Customer Uplift Modelling for Targeted Email Campaigns Using Causal Machine Learning

A classical e-mail campaign mails the customers who are most likely to buy - but many of them would have bought anyway. Uplift modelling estimates, for every customer, the difference the e-mail makes (extra visit probability, extra dollars) and mails those with the largest difference. This project compares both ideas on the public Hillstrom MineThatData test (64,000 customers randomised to a Mens e-mail, a Womens e-mail or no e-mail): a random list, an RFM rule and response models (logistic regression, gradient boosting) against six uplift learners (T-, S- and X-learner, class transformation, uplift random forest). Models are judged on held-out customers with Qini / uplift curves, AUUC and bootstrap confidence intervals, and budget decisions are valued by inverse-probability weighting. A web application scores customer lists, chooses the better campaign per customer and plans a send for a budget, showing the model estimate next to the value measured in the study. An Excel decision workbook, a base R script and Power BI files are included.

Category: Marketing analytics

Master of Business Administration Project MBA01 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.