Our client is a major provider of health insurance plans in the US. Its command center team receives over 400,000 alerts in a year. The current rule-based approach of categorizing alerts into potential incidents required some adjustment, as the client was missing on identifying a significant number of (high risk) alerts that led to major incidents.
More specifically, the client wanted to:
The developed model is currently being used to tag high-risk alerts and guide alert handling team with tip-offs to resolve the issue quickly
The algorithm-based risk score identifies alerts linked to major incidents more accurately
The typical nature of predictive variables in the model (alert type, alert duration, no. of events) made the solution very scalable by applying it to other applications, not in the scope