MWITA-MIO-2026-006 · Evidence A · P1
Twenty-five major-advertiser field experiments showed that even very large randomized ad tests often leave ROI too imprecise for confident budget decisions.
What this does not establish
Does not say experiments are uninformative or should be replaced by attribution; it establishes power and economic constraints under small effects and volatile purchase outcomes.
Counterevidence & uncertainty
Randomization removes selection bias but not sampling uncertainty, contradicting dashboards that present incremental ROI as precise merely because the campaign population is large.
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.