MWITA-MIO-2026-003 · Evidence A · P1
A randomized ghost-ad retargeting experiment found 17.2% more website visits and 10.5% more purchases while attaining comparable lift precision at at least an order of magnitude lower control-group media cost than intent-to-treat or PSA designs.
What this does not establish
Does not imply all retargeting is incremental, that the 10.5% purchase lift transports to other advertisers, or that predicted ghost assignment remains unbiased under every auction implementation.
Counterevidence & uncertainty
The positive randomized lift contrasts with blanket claims that retargeting only claims credit for inevitable purchases, while its magnitude remains campaign-specific.
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.