Data Governance Practices at Meituan Hotel & Travel Platform

Data has become a core asset for many companies, and during data development various quality, efficiency, and security issues arise. Data governance aims to continuously eliminate these problems, ensuring data accuracy, completeness, and security, thereby creating business value while strictly managing data permissions to avoid leakage risks. This article introduces Meituan’s hotel‑travel platform’s data‑governance practice.

1. Background

1.1 Why Data Governance

With the rise of mobile internet, offline business activities have moved online, dramatically increasing data generation speed. Companies recognize data as a core asset that drives business. Data governance has become a hot topic, especially for large internet companies, because every stage—from data generation, collection, processing, storage, usage to destruction—can introduce quality, efficiency, or security problems. Early in a company’s growth, these issues are tolerable, but as business scales, higher data quality and stability are required, revealing many governance needs.

1.2 Issues to Govern

Quality problems (timeliness, accuracy, consistency of metrics)

Cost problems (rapid data growth leading to high infrastructure costs)

Efficiency problems (manual, blind labor in data development and management)

Security problems (risk of user‑data leakage)

Standard problems (inconsistent data standards across business units)

1.3 Meituan Hotel‑Travel Data Status

Since 2014 the hotel‑travel business became an independent department; by 2018 it was a major online reservation platform. Data volume grew faster than business, with production tasks doubling each year. By 2019 the following problems were observed:

Severe data‑quality issues (redundancy, inconsistent metric definitions)

Rapid cost growth (big‑data storage and compute >35% of expenses)

Low operational efficiency (engineers spending 5‑10% of time on ad‑hoc queries)

Insufficient data‑security controls (no unified permission standards)

Lack of development standards ("silo" development, duplicated data)

1.4 Governance Goals

Establish full‑link data‑development standards to improve quality and metric consistency.

Control big‑data cost by reducing redundancy, archiving cold data, and optimizing resource usage.

Manage data‑usage security with approval processes and usage standards.

Improve engineer development and operation efficiency through automation.

2. Data Governance Practice

2.1 Governance Strategy

The governance covers the entire data lifecycle and is divided into four major parts: organization, standards, technology, and measurement. Standardization and organizational guarantees are prerequisites; technology systems implement the strategy, and a measurement framework monitors effectiveness.

2.2 Standardization and Organizational Guarantees

We defined a full‑link data standard covering collection, warehouse development, metric management, and lifecycle management, and established a Data Management Committee with business and technical teams to enforce these standards.

2.1 Standardization : includes standard creation, execution, and organizational support to ensure alignment across data‑tech, business, and analysis teams.

2.2 Organizational Support : a functional governance organization (Data Management Committee) coordinates business data product groups and technical data development groups to drive standards, security, and lifecycle management.

2.3 Technical Systems

To improve quality, cost, security, and operational efficiency, we built several unified systems:

One Model : unified warehouse modeling to enforce standards and eliminate redundant/obsolete tables.

One Logic : centralized metric definition and management, ensuring consistent metric logic across products.

One Service : unified data‑service layer decoupling logic from APIs, enabling monitoring and stable service.

One Portal : unified product entry for different user groups (decision analysis, sales, data asset lookup).

Data Quality : We introduced a unified modeling system, indicator logic management, and rule‑based validation (Dataman) to enforce standards, generate ETL templates, and automatically scan for non‑compliant tasks.

Indicator Management : Standardized metric taxonomy (atomic, computed, composite), SOP for metric creation/modification, and a system that links metrics to physical models, generating SQL automatically. An intelligent parsing module selects the optimal model for complex queries.

Unified Data Service (One Service) : A configurable API platform (Buffalo) decouples data logic from service interfaces, provides usage monitoring, and reduces development time from days to about one hour per interface.

Unified Portal (One Portal) : Three portals serve decision analysis (“大圣”), sales data queries (“天狼”), and data‑asset lookup (“大禹”), providing consistent data logic for different user groups.

2.4 Data Operation Efficiency

Engineers spend 5‑10% of time on ad‑hoc queries due to information asymmetry. We built a data‑usage guide system (knowledge white‑paper) and a data‑Q&A robot (Meituan AI “Moses”) to automate 80% of repetitive queries, reducing engineer interruption by over 60%.

2.5 Data Cost Optimization

Big‑data cost consists mainly of compute and storage. We applied measures such as cleaning invalid tasks, optimizing long‑running jobs, redistributing compute load, consolidating tenants, compressing storage formats, and implementing data‑lifecycle policies (hot/cold data, retention periods).

2.6 Data Security

Security controls are applied from data generation to usage: encryption at source, tiered masking in the warehouse, and double‑layer permission checks in applications. A three‑layer system and five usage principles protect high‑sensitivity data.

2.7 Measurement Indicators

We built a comprehensive metric system covering quality, cost, security, usability, and efficiency across the data lifecycle (collection, production, storage, metric management, application, destruction). These indicators are monitored daily to drive continuous improvement (PDCA cycle).

2.8 Governance Effect Summary

Data Quality : After standardization, quality improved; 2019 task growth rate dropped ~60% vs. 2018.

Data Cost : Per‑unit cost reduced ~40% while supporting rapid growth.

Data Security : No serious security incidents after encryption and masking.

Operational Efficiency : Automation reduced engineer query time by >60%.

3. Future Plans

Data governance will evolve through three stages: passive governance, proactive governance, and autonomous (intelligent) governance. Meituan’s hotel‑travel platform is currently transitioning from proactive to autonomous, aiming to embed policies into metadata and automate issue detection and resolution.

4. Author Introduction

Jian Shu – Data Engineer, Meituan Data Science & Platform Department (joined 2015)

Wang Lei – Data Engineer, Meituan Data Science & Platform Department (joined 2017)

Luo Xi – Data Product Manager, Meituan Data Science & Platform Department (joined 2017)

Recruitment Information

Meituan Data Science & Platform Department is hiring Data Warehouse Engineers and BI Backend Engineers. Interested candidates can send resumes to [email protected] (subject: Data BP Center).