How Xiaomi Built a Scalable Metric System: Best Practices and Methodology

Metric System Overview

Metrics are quantitative measures derived from subdivided business units, used to describe, measure, and evaluate performance; a metric system combines these metrics with multiple dimensions to provide a comprehensive view.

Business Pain Points and Solutions

Three major challenges are identified: (1) Business view – multiple data sources and inconsistent metric definitions require repeated validation; (2) Technical view – confusing metric naming, duplicate metric production, and inconsistent consumption across systems; (3) Product view – lack of an integrated solution that connects business systems, data warehouses, metric management, and BI visualization.

Metric System Construction Method

Xiaomi addresses the above pain points with two core methodologies: the OSM model and the MECE principle.

OSM Model

The Object‑Strategy‑Measure (OSM) model links business objectives (O), strategies (S), and measures (M). For example, improving inventory turnover efficiency (objective) leads to strategies such as clearing old stock and minimizing inventory, which are measured by metrics like age‑of‑stock days and days‑of‑stock (DOS).

MECE Principle

MECE (Mutually Exclusive, Collectively Exhaustive) ensures categories are non‑overlapping and cover all possibilities. Methods include binary division, process‑based division, element‑based division, formula‑based division, and matrix‑based division.

Xiaomi Best Practices

By applying OSM and MECE, Xiaomi has established a group‑wide data warehouse and a unified metric system.

Implementation Path

The construction process consists of four stages:

Model design – use MECE to split data domains and build core models.

Data‑warehouse construction – follow a unified architecture and standards to create a stable, secure group warehouse.

Metric management – create a unified data dictionary to ensure consistent definitions.

Data application – build a semantic model and “Data Encyclopedia” product, enabling self‑service BI dashboards and rapid data consumption.

Model design follows seven principles:

High cohesion and low coupling.

Separate core and extension models.

Centralize common processing logic.

Balance cost and performance.

Data rollback capability.

Consistency of field names across tables.

Clear, understandable naming.

Summary and Outlook

The metric system links data development → metric management → data analysis, providing full lineage visibility. The methodology has been rolled out to 55 data domains and 520 core atomic metrics within Xiaomi, and will continue to be iterated and expanded.