Presented By O'Reilly and Cloudera
Make Data Work
5–7 May, 2015 • London, UK

Measuring the benefit effect for customers with Bayesian predictive modeling

14:55–15:15 Wednesday, 6/05/2015
Data Science
Location: King's Suite - Balmoral
Average rating: **...
(2.67, 9 ratings)
Slides:   1-PDF 

Prerequisite Knowledge

Basic R and basic CRM


Offering benefits is a classic and important strategy for acquisition of new customers and churn management. This strategy consists of customer targeting and offering benefits, and its purposes include improving awareness and impressions of services/products, and attracting new customers.

Methods for customer targeting procedures have been significantly improved by various statistical methods and algorithms. Unfortunately, the benefits offering step has suffered from a lack of effective methods due to insufficient customer data, a wide variety of benefits, and most of all, requiring predictions of customer reactions.

Recently, several alternatives have been proposed such as A/B testing, which is easily applicable and modifiable. However, these methods have several issues such as lack of long-term prediction, mandatory modification of data/benefits, and limitation in quantitative comparison of testing options and the influence of external factors/treatment.

In this presentation, a new approach is proposed that integrates multivariate testing and a Bayesian time series prediction model to measure casual inference. This approach is implemented in R and Google’s CausalImpact package. An empirical application of the approach is presented. A discussion will follow the conclusion.

Photo of JeongMin Kwon

JeongMin Kwon


JeongMin is a data scientist. JeongMin finished M.S. in Industrial Engineering and and has worked as data scientist and database engineer in the industry. Her current interests focus on anything to do with data analysis and measurement -finding insights from data and playing with data.