How Data-Driven Metrics Transform Product Analytics and Decision-Making

01 End‑to‑Start Metric System

Metric systems consist of basic indicators and derived dimensions. Two classic examples are shown: an e‑commerce app that breaks down revenue into order value and customer count, and a content app that starts from DAU and splits into channel and user cohorts. Metrics provide a hierarchical, structured view that enables quantification and evaluation of business performance.

Metrics help diagnose problems, set priorities, and guide solutions, making them indispensable for business management.

02 Business‑Driven and Data‑Driven

Beyond business goals, data should drive metric construction. First, define a long‑term North Star goal (strategic). Then decompose it into sub‑metrics using frameworks such as OSM (Objective‑Strategy‑Metric). Real‑world cases show that simple UI tweaks can increase raw metrics (e.g., average stay time) without improving user experience, highlighting the need for data‑driven signals.

Data‑driven methods include statistical analysis, machine‑learning (Shapley values, decision trees, clustering), and causal analysis (Uplift, DML) to select high‑impact signals and dimensions.

03 Data‑Driven Metric System Implementation

Case 1: Video app auto‑play increased raw play counts but polluted the metric. By analyzing play‑time distribution and experiments, a threshold was set to filter out accidental plays, yielding a cleaner metric.

Case 2: Using causal inference (Uplift meta‑learning) to determine the playback completion rate threshold that most improves user retention. Experiments confirmed that data‑driven metrics have higher directionality and sensitivity than raw counts.

04 Data‑Driven Metric System Application

The metric system supports three practical steps: alert (detect anomalies via statistical rules or predictive models like XGBoost), diagnose (root‑cause analysis using multi‑dimensional breakdowns and Gini coefficients), and strategy (targeted actions on audience, scenario, or traffic). A full‑chain “in‑play intervention” workflow is illustrated, from potential identification, prediction, diagnosis, to targeted audience outreach, enabling timely, data‑driven decisions.