Our client was a large Communications Service Provider (CSP) in Africa. This client had already implemented models for churn identification/ reduction utilizing traditional methods that score customers based on their individual behaviors. They wanted to expand on these traditional models using network behavior (Who’s communicating with Who? What type of communication? How frequently and How long?).
CoreCompete was responsible for defining the data model, developing the ETLs and building the models that would allow the client to understand the role a particular subscriber plays in the network (influencer, follower). This system was required to work in an automated manner and needed to aggregate a large volume of call detail record (CDR) data. The system included capabilities for detecting issues such as super-nodes and other outliers and ensure that the recommendations were meaningful.
CoreCompete completed this project in 12 weeks. The models incorporating this new input were generating 28% more lift than the previous models (that were based solely on a single subscriber level data). The system scales to deal with millions of CDRs on a regular basis. The entire process is automated to minimize the operational impact on the analytical and operations teams.