|
Abstract:
|
Customer Lifetime Value (CLV) predicts the future revenue each customer may generate and serves as a foundational metric for any organization. While these models are well-developed for consumer customers, the business-to-business use case imposes unique challenges. From the customer side, the wide range of revenue values across a diverse spectrum of business customers requires special attention in developing a reasonable statistical model. From the company side, multiple product offerings allow for a customer to change their product mix at any point in the future which can lead to sudden changes in the revenue stream data. Lastly, the availability of historic data presents constraints on the training data for a reliable forward looking model. To tackle these issues, we first redefined the CLV as the total future expected discounted revenue in the next 2-years. We then proposed a hierarchical model to derive a 5 year prediction which can be generalized over longer time horizons. This provides us with more accessible tools in supervised machine learning and allows for richer features to be used. Our empirical results demonstrate dramatic improvements in the CLV prediction.
|
