Top U.S. Bank Improves Quality of Real-time Fraud Monitoring Alerts While Cutting Process Time in Half
A top U.S. bank was receiving an unmanageable amount of alerts with various degrees of quality from its real-time fraud monitoring system. Data quality issues were difficult to detect and long data processing times were hindering the ability to effectively monitor fraud in real-time, exposing the bank to increased risks from fraudulent activities.
Challenge
A top U.S. bank was receiving an unmanageable amount of alerts with various degrees of quality from its real-time fraud monitoring system. Data quality issues were difficult to detect and long data processing times were hindering the ability to effectively monitor fraud in real-time, exposing the bank to increased risks from fraudulent activities.
The Exzac team developed and implemented an automated data validation tool to precisely detect any anomalies, errors and omissions in the data fed into the surveillance system. Using database performance diagnostic tools, the team analyzed the data loading and manipulation processes to accurately identify and resolve ‘bottlenecks’ and other inefficiencies.
Results
Using the data validation tool, some 10,000 unique semantic issues and over 25,000 syntactical constraint errors were immediately identified. Resolving the issues and errors substantially reduced the number of false-positive alerts generated which led to a manageable amount of high-quality alerts. The ongoing data validation process put in place by the Exzac team ensures that the data quality will remain high by detecting and alerting on any issues. System optimization and fine tuning reduced data record reading time from approximately 400 milliseconds to 200 milliseconds allowing for a more effective monitoring capability and thereby reducing the bank’s risks from fraudulent activities.
Next Steps
Building upon success, the Exzac team is implementing further system enhancements including an extensive management reporting tool to accurately track the workflow activities of generated alerts. Also in development, a tool that can automate the unit testing of individual fraud detection rules for more efficient and effective future implementations of alert-generating fraud detection models.
