In several industries (e-business, telcos…), a common approach to diminishing user churn is to use machine learning to score individual customers by churn probability and target them with specific messaging or offers. However, this approach may be ineffective since it does not optimize what is called "true lift" or "uplift": the effect of an action on churning probability. This talk aims at introducing uplift modeling in a tutorial like format. We’ll cover the basics of the theory as well as how to make it work in practice. We will illustrate the talk with examples from real life.
Pierre Gutierrez is a senior data scientist at Dataiku. As a data science expert and consultant, Pierre has worked in diverse sectors such as e-business, retail, insurance or telcos. He has experience in various topics such as fraud detection, bot detection, recommender systems, or churn prediction.