VAM: Value Propensity Score For User Acquisition Marketing Campaign

Propensity Score, proposed by Rosenbaum and Rubin in 1983, is the conditional probability of assignment of a particular treatment or marketing campaign to a user given a vector of observed features. In online advertising, propensity score has been used to capture various customer responses to a marketing campaign like click on advertisement, purchasing a product or subscribing to a service. While propensity score is essential to find the affinity of a user to be exposed to an ad, it does not ensure that we target users who are valuable to the business in terms of customer lifetime value. Our approach of using Value Propensity score could solve this problem for marketers and researchers. Our study has two major contributions 1. A composite Value Propensity score that could be used as a threshold to select potential customers instead of propensity score for observational studies, 2. A Customer Lifetime Value (LTV) based ML framework which could be used to calculate Value Propensity score. In this framework, we have used a probabilistic model, Beta-Geometric Negative Binomial Distribution (BG NBD), and a deep learning model, Long Short Term Memory Networks (LSTM), to create inferred features which are fed into a decision-tree based classification model. The use of these inferred attributes results in 1% to 2% increase in ROC AUC score compared to baseline.