Customer lifetime value applied to mobile apps

Using customer-level data from 2017 for two mobile app word games in France and the US, this study explores the adaptability of customer lifetime value (CLV) models, widely used for e-commerce data, to the mobile app domain. We evaluate the Pareto/NBD model and its four extensions, focusing on dropout process simplification, transaction regularity, dropout-transaction rate correlation, and the inclusion of predictive covariates like gameplay and video views. Although the Pareto/NBD model has strong predictive performance, an extension accounting for dropout-transaction rate correlation excels in out-of-sample performance. By comparing with the e-commerce-based CDNOW dataset, we highlight significant disparities in the relative performance of different models, emphasizing the distinctive nature of mobile app data. We find also slight variations in estimated parameters across different markets, platforms, and games. • We adapt classic CLV models to freemium mobile app data from two countries. • Correlations of purchase frequency and dropout significantly boost predictions. • Abe's correlated Pareto/NBD outperforms simpler BG/NBD in mobile game contexts. • Rewarded-video covariates enhance forecast accuracy in short-lifecycle settings. • Our results guide marketers in targeting, retention, and monetization strategies.