Acquiring new customers is more expensive and challenging than retaining existing customers. Therefore, companies should take care of their existing customers in order to retain them. These so-called retaining actions should only be used for unsatis ed customers who are about to transfer their business somewhere else.
In this thesis we analyze the Finnish National Opera’s (FNO) customers and develop a model to identify churning customers. Furthermore, additional continuous variables are predicted from the customers, e.g. the number of tickets purchased next year. These additional predictions are intended to further increase the efficiency of retaining actions by being able to personalize and more accurately target individual customer needs. A random forest model is developed and benchmarked to other popular models in churn modeling. All the models are built using the FNO’s sales system data.
Random forests outperform other models in churn modeling and in most of the continuous variable prediction problems. The random forest model has high accuracy when used for churn modeling, but its performance is poorer when predicting continuous variables. Our model is the most accurate when predicting how many days in advance customers purchase their tickets. The churn model is able to predict the churn of all customers with equal accuracy, and while it is accurate in identifying churners, it is not able to find all of them.
Hung Ta (Aalto University): Customer Churn Modeling Using Sales System Data