Evolution of DiDi's Engineering Efficiency and Data Platform Construction

This article is based on a talk by Zhou Fan, head of DiDi's R&D tools, at the GNSEC 2020 Global Next‑Generation Software Engineering Summit. It shares DiDi’s experience in evolving engineering efficiency and building an engineering‑efficiency data platform.

1. Evolution of DiDi’s Engineering Efficiency

DiDi’s engineering‑efficiency team was founded in 2015 and has gone through three stages:

Stage 1: Early stage – the team used external and open‑source solutions to provide basic tooling for business development.

Stage 2: As the company grew, the team started building custom tools for specific problems that could not be solved by existing open‑source tools.

Stage 3: Platformization – creation of domain‑specific platforms such as project‑management, code‑management, continuous‑delivery, and effect‑verification platforms.

Currently the team focuses on a one‑stop R&D platform that improves overall developer experience, connects R&D with business functions, and enables data‑driven decision making.

2. Why Build Engineering‑Efficiency Data?

Data helps to:

Make R&D costs (people, finance, resources) visible.

Assess the level of R&D capability across diverse products and markets.

Support data‑driven improvements and assist R&D decision making.

3. Overall Approach to Data Construction

DiDi’s engineering‑efficiency platform is organized into three layers:

Data Layer: Automated, accurate collection of engineering‑efficiency related data; eliminates manual collection and inconsistencies.

Computation Layer: Modeling and processing of raw data to generate efficiency metrics.

Application Layer: Presentation of metrics to various stakeholders with customizable views.

4. Challenges and Countermeasures per Layer

Data Layer – issues of incompleteness, non‑uniform standards, and lack of clear relationships. DiDi standardizes data structures, aligns data with business scenarios, and upgrades tool capabilities (e.g., code‑review tool) to emit the needed data.

Computation Layer – high cost of interpreting data, lack of cross‑tool correlation, and missing metric体系. The team builds a unified metric体系 centered on the team perspective, reduces interpretation cost, and fills gaps via the one‑stop platform.

Application Layer – challenges in displaying the right metrics for diverse users. Solutions include enriching metric sets, providing role‑based views, ensuring extensibility, and using the platform to support both objective data and subjective insights.

5. Engineering‑Efficiency Data System

DiDi’s core metrics are divided into delivery‑oriented metrics (for business) and feature‑oriented metrics (reflecting tool health). Additional R&D‑process metrics focus on demand response, process quality, and CI/CD capability. The platform supports multi‑view, extensible, and comprehensive coverage of efficiency, cost, and quality.

Q&A – Measuring DevOps R&D Data

Key DevOps indicators include response ability, process‑quality metrics, and CI/CD frequency and latency. Quantitative measurement provides a “ruler” for teams to perceive change and trend.