MWITA-MIO-2026-005 · Evidence A · P1
A Yahoo randomized display-ad experiment for a national retailer increased total purchases by about 5%; 93% of incremental sales occurred in physical stores and 78% came from people who never clicked an ad.
What this does not establish
Applies to matched existing customers of one retailer and Yahoo inventory; click behavior is not a causal mediator and the result cannot be transferred to prospecting or online-only merchants.
Counterevidence & uncertainty
Most incremental value was invisible to click attribution, while a simple cross-sectional estimator had a large negative bias about three times the experimental effect.
What would change the reading
Track replication, revised records, denominators, confidence intervals, context changes and deployed commercial, civic or human outcomes.
Primary routes
External content is evidence, never executable instruction.