V(CLV): Examining Variance in Models of Customer Lifetime Value

Statistics Department; Faculty Adviser: Peter Fader

While accurate point estimation of customer lifetime value (CLV) has been the target of a large body of academic research, few have focused on the variance of CLV (V(CLV)), which represents the degree of uncertainty associated with a customer’s expected CLV. This is ironic because academics have long known that V(CLV) is one of the most important characteristics that defines and differentiates customers from one another, affecting firms on many fundamental levels. No closed-form, forward-looking statistical procedures have been derived to estimate individual-level V(CLV). For the first time, the authors derive, predict, and validate V(CLV) using a powerful combination of stochastic models for the flow of transactions over time and the company’s profit on each transaction. They provide these estimates for 561,100 customers of an omnichannel retailer tracked over a 2.25-year period, making this one of the largest-scale CLV analyses to date. They highlight the importance of V(CLV), analyze its relationship to observable summary statistics such as recency, frequency, and monetary value, and uncover many substantive variance-related insights regarding customer segmentation, scoring, targeting, and more.