YY Live Business Metric Governance Practice

The speaker, Deng Qinfeng, data product lead at YY Live, introduces a sharing session on the practice of business metric governance from a data product perspective.

Problem background includes three typical scenarios: inconsistent report definitions, difficulty locating required data, and long lead times for new report requests, all of which cause confusion and inefficiency for product, operations, analysts, and engineers.

Analysis of these scenarios reveals three main issue layers: lack of unified metric management leading to duplicate or ambiguous definitions, uncontrolled proliferation of reports making retrieval hard, and inconsistent metric calculations across data models.

The governance goals are to provide a standardized, searchable metric system, improve delivery speed, and reduce uncontrolled resource consumption, thereby enhancing user experience while increasing development efficiency.

Collaboration involves three key roles: the data product team as owners of metric definition and standards, the data development team responsible for applying these standards in data models and reports, and business users who consume the metrics and provide feedback.

Implementation steps focus on (1) defining consistent naming and descriptions, (2) building hierarchical classification indexes for quick search, (3) documenting metric dependencies and lineage, and (4) establishing a lifecycle process of registration before usage.

Product features built to support the governance include an indicator management UI for naming, classification, and lineage, as well as a self‑service analysis application that lets users select metrics and dimensions, automatically generates SQL, and visualizes results without developer involvement.

Measured outcomes show significant improvements: faster data retrieval and self‑service analysis, reduced development cycles from days to hours, and lowered storage/computation waste by decommissioning redundant assets.

Future evolution aims to integrate large language models as a data Copilot, leveraging a knowledge base of metric definitions to enable natural‑language query, automatic SQL generation, chart recommendation, and continuous feedback loops for iterative improvement.