Improving R&D Efficiency in the Used‑Car Division: Process, Metrics, and Practices

The used‑car market in China has entered a "stock era," prompting rapid growth of online platforms; Autohome’s second‑hand car division leveraged its integrated resources to build a C2B2B2C ecosystem and shift from lead generation to value‑added digital services.

To boost efficiency, the team adopted principles of cost reduction, productivity gains, and rapid strategy execution, emphasizing timely demand response, flexible handling, and quick releases, while establishing a methodology for performance improvement.

Key efficiency measures include standardizing R&D hand‑off processes, introducing work‑order boards, formalizing requirement breakdown and task estimation, digitizing resource allocation, adopting advanced technologies, and ensuring both technical and business integration.

Efficiency is defined as the ratio of output quality to effort; the team uses time (hours) as a unified metric for task cost, breaking down requirements into sub‑tasks of 0.5–4 hours, employing a "black‑box" estimation method to reach consensus and using median estimates as final values.

Demand quality is ensured through a two‑stage review process covering business logic, prototypes, data needs, and design completeness, with clear guidelines on documentation and review participation.

Version release follows a dynamic online model without fixed iteration cycles; each demand follows a waterfall‑like flow, with high‑priority items released immediately and lower‑priority items bundled.

Code management uses a flexible branching strategy: one branch per demand, base branches for dependencies, a shared "testing" branch for integration, and master for final releases, with tagging for release points.

CI/CD pipelines, built on Jenkins, automate testing and deployment for iOS, Android, React Native, and H5 applications, saving roughly one person‑year per app.

Task boards (Lean Cloud) track progress across stages—from review to deployment—capturing metrics such as task count, actual hours, bug rates, and throughput, enabling data‑driven continuous improvement.

Results show reduced average sub‑task time, higher estimation accuracy, increased throughput (from 6.4 to 7.2 hours/person/day), improved demand quality (review pass rate from 33 % to 67 %), and lower bug rates (from 5.98 % to 2.08 %).

The team emphasizes balancing quantitative metrics with qualitative value creation to sustain healthy, sustainable growth.