author Background Li Ning, senior data & product manager at Ctrip Flight Tickets, shares extensive experience in user behavior analysis, user profiling, and A/B testing.

Why Tracking Matters If event tracking (埋点) is ambiguous, data analysis foundations become unstable, leading to unreliable conclusions. Analysts must leverage existing tracking data effectively rather than waiting for perfect tracking.

1. What is UBT? UBT (User Behavior Tracking system) was initiated by Ctrip's former CTO Eric Ye. It started with online event collection and expanded to mobile apps, hybrid, H5, and later added AB testing. UBT defines data formats, contracts, storage, ETL, and reporting for Ctrip's analytics.

2. Client‑Side Tracking Ctrip Flight Tickets uses five major client‑side tracking categories: ctm, action, trace, pv, and server‑side tracking. Over time, these have been added cumulatively, leading to some overlap and redundancy, which is periodically cleaned when app size reduction is needed.

2.1 ctm Tracking Similar to Google Analytics UTM, ctm records page source via GET parameters and is used for online and H5 platforms. It helps evaluate page UV and conversion but only passes data once, limiting direct order association.

2.2 pv Tracking PV tracking is the longest‑standing and simplest method (logpage with pageid). It serves as a baseline for validating new tracking data and is used across native, hybrid, and RN pages.

2.3 Action (Click) Tracking Click events are named with a "c_" prefix (e.g., c_search). The Hive table stores pageid, and developers must avoid duplicate names on the same page. Click tracking is essential for detailed behavior analysis.

2.4 Trace Tracking Trace events capture all business information on main‑flow pages (home, list, detail, fill, complete) with a single event per page, simplifying conversion calculations.

2.5 Server‑Side Tracking Server‑side events record policy tags and other display information from list and intermediate pages, enabling analysis of policy impact on sales and future machine‑learning‑driven optimization.

3. Metric Understanding Key metrics include UV, visits, PV, exit count, and page stay time. UV can be defined by device ID (vid/clientcode) or user ID (uid) depending on the analysis dimension. Stay time is calculated as the median difference between start times of consecutive pages.

4. Data Association and Loss Associating behavior with orders relies on device IDs and temporary order tables. Different tracking types have varying loss rates; PV is the most stable, while trace and server‑side tracking have higher accuracy but may suffer from format issues.

5. Common Tracking Issues Issues include unintended triggers, mutual suppression of events, mismatched logic between developers and analysts, and framework constraints that cause data loss.

6. New vs. Old Users New users are defined by device history (no prior visits) or order history (no prior orders). UV differences arise from device‑based vs. order‑based counting, and platform distinctions affect user classification.

7. Retention, Repurchase, and Stay Time Quarterly repurchase rate and monthly revisit rate are used for flight tickets. Stay time is an auxiliary metric; its interpretation depends on conversion outcomes and must be considered alongside core metrics.

8. Behavior Flow Visualization Building user behavior flow tables enables PMs to trace actions by UID or phone number, facilitating problem discovery and hypothesis testing with SQL.

9. Summary Principles 1) Interaction tracking should be placed on the front‑end as it is closest to the user. 2) Group similar concepts to avoid confusion (e.g., exit vs. bounce rate, UV vs. PV, repurchase vs. revisit).

Appendix Definitions of identifiers: vid/clientcode/clientid (device IDs), sid (session ID), pvid (page view sequence), starttime (event timestamp), UV (device‑based unique visitor count).