No abstract

Managing marketing funds in order to maximize profits while minimizing costs is an inherent objective of any business. A key to optimizing this relation is to select a subset of customers to target marketing expenses. An approach is presented, which helps identifying such a subset at an early stage of their life cycle. We suggest a measure to rate the users' value relative to others, and present an approach how to predict the user rating. This work compares and evaluates two algorithms (naive Bayes and random forest) which predict the user rating. Two groups were used, the first representing the top rated users, which are the part of the subset on whom the marketing funds should be focused, and a second one containing the users which may be neglected. The dataset is skewed, due to the amount of users of interest being low. It is shown how the algorithms can deal with such an imbalanced set. The algorithms are evaluated with a precisionrecall curve. The framework is compared to other prediction models and metrics such as recency, frequency, monetary model and customer lifetime value. Since the framework is flexible, it encourages including and expanding other established metrics.