How NetEase Yanxuan Built a Scalable Data Product System: Lessons & Practices

Background

The author’s written summary originates from a 2020 product‑manager conference talk titled “NetEase Yanxuan Data Product Practice and Methodology.” Over three years, driven by user demand, the team built a data‑product ecosystem that evolved through four overlapping stages.

1. Business‑Data Visibility

In 2017 the team faced three classic problems: many business units, a small data team, and inadequate tools. To meet the surge in reporting requests, they built a data warehouse on the NetEase Mammoth platform and deployed a private, agile BI solution called “Yanxuan YouShu.” Data engineers used SQL to model the warehouse, analysts built data marts, and the BI tool enabled drag‑and‑drop report creation. The platform now hosts over 80,000 charts with daily PV > 6,000 and weekly UV > 900.

Performance bottlenecks caused by massive concurrent access were mitigated through a two‑step caching strategy: first, a timed active cache at 7 am, then a data‑driven cache that pre‑loads charts as soon as their source tables are materialised, raising cache‑hit rates above 80 % and delivering sub‑second response times.

2. Data‑Quality Assurance

Multiple data sources, long pipelines, and ambiguous metric definitions led to three quality issues: latency, errors, and inconsistency. The team introduced a suite of middle‑platform products to guarantee completeness, accuracy, and stability:

All business DBs and logs are ingested into the warehouse, with a supplemental Excel‑based data‑entry system for non‑digitised processes.

Metric definitions are managed in an indicator‑management system; data engineers design warehouse tables based on these definitions, ensuring consistent metric semantics.

Extensive validation rules are applied at both table‑level and field‑level during Excel imports.

Event‑tracking data quality is enforced via an event‑management system that provides definition, workflow, unit‑test, and regression‑test capabilities.

A joint task‑operations center with Hangzhou Research Institute offers task governance, intelligent monitoring, and impact analysis to keep the >10,000‑task pipeline stable.

3. CXO‑Level Data Access

Although the BI platform contained abundant reports, senior executives (CXOs) struggled to locate relevant dashboards quickly. To address this, the team built “VIPApp,” a mobile data workbench embedded in the Yanxuan app. VIPApp provides on‑the‑go KPI monitoring and direct links to product, category, and traffic data, turning the “see‑and‑get” interaction into a seamless experience for executives and eventually for all employees.

4. Scenario‑Based Data Products

Beyond generic reporting, the team delivered vertical, scenario‑specific solutions:

E‑commerce Big‑Screen : Real‑time visual dashboards for major sales events (e.g., Double 11) with animated, high‑impact UI.

Customer‑Service Dashboard : Real‑time queue, agent, satisfaction, and alert metrics, used continuously by the support team.

Sentiment Insight Center (DiTing) : Converts unstructured user comments and service chats into quantifiable sentiment categories, linking them to business units for targeted UX improvements.

Promotion Channel Management (XingTian) : Tracks acquisition KPIs and ROI across marketing channels.

Conclusion & Reflections

After more than three years of iterative development across the four stages, NetEase Yanxuan has established a comprehensive data‑product system that supports large‑scale reporting, high‑quality data management, and scenario‑driven analytics. The experience demonstrates that combining a robust data warehouse, agile BI tooling, proactive caching, rigorous quality controls, and mobile‑first access can deliver enterprise‑level data products that scale with business growth.