Capstone Audit a Causal Claim

The one idea: When people choose their own group, "group A outperforms group B" usually describes who those people already were, not what the group did to them. Before betting money on turning Bs into As, ask whether the gap would survive if the two groups had been the same kind of people to begin with.

⚠️ The trap: "Group A brings in far more revenue than group B, so converting Bs into As is worth a fortune." That feels right and it's wrong. When people chose their own group, the gap mostly describes who they already were, not what the group did to them — and it shrinks toward zero for the reluctant people an expansion actually recruits.

One slide, full audit, honest verdict. This is the track's final exam, worked end to end the way you'd actually do it — the evening before the strategy review, with an AI assistant open in one window and the deck in the other.

The slide:

Loyalty-program members generate 2.4x the revenue of non-members. Members average $1,200/year vs $500/year for non-members. Expanding enrollment 3x projects to $12M incremental ARR (annual recurring revenue).

Everything this track has built — selection, the bias equation, DAGs (causal diagrams of what causes what), colliders, quasi-experimental designs, LATE (local average treatment effect — the effect only on the people a lever actually moved), the AI audit patterns — gets used here. By the end you'll have the transferable artifact: a three-part verdict for any claim shaped like "group A outperforms group B, so converting Bs into As is worth $X." Which is, conservatively, half the causal…