AI-Powered Anomaly Diagnosis and Root Cause Analysis for Gaming Business Intelligence

Introduction – The session shares how 37 Mobile Games applies AI for intelligent analysis across five major parts: abnormal diagnosis & root‑cause analysis, QBI fluctuation analysis & insights, intelligent building & questioning, AI data analysis reports, and a Q&A.

Speaker – Shi Feixiang, senior big‑data platform engineer at 37 Mobile Games, presented the technical details.

1. Abnormal Diagnosis & Root‑Cause Analysis – With rapid AI advances, the platform evolved from simple querying and charting to intelligent summarization and fluctuation attribution. A production‑grade intelligent BI must provide comprehensive BI functions, sub‑second query speed, and 100% accurate results. User research showed that business users prefer AI‑driven analysis over simple Q&A, seeking insights into why data anomalies occur.

2. Life Case Example – An illustration shows that correlation‑based predictions (e.g., leaf greenness + temperature → more mosquitoes) can be misleading; causal analysis correctly identifies temperature as the true driver.

3. Current State of Data Anomaly Investigation – Challenges include high manual cost, reliance on subjective judgment, and long analysis cycles. Business users must manually cross‑analyze dimensions (games, channels, etc.) when anomalies like sudden DAU drops appear.

4. Analysis Approach – A layered strategy is proposed: (1) Business layer – define monitored metrics and dimensions; (2) Data‑warehouse layer – develop dimensions and metrics; (3) Root‑cause algorithm layer – use contribution‑rate models to locate anomaly sources; (4) Presentation layer – generate root‑cause reports and drill‑down dashboards. The process involves multi‑level dimension pruning and metric decomposition to pinpoint core factors.

5. Contribution‑Rate Calculation – Basic metrics use month‑over‑month differences; derived metrics combine group‑level and inter‑group contributions. The system flags daily anomalies, identifies affected games or ad groups, and pushes diagnostic reports to business users.

AI Data Analysis Report – A multi‑agent pipeline (editor, researcher, reviewer, reviser, writer, publisher) generates detailed analysis reports. Agents collaborate via tools such as web search and data analysis, following a structured workflow (CrewAI). The report includes overall effects, service quality metrics, trend analysis, and actionable suggestions.

Q&A – Discussed the necessity of providing report outlines to large language models for higher‑quality analysis, emphasizing the need for solid data‑analysis expertise.

Overall, the sharing demonstrates how AI, big‑data technologies, and multi‑agent systems can automate and enhance business intelligence, root‑cause diagnosis, and reporting in the gaming industry.