What problem does it solve? Headquarters teams often cannot tell whether a coupon, price change, or promotion actually worked, or whether it is safe to roll out to other stores. This Skill turns "did the campaign work, and should we expand it?" into a rigorous decision flow instead of a single vanity metric. ## Core Features & Use Cases - Causal method routing: Splits interventions into holdout-based campaigns (get_campaign_roi) and non-holdout interventions (get_intervention_effect with CausalImpact), then labels each as win, inconclusive, or negative using posterior probability thresholds. - Winner attribution and rollout selection: Decomposes what drove the win (product, time slot, offer, or baseline strength) via menu engineering and business summaries, then selects similar stores that have not yet run the tactic and checks menu/inventory compatibility per store. - Guarded two-stage rollout: Presents target stores, operation counts, SQL equivalents, and reversibility, waits for human approval, then dry-runs one store before applying to the rest, and later measures the rollout's own lift against non-rollout control stores. - Use Case: An HQ manager asks "did last month's coupon work, and can we expand it?" The Skill verifies significance, attributes the win to a specific menu item, picks three similar stores, dry-runs one, applies the rest after approval, and schedules a two-week follow-up comparison. ## Quick Start Ask the assistant to verify whether last month's coupon campaign had a statistically significant lift and, if so, plan a store-by-store rollout to similar locations.
