Abstract:
|
Customer acquisition for online subscriptions often involve offering price discounts to convert new customers. Setting new member pricing represents a tradeoff between running discounts steep enough to induce trial while attracting customers into contracts that maximize long-run customer lifetime value (CLV). However, problematic for measuring long-run causal effects of initial pricing, i.e., via A/B tests, is their inherently time-consuming nature.
To address this shortcoming, we introduce a Bayesian nonparametric data fusion framework that expedite inference on the long-run effects of initial pricing using parsimonious ‘one-shot’ experiments on initial conversions, augmented with existing resubscription patterns found in the firm’s longitudinal CRM database. We develop a variational Gaussian process data fusion model to share information across the datasets at the customer-level. Our results show that compared to common longitudinal inference strategies for database marketing, the proposed method improves CLV estimates by better capturing downstream preference heterogeneity, and providing identification on heterogeneous treatment effects (HTE) of initial pricing.
|