Leveraging Network Effects for Predictive Modelling in Customer Relationship Management

Abstract

Predictive modelling and classification problems are important analytical tasks in Customer Relationship Management (CRM). CRM analysts typically do not have information about how customers interact with each other. Phone carriers are an exception, where companies accumulate huge amounts of telephone calling records providing information not only about the usage behaviour of a single customer, but also about how customers interact with each other. In this paper, we do not measure network effects, but we analyze techniques to improve classification tasks in CRM leveraging network effects. In contrast to traditional classification algorithms, we try to take into account the information about a customer's communication network neighbors in order to better predict usage behavior. The presumption in our ex-periment is that a customer's SMS (Short Message Service) usage also depends on the SMS usage of his social network. However, analysing huge amounts of call detail data which exhibits a graph structure poses new challenges for predictive modelling. In our work, we focus on ways to improve predictive modelling and classification leveraging data about the social network of a customer. We describe the re-sults of an experiment using real-world data form a cell phone provider and benchmark the results against traditional approaches.

Keywords: Customer Relationship Management, Network Effects, Propositionalization, Relational Data Mining, Telecommunication Industry