A leading telecom operator wanted to predict call volumes for planned or unplanned events, and reduce the number of calls to the call center by identifying customers likely to call and the reason for calling. Enhance customer experience with proactive outreach and faster first-call resolution.

Developed call reason group logic through multiple rounds of model refinement, including multi-level models for different group of reasons granularity. Included a framework of ensemble models, clustering, and reinforcement learning to determine customer treatment and channel. Implemented a real-time model scoring to predict call reasons and engage before a customer calls. Proven framework and models for call classification and operations optimization.