01 — Causal inference
A $300M renewal book rested on the belief that advancing customer maturity protects renewal. Everyone could see the correlation; no one had shown causation. Part I establishes the causal estimate. Part II answers the harder question: for which accounts, and which dollars?
Part I — Identification and estimation
An identification design that blocks reverse causation by construction, five estimators run independently on 11,166 renewal events, and stress tests chosen to give the confounding objection its best shot. The result: a +12.9pp causal renewal lift from a 0.5+ maturity advance, adopted as the standing planning value.
Part II — Heterogeneity and targeting
A causal tree grown on doubly robust scores partitions the book into four leaves by estimated impact, from a +25.5pp primary target to a null leaf excluded by rule. 82% of flat ARR sits where the effect is weak or absent. An incrementality screen, not effect size, picks the pilot.
Note
All figures and data shown in both decks are synthetic and anonymized to preserve confidentiality. The analytical patterns and methodological framework reflect real work; specific numbers do not.
