September, 15 2017
David Steinberg
Dew and Ansari look at how to automate customer analytics. This can be a crucial activity for companies that manage distinct customer bases. In these data-rich and dynamic settings, visualization is essential for understanding events of interest. The value of visualization has led to the popularity of analytics dashboards. Although popular in practice, this is an area that has not attracted much academic research.
This article develops a probabilistic, nonparametric framework for understanding and predicting individual-level customer spending. Dew and Ansari propose the Gaussian Process
Propensity Model (GPPM), which uses Gaussian process priors over latent functions to model spending. The model is able to describe customer activity at scales that relate to calendar time, interpurchase time, and customer lifetime. The model generates curves that can be conveniently summarized in a dashboard, providing a visual, and easy to understand, model-based representation of purchasing dynamics. The model flexibly and automatically captures the form and duration of impact from events that affect the propensity to spend.
Dew and Ansari illustrate the use of the GPPM by analyzing data from two popular mobile games. They show that the GPPM generalizes hazard and buy-till-you-die models by incorporating calendar time dynamics while simultaneously accounting for recency and lifetime effects. It thus provides useful additional insights about spending propensity. Finally, they show that the GPPM outperforms these benchmark models both in fitting and forecasting real and simulated spending data.
Read the paper:
A Bayesian Semiparametric Framework for Understanding and Predicting Customer Base Dynamics. Ryan Dew and Asim Ansari.