How Meituan’s R&D Team Cut Tens of Millions in Resource Costs: A Practical Guide

Practice

In Meituan, multiple teams provide resources (SRE, big‑data, risk control, advertising, etc.). After receiving monthly cost statements, the team dissects each line item to devise targeted solutions. The process is illustrated below.

1. Define Methodology

“Plan‑Do‑Check‑Act” (PDCA) is applied, with secondary iterations at each stage.

Plan/Standard : Build awareness → set goals → analyze current state → define evaluation metrics.

Do : Break down atomic projects → choose solutions → assign owners → optimize atomic metrics.

Check : Define actions → evaluate results → locate system issues → correct standard actions.

Act : Regular retrospectives → produce reports → refine knowledge → upgrade methodology → set next goals.

2. Planning (Plan & Standard)

The core aim is to measure work with 2‑3 quantitative indicators. Without measurable metrics, work remains qualitative.

Establish awareness : Leaders must ensure the team understands cost significance.

Set goals : Use SMART principles to clarify what cost problems to solve.

Analyze current state : List data to judge the team’s cost‑optimization stage.

Define evaluation metrics : For Meituan’s team, total cost and cost per order were chosen.

3. Execution (Do)

3.1 Build Thinking Process

Decompose atomic projects, assign responsible owners, and optimize metrics using a pyramid “MECE” structure.

3.2 Practical Analysis Framework

A second pyramid visualizes the hierarchy of cost categories.

3.2.1 Business Host Cost

Includes physical/virtual machines, containers, storage, bandwidth, cloud, CDN, etc. Optimizing host utilization and refactoring high‑utilization services reduced costs dramatically.

Note: later cost rose slightly due to new services.

3.2.2 Big‑Data Cost

Big‑data workloads consume YARN containers (vCore+Mem) and HDFS storage. The team reduced storage by tiering data (cold, warm, hot) and compressing traffic logs, saving over 20% of space.

3.2.3 Shared Cost – Risk Control & Anti‑Scraping

Costs are allocated proportionally to usage. Teams must monitor both absolute request volume and relative share to avoid hidden cost increases.

3.2.4 Shared Cost – Secure Data Warehouse

Cross‑team offline data exchange is handled by a secure warehouse; cost is shared by resource proportion, requiring efficiency improvements in data models.

3.2.5 Shared Cost – Advertising

Advertising costs are split by CPM or CPC metrics; teams should track “cost per mille impressions” and “cost per thousand revenue” to evaluate impact.

3.2.6 Other Costs

New cost items appear monthly; spikes must be investigated for new projects or metric changes.

4. Check

Focus on action compliance, result evaluation, system issue location, and corrective actions.

Action compliance : Verify that planned actions are executed.

Result evaluation : Assess whether metrics improved.

System issue location : Iterate quickly between solutions.

Correct standard actions : Encourage continuous questioning and improvement.

5. Act – Review and Iterate

Regular retrospectives : Strengthen abstract, thinking, and management skills.

Report generation : Enable others to learn cost‑optimization.

Iterative knowledge : Apply PDCA repeatedly, turning occasional successes into habitual ones, and earning technical patents.

Conclusion

Cost optimization is often overlooked but remains critical. Meituan’s R&D team saved tens of millions of yuan over a year by systematically applying PDCA, focusing on resource utilization, and collaborating across teams. The experience offers practical insights for any organization seeking efficient, cost‑effective engineering.