Abstract

Customer churn is a term used to describe customers leaving a company by no longer using their services
or products. Companies should develop and target retention strategies towards customers at risk of
churning, because customer acquisition is more costly than customer retention. At-risk customers can be
identified using predictive machine learning. Previously, predictive churn modelling has typically been made
for companies offering contractual products, where payments are made on a regular basis following a
subscription or other contract. In these cases, the moment a customer churns is intuitively identified.
Defining when a customer churns from a company offering non-contractual products, where the purchase
occasions are sporadic, is more difficult, as the exact churn moment is both subjective and hard to identify.
No studies of non-contractual customer churn have been made in the winter tourism industry, the industry
in which non-contractual churn is defined and predicted in this thesis.
The purpose of this thesis is to define and predict non-contractual customer churn in the winter tourism
industry. The purpose is fulfilled by creating two different definitions of customer churn; one where the
complexity of non-contractual churn is captured through the integration of industry knowledge and the
theoretical background, and one that is based solely on the theoretical background. Five frequently used
machine learning classifiers are evaluated for the prediction, revealing that our first definition of churn yields
the highest AUC performance when predicting customer churn in this case. We conclude that if the
definition of churn is sufficiently complex, non-contractual churn in the winter tourism industry can be
predicted with a high performance using an XGBoost classifier. When data of previous reservation and
purchase patterns is considered, the classifier achieves what is considered to be an excellent AUC
performance at nearly 86%.