A Recommender System for the Upselling of Telecommunications Products

Telecommunication providers are always seeking ways to upsell products to corporate customers. Traditionally, the telecommunication provider’s Account Managers build a business relationship with the customers and try to persuade them to upsell. However, only some instances result in a successful upsell while others are unsuccessful. First, we focus on a binary classification framework for predicting the successful upsell of products and services, using data from one such telecommunications provider. Through this prediction model, we illustrate a recommender system for voice products/services to corporate customers of the telecommunications company. We use a logistic regression classifier to automate the selection of customers that are most likely to upsell. We also acknowledge that there may be monetary costs associated with misclassification errors. Note that minimizing losses (or maximizing revenue) may conflict with the objective of minimizing errors and so we address this trade-off. We apply our predictive model to recommend a set of target customers to approach for upsell, illustrating the different accuracy results for different cost weightings. We also show that the success rate of upselling products to the selected customers is dramatically improved when compared to the traditional approach.