Data-Driven Data Governance and Asset Health Scoring at Xiaomi

In this presentation, Xiaomi shares a data‑driven approach to efficient data governance, emphasizing the importance of managing growing data volumes to improve quality, value, security, and compliance.

The talk outlines a three‑level "rocket" governance plan: Level 1 builds baseline capabilities and a metadata warehouse; Level 2 develops product tools for comprehensive governance, including data quality monitoring, security classification, cost control, and development standards; Level 3 introduces an asset health scoring model that quantifies data health across storage, compute, quality, security, and compliance dimensions.

The asset health model assigns a 100‑point score to each table, aggregates scores to higher‑level entities, and calculates sub‑scores for storage (including hot‑warm‑cold tiering and lifecycle management), compute (identifying skew, failures, duplication, and resource consumption), quality (content and generation quality with monitoring rules), security (classification, grading, and field‑level controls), and compliance (metadata completeness, naming conventions, and documentation). The overall health score is a weighted sum of these sub‑scores.

Productization of the model results in a unified governance platform that displays overall and detailed health scores, highlights problematic assets, and suggests remediation actions. Future plans aim to enrich health features, incorporate real‑time data, build an asset graph for value scoring, enable proactive governance, and provide unified data services for various application scenarios.

The Q&A addresses concerns such as avoiding notification overload, determining table access and compliance, and details on data‑application governance through a semantic data service layer that standardizes data access across OLAP, KV, and other use cases.

The session concludes with thanks to the audience.