causalml

Why we included this project

A typical A/B test reports the average lift, which is useful but sidesteps a harder question: which customers should actually receive the treatment? CausalML tackles that directly by estimating treatment effects at the individual level, so you can see how an intervention plays out differently across your population instead of assuming one average fits everyone. That makes it most valuable for campaign targeting and personalized engagement, where the point is to spend budget on the customers who will actually respond to an ad or offer. It bundles a wide range of uplift models, from tree-based learners to meta-learners, behind a common fit-and-predict API, and includes evaluation tooling for checking how well a model separates responders from non-responders, whether the input is a randomized experiment or observational data. If you have outgrown plain response prediction and want defensible treatment-effect estimates, this is a solid place to start.