Geo-lift testing is an experimental method that splits geographic regions into treatment (ads on) and control (ads off) groups to measure a channel's true incremental sales effect. It's the post-cookie gold standard for causal measurement.
detail
As cookie loss weakens MTA (multi-touch attribution), MMM alone isn't enough for strategic decisions. Geo-lift measures a channel's effect experimentally: 'what happens to sales if we turn this off?'
The flow:
- Split geographic regions (provinces in Turkey, DMAs in the US) into two groups.
- The treatment group runs the channel active or at increased budget; the control group runs it off or reduced.
- 2 weeks pre-period (calibration) + 4 weeks treatment + 1 week wash-out.
- Synthetic control builds a synthetic 'what would have happened' curve from the control group.
- Treatment-group sales vs. synthetic control equals pure incremental lift.
Open-source tools: Meta GeoLift (R), Google CausalImpact (R/Python), LightweightMMM.
Cost: roughly 5-10% of the monthly ad budget (opportunity cost). Output: a cookie-independent 'true lift' number, used to cross-validate MMM and de-risk large budget decisions.
Example: a brand's MMM estimated Google Search ROI at 3.5x. A 4-week geo-lift test: treatment spend raised 50% across 8 provinces, control held flat across 70. Result: incremental ROAS 4.25x, p=0.03 (significant). MMM had been slightly underestimating; budget was increased.