Causal G-Compution

Bayesian observational causal inference & CATE estimation with PyMC to evaluate product feature uplift without A/B tests.

Code & Infrastructure: Full implementation, Docker environment, and tests are available on GitHub: github.com/krystiankorzec/bayesian-causal-feature-impact

In B2B SaaS, randomized controlled trials (RCTs) are often impossible. You cannot forcibly disable a major workflow feature for 50% of your enterprise accounts just to run a clean experiment. When forced to rely on observational telemetry, standard regression and naive \(p\)-values frequently mislead product roadmaps. This post demonstrates how combining causal DAGs with Bayesian G-computation recovers true conditional average treatment effects (CATE) from noisy product data.

TBC