Diagnostic Analytics: Methods, Case Studies, and Common Pitfalls in Business Data Analysis

Why perform diagnostic analytics? It helps identify root causes ("finding the disease") and highlights positive factors ("finding the bright spot"), enabling data‑driven decision making in competitive markets.

What is diagnostic analytics? According to Gartner, it answers "Why did it happen?" In practice, it means breaking down a problem, comparing current and expected states, and pinpointing the reasons behind metric changes.

Problem definition and types – A problem is the gap between current status and expectation. Three common categories are:

Occurrence problems (e.g., a 10% drop in orders compared to yesterday).

Potential problems (e.g., assessing whether weather will affect staffing).

Ideal problems (e.g., achieving a target KPI increase).

Methodology: logical tree + hypothesis‑driven analysis – Build a logical tree to structure the metric, then generate hypotheses to guide the investigation. The process involves clarifying the gap, decomposing the problem by expected vs. actual composition, and comparing two trees to locate the cause.

Case Study 1: Weather Impact Index – Goal: model how weather influences DAU (daily active users) for Meituan delivery across cities. Steps include data collection, preprocessing, constructing a target variable, and using XGBoost to predict a weather‑impact index. City‑level clustering is applied to handle varying weather effects and limited samples.

Case Study 2: Intelligent Diagnostic Analysis System – Goal: build an automated system that detects metric anomalies and diagnoses their causes. The system consists of anomaly detection (identifying abnormal nodes) and anomaly diagnosis (quantifying contributions via contribution and composition analysis). It leverages logical trees, hypothesis generation, and machine‑learning models to automate the workflow.

Common pitfalls – Failures often stem from unclear problem definition (misidentifying current state, expectation, or gap) and staying within a "comfort zone" that limits the scope of investigation, leading to missed or incorrect diagnoses.

In summary, a systematic, hypothesis‑driven diagnostic approach, combined with appropriate modeling techniques, can significantly improve the efficiency and accuracy of business problem solving.