Counting Your Customer from an ‘Always a Share’ Perspective

The Pareto/NBD or BG/NBD model are very popular in customer base analysis in the non-contractual setting these days. But both models assume that customer will not back once gone. This is not true in the real world, for customer always want to try different brands and always have chances to come back even after a long silence period. This paper proposes a new approach for customer base analysis in the non-contractual setting under more realistic “always a share” assumption. The new framework, called the Hierachical Interrupted Poisson Process model (HIPP), assumes that customer go through two stages constantly in their “lifetime” with a specific firm: being active for some period of time, and then being inactive for some period of time, which is modeled by a two state continuous time Markov process. While active, the number of transactions made by customer follows a Poisson process; while inactive, there would be no transactions made by the customer. Some useful expressions for customer base analysis is derived. Bayesian simulation methods for parameters estimation is developed and implemented via Markov chain Monte Carlo (MCMC) simulation. By implementing the proposed HIPP model and contrasting models in various simulative purchasing environments, I find the HIPP performs well under different scenerios in terms of prediction of total repeat transactions, while the BG/NBD or NBD perform well only under limited scenerios. The empirical study further shows that HIPP performs better in terms of predicting customer purchase than existing benchmark model