Geo Holdout Experiments — The Methodology That Replaced MTA
Geo holdout experiments — randomly assigning regions to holdout vs. control — produce iROAS (incremental ROAS), revealing that platform-reported ROAS often significantly overstates true incremental value.
Scale
Industry-standard methodology; table stakes for sophisticated marketers
Before
Multi-touch attribution (MTA) counting all attributed conversions as platform ROAS — overstating incremental value
After
Geo holdout experiments: randomly assign regions to holdout vs. matched control → iROAS (incremental revenue ÷ incremental spend)
Key Insight
iROAS ≠ platform ROAS. Example: Meta shows 2.5 ROAS, iROAS comes in at 1.2 — much of the 'attribution' was capturing organic demand that would have happened anyway.
In a Snowflake Conversation
Every sophisticated marketer now uses geo holdouts to calibrate their channel mix. This is table stakes knowledge for any measurement conversation.
My Read
Practitioner commentary coming soon.
Relevant Conversations