Diagnose and fix low conversion rigorously

After a new promotional banner went live, checkout conversion fell from 42% to 35% week over week; traffic stayed flat, the marketing mix moved 10% toward paid social, and payment failures ticked up slightly. Lay out a step-by-step plan to isolate the root cause: check the instrumentation, pin down the conversion definition, construct a funnel, break the data down by device/geo/new versus returning, examine latency and errors, and apply holdouts or difference-in-differences to disentangle the promo effect from channel shifts. State the minimal queries/metrics you would pull, the statistical power required for an A/B rollback or fix-forward test, and the executive-level insight you would present, including expected revenue impact and next actions.

Overview: This question tests a Data Scientist's skills in analytics and experimentation, including instrumentation validation, funnel and segmentation analysis, causal inference methods (holdouts/DiD), system reliability diagnostics, statistical power and sample-size estimation, and the ability to turn findings into executive-level impact statements. It is often used to see whether a candidate can tell measurement or product regressions apart from marketing-mix shifts and to show both conceptual understanding of causal inference and practical application of data querying, metric definition, and experiment design in the Analytics & Experimentation domain.