Design and Implementation of a Quality Score Model for the Beike Transaction Platform

"Quality Score" is a data‑driven quality metric model built by Beike's transaction platform based on business characteristics and targeted quality indicators, aiming to identify system quality issues, improve quality, and raise personnel awareness.

1. Why build a quality‑score model? Different roles in R&D have varying quality requirements, and periodic quality analysis often relies on vague terms like "excellent" or "average" without quantitative insight. Quantifying quality indicators with data provides an objective view of product quality.

2. Current industry quality‑measurement landscape – Process and efficiency metrics are mature across major companies; a typical set of indicators is shown (e.g., Meituan's R&D measurement system). Beike’s advantage is a business‑specific model using weighted‑deduction design to reflect quality changes more intuitively.

3. What is a quality score? Quality is divided into online (production) quality and process (offline) quality, with key measurable indicators abstracted, often referencing standards such as ISO 9126.

4. How are indicators determined? Indicators are chosen from industry‑standard metrics and business‑specific pain points; high‑impact, high‑loss, or frequently recurring issues receive higher weight, while less critical ones receive lower weight.

5. How is the calculation formula set? After defining indicators, a formula must be simple enough to understand yet expressive enough to measure quality; Beike adopts a "weighted‑deduction" scheme. The specific rules are illustrated in the following diagrams.

Online‑issue scoring deliberately excludes operational metrics like monitoring coverage or slow‑SQL follow‑up rates to keep the score focused on true quality assessment.

6. Implementation plan

Identify data sources and acquisition methods (e.g., internal platforms keones, fast, xmen).

Define data cleaning, calculation rules, and storage solutions.

Design data analysis processing workflow.

Design data visualization and reporting.

7. Results

Configuration can be completed within minutes for certain indicators.

Supports personalized configurations.

Quality score visualizes KPI‑level data, highlighting key issues.

Data analysis accumulates knowledge, continuously improving product quality.

System has been running stably online for over a year.

Over 200 project retrospectives and improvement actions have been recorded.

8. Using data to control risk – Example: two severe online incidents in May‑June were caused by UI bugs that escaped monitoring and were only discovered via user feedback. A retrospective led to the creation of an automated UI inspection topic, now under continuous case addition.

9. Using data to improve quality – General capability building includes consolidating business/technical expertise, raising personnel awareness, and strengthening quality operations.

Organize Quality Month activities to boost awareness and timely issue handling.

Build a virtual city for online environment acceptance testing.

10. Conclusion – Scoring quality is an experiment in quality‑operation practice. This article covers the quality‑score model; other parts of the series (full‑process assurance, exemption testing, Quality Month, Gold Bug) will be released later. Feedback and suggestions are welcome.