Machine-Learning Techniques for Customer Recommendations

Abstract: Today, there is a demand for automated procedures for predicting future customers using recommendation engines in the customer relationship management market. There are already functions commonly available for finding “twins”, i.e., possible customers that are similar to existing customers, and for browsing through lists of customers partitioned into categories such as locations or lines of business. Current recommendation engines are typically built using machine-learning algorithms. Thus, it is of interest to determine which machine-learning algorithms that are best suited for making a recommendation engine aimed at customer prediction possible. This thesis investigates the prerequisites for determining suitability, and perform an evaluation of various off-the-shelf machinelearning algorithms. The supervised learner models are shown to have promise, as a direct method of identifying new potential customers. A classifier algorithm can be trained using a set that contains existing customers, and be applied on a large set of various companies, to classify suitable prospects, provided there is a sufficiently large number of existing customers.