UK Retail Financial Services

The client faced significant challenges in identifying unusual customer orders and financial irregularities hidden within high-volume transactions. Risky operational behavior was not visible through traditional reporting methods. Additionally, there were no labeled anomalies available, making supervised detection approaches impractical. The business needed a way to automatically detect and prioritize risky transactions without manual rule-based checks.

We constructed a process-aware event log where each invoice represented a customer order and each transaction line an event. Using process mining and unsupervised machine learning, we engineered behavioral features including process duration, event count, and order value. We created a composite process risk score and applied anomaly detection and clustering techniques to isolate abnormal cases. The solution visualized risk across the entire order population, enabling data-driven decision-making.

• Process-aware event log construction
• Behavioral feature engineering
• Composite process risk scoring
• Unsupervised anomaly detection
• Behavioral clustering for prioritization
• Explainable risk indicators

• Constructed a process-aware event log where each invoice represented a customer order and each transaction line an event.
• Engineered behavioral features including process duration, event count, and order value.
• Created a composite process risk score combining multiple indicators.
• Applied anomaly detection and clustering techniques to isolate abnormal cases.
• Visualized risk across the entire order population for decision-makers.

• Identified approximately 5% of customer orders as anomalous, based on extreme financial values and unusual execution patterns.
• Discovered that financial anomalies (very high or negative order values) were a stronger indicator of risk than processing time alone.
• Revealed that anomalous orders clustered into distinct behavioral groups, enabling prioritization and investigation.