AI-Driven Risk-Based Testing and Intelligent Delivery System at Baidu

The article discusses the importance of risk-driven delivery in software testing, highlighting three observations: most projects have no associated risk or online bugs, a large proportion of testing activities are ineffective, and testers may misjudge or miss tests.

To address these issues, the authors propose introducing an AI‑based intelligent decision system that spans the full delivery lifecycle. The system structures historical project knowledge, automates task allocation, precisely analyzes test scope, optimizes test execution, intelligently locates problems, and supports risk‑based release decisions.

Key components include: a task auto‑distribution system that uses personnel profiles and historical data to decide who should test; a test‑scope precision analysis that leverages code call‑chain relationships and a knowledge base to recommend appropriate testing strategies; intelligent test execution that dynamically creates pipelines and minimizes test cases based on code changes; a problem‑location rule base that combines real‑time logs, traces, and configuration data to prioritize and merge issues; and a quality assessment model for release decisions, trained on historical defect, coverage, and project data.

Results from internal deployment show that the quality model identified 1,123 projects that should not be released, intercepted 318 bugs; converted 4,345 projects to self‑testing, saving approximately 2,172 person‑days; and reduced assessment waiting time from 50 hours to 2 hours.

The conclusion emphasizes that reducing reliance on human subjective judgment through AI‑driven decision making improves both testing efficiency and delivery quality.