Design and Implementation of a Full‑Chain Marketing Data Product at NetEase Yanxuan

NetEase Yanxuan is a self‑operated e‑commerce brand covering the entire product lifecycle from R&D to delivery, which creates a strong need for data‑driven decision making.

The marketing landscape involves multiple channels and agencies, strict ROI‑based assessment, and dynamic adjustments as the business evolves from user acquisition to long‑term value.

Based on these characteristics, Yanxuan defined four core data product requirements: security (protecting user data and preventing fraud), high timeliness, comprehensive business coverage, and complete end‑to‑end linkage.

Full‑Chain Data Product Design

Security and high timeliness: integrated with 30+ media sources, built activation and order attribution, and risk monitoring.

Comprehensive scenario coverage: established a rich metric evaluation system and analysis framework covering external channel factors and internal e‑commerce dimensions (people, goods, place).

Complete business linkage: ensured connection from media‑side ad delivery to Yanxuan's backend conversion, and covered the entire internal workflow from project initiation to settlement.

The core of the product consists of three systems: an attribution system, a metric evaluation system, and an analysis system.

Attribution System

Yanxuan distinguishes between APP and non‑APP users, tracks cross‑device behavior, and adopts a last‑click priority model. Attribution windows are shorter for APP interactions and longer for non‑APP to reflect user stickiness.

Metric Evaluation System

Metrics are derived from strategic goals and tactical execution, split into scale (e.g., payment users), efficiency (e.g., conversion cycle), and quality (e.g., repeat purchase, user health). Targets are cross‑validated and iteratively refined.

Analysis System

The analysis framework progresses from descriptive (what happened) to diagnostic (why it happened), predictive (what will happen), and prescriptive (how to act). Diagnostic analysis drills down to channel‑level anomalies, while predictive analysis uses historical user data to forecast long‑term ROI.

Practical applications include pre‑project planning, online link generation, real‑time intervention during campaigns, and post‑campaign cost management with risk‑adjusted settlement recommendations.

Scenario‑Based Recommendations

The design follows five layers: strategic (define business goals), scope (establish metric taxonomy), structure (identify core impact factors), framework (monitor‑diagnose‑decide workflow), and presentation (deliver insights to end users).

Key advice includes deep business immersion when building data products, extensive stakeholder education, and targeting senior decision‑makers for faster adoption.

Q&A Highlights

Diagnostic analysis uses a generic framework; personalized algorithms appear in predictive analysis.

Yanxuan offers various data products beyond marketing, such as product‑side and supply‑chain analytics.

Markov models are not used for attribution due to relatively simple conversion paths; a time‑decay weighting approach is preferred.

The session concluded with thanks and a reminder to like, share, and follow the DataFunTalk community.