CATE Toolkit · Causal uplift modeling

Who actually responds to the campaign — and who would have anyway?

A lightweight, scikit-learn-compatible package for individual causal treatment effects (CATE) in campaigns, A/B tests and feature rollouts — one API across meta-learners plus a reproducible multi-dataset benchmark of when each one works.

ConceptCausal AIMIT

The question

Who actually responds to the campaign — and who would have anyway?

Average treatment effects tell you whether a campaign worked on average. Budget decisions need to know for whom — and existing libraries are either methodologically broad but hard to use, or easy to use but thin on transparent benchmarks.

Approach

How it works

Wrap, don’t rebuild. A single API over proven implementations of X-learner, causal forest and DR-learner, reducing error risk compared with re-implementing core algorithms.
Marketing diagnostics. Segment reports, a budget-allocation simulator and Qini-based campaign prioritisation.
A benchmark as the contribution. A reproducible comparison of which learner works under which data characteristics — sample size, treatment share, effect heterogeneity and noise.

Stack & sources