What problem does it solve? When a store manager asks "why did sales drop yesterday?", the risk is jumping straight to a fabricated explanation or chasing a false alarm caused by normal weather or weekday variance. This Skill first verifies whether the drop is statistically abnormal using similar business days, suppresses false alarms, and only then drills down to a single root cause with one concrete countermeasure. ## Core Features & Use Cases - Counterfactual Anomaly Gate: Compares actual sales against an expected range (mean ± std) built from similar business days matched on weekday, weather, season, and holidays, with confidence levels based on sample size. - Ordered Root-Cause Drill-Down: When a drop is significant, it investigates in a fixed sequence—cancelled/lost items, hourly heatmap, menu engineering quadrants, and organic CSAT/NPS—then attributes one primary cause using time-order, magnitude, and confidence rules. - Fixed Output Templates: Returns one of three fixed formats (not anomalous, indeterminate due to low confidence, or significant drop with cause and next action), keeping responses consistent and preventing emotional overstatement. - Use Case: A manager asks "why was last Tuesday so bad?" The Skill finds the drop is within the expected range for rainy Tuesdays and replies with evidence that no anomaly occurred—closing the conversation without triggering downstream winback campaigns. ## Quick Start Ask why sales dropped yesterday at your store and let the Skill verify against similar business days whether it was actually abnormal before investigating the cause.
