Estimation of Dynamic Treatment Effects from Marketing Campaigns

To date, conduction of marketing campaigns using personalized treatments is a widespread practice for companies in various fields around the world. However, despite the popularity of such campaigns, evaluation of their effectiveness over time and with respect to each specific customer is a non-trivial task. Often, a randomized experiment, allowing you to clear the analyzed effect from the effects of various factors, is used to solve this problem. In this paper, we study the effect of marketing treatment in dynamics and taking into account its heterogeneity among individuals, using the data of a randomized experiment of one marketing campaign conducted by one Russian restaurant chain and aimed at reactivating churn customers. As a result of the study, we are developing the methodology based on the Two Models Method, which allows, firstly, to get an idea of the variability of the treatment effect over time; secondly, to divide the investigated churn segment of consumers into sub-segments depending on the nature of the treatment effect (positive, neutral, negative); and thirdly, describe each of the received sub-segments and give recommendations to the restaurant chain on interacting with each of them.