Heterogeneous treatment effect estimation of an advertising campaign: A casual impact estimation from a randomized control study

Abstract

The main objective of this project is to estimate the causal impact of an advertising campaign with a randomised control study on customers, As businesses are continuously trying to improve the customer experience and identifying the right way to target customer for maximum impact (ROD), the study of causal effect from the campaign will determine its effectiveness. The impact in this study is measured as visits and conversions that happen as a result of the advertising campaign. Though it is a random control experiment, the users in the study differ in their underlying characteristics such as age, sex, income etc. Due to these differences, the causal effect of the campaign when calculated could be affected by these confounding factors. Therefore, the treatment effects at various levels of similarity grouping are examined for accurate measure for causal effect. Two modelling approaches, logistic regression and decision trees are used to define stratification of users such that there is similar type of users in each stratum. It is observed that the treatment effect varies between the strata. Through the use of decision tree method, user segmentation based on their features can be obtained directly using which those users where the causal impact is high can be targeted for maximum benefit.