How Meituan Scaled Data Governance: Practical Lessons for Enterprise Data Management

01 Background Introduction

Why implement data governance? The author believes that throughout the data lifecycle—generation, collection, processing, storage, usage, and destruction—various problems can arise. In early stages these issues have little impact, but as business grows and data volume and quality requirements increase, many problems become evident and need systematic governance.

1. Issues to Govern

The data governance process must address five major problem categories: quality, cost, security, standardization, and efficiency.

02 Governance Practice

2. Meituan Data Current State

The main problems are lack of standardized specifications, numerous data quality issues, rapidly growing costs, insufficient data security controls, and low efficiency in data management and operations.

3. Governance Implementation

Data governance is divided into four major parts: organization, standards, technology, and measurement metrics. The overall implementation path relies on standardized norms and organizational guarantees, supported by a technical system that ensures the governance strategy is realized. A measurement system continuously monitors governance effectiveness to ensure long‑term improvement.

3.1 Standardization and Organizational Assurance

The management committee is a virtual organization composed of technical and business departments. The technical side handles data development, while the business side manages data products. Both act as responsible parties, coordinating technical and business teams.

3.2 Technical System

Data quality is the most critical issue; four major problems are identified:

Data warehouse lacks comprehensiveness and relies on personal interpretation of guidelines.

Data consistency issues, especially in metric management, due to undocumented and non‑systematic logic.

Numerous data applications (table sync, message push, OLAP queries) cannot guarantee consistency across endpoints.

Multiple product entry points (over ten) without unified control lead to divergent usage.

3.3 Data Warehouse Modeling Standards

Standard documents are provided in advance, and many standardization items are enforced through configuration constraints. Post‑deployment checks verify compliance, prompting timely corrections.

3.4 Unified Metric Management System

The system standardizes process management, metric definitions, and metric usage.

3.5 Unified Data Service

A unified data service platform aims to improve efficiency, accuracy, monitoring, and to connect the entire data warehouse and application chain. It offers two consumption modes: on‑demand B‑end calls (tens of thousands per day) and C‑end push (daily updates).

3.6 Unified User Product Entry

Through standardized data modeling and metric logic, three major data marts are kept consistent, ensuring overall data consistency.

Overall System Architecture

The technical architecture consists of three layers: unified data modeling, unified metric logic, unified data service, and unified product entry, collectively ensuring data quality.

Data Operation Efficiency

Operational data issues are addressed with a systematic data guide covering metrics, warehouse models, and recommended usage.

Data Cost

Meituan's data cost grows rapidly each year, roughly 70% computation, 20% storage, and 10% log collection.

Data Security

Security is handled through pre‑prevention, in‑process monitoring, and post‑incident tracing, following five principles: ciphertext handling, latest decryption, minimal extraction scope, minimal authorization, and full‑process auditing.

3. Measurement Metrics

A comprehensive data governance measurement system has been established, covering five categories: quality, cost, security, usability, and value.

03 Future Planning

Data governance evolves through three stages: passive governance (ad‑hoc, project‑based), proactive governance (long‑term planning, systematic processes), and automated governance (intelligent, policy‑driven automation based on metadata and continuous monitoring).