Incrementality testing measures the TRUE causal impact of marketing by comparing test regions (ads on) against control regions (ads off). Uber ran a 3-month geo-lift test and discovered Meta ads weren't driving new riders—saving $35M annually. Over 52% of brands now use incrementality, but most marketers still rely on flawed attribution. Free tools: Meta GeoLift and Google Meridian (both open-source).
The $35 Million Problem With Attribution
Your attribution dashboard says Meta ads drove 10,000 conversions last month. But how many of those people would have converted anyway—through organic search, direct traffic, or brand awareness built over years?
Attribution can't answer this. It only tells you WHO converted after seeing an ad, not IF they converted BECAUSE of it. This is why Uber's marketing team ran a test that changed everything.
The Uber Case Study
Uber's Rider Performance Marketing team paused Meta ads for three months in selected US and Canada markets. The result? No measurable negative impact on rider acquisition. Signups followed the exact same seasonal pattern with or without ads.
This meant Meta's attribution was claiming credit for conversions that would have happened organically. Uber reallocated the $35M annually to more effective channels—including Uber Eats expansion.
"Uber's team overlaid seasonality data, discovering that signups followed an identical pattern despite fluctuations in ad spend. This suggested that paid ads were not actually driving new rider growth."— Growth-onomics
How Geo-Lift Testing Works
Geo-lift experiments divide your market into test and control regions. You run ads normally in test regions while pausing (or reducing) in control regions. After 4-8 weeks, you compare outcomes.
The math is simple:
Incremental Lift = (Test - Control) / Control × 100
If test regions show 15% more conversions than control regions, your ads delivered 15% incremental lift. If they show the same results? Your ads aren't working—regardless of what attribution claims.
Free Open-Source Tools
Two major platforms have released free, open-source incrementality tools:
Meta GeoLift
Released by Meta's Marketing Science team, GeoLift uses Synthetic Control Methods to create a statistical "twin" of your test regions from control data. It handles market selection, power calculation, and causal inference automatically.
- Language: R package
- License: MIT (free for commercial use)
- GitHub: facebookincubator/GeoLift
Google Meridian
Released January 29, 2025, Meridian is Google's open-source Marketing Mix Model with built-in incrementality calibration. It can combine geo-level data with incrementality experiments to validate and fine-tune model outputs.
- Language: Python
- Minimum: 8GB RAM, 2+ years of data
- Docs: developers.google.com/meridian
How to Get Started
A basic geo-lift test requires:
- Historical data: 12+ months of conversion data by region
- Geographic granularity: State, DMA, or city-level reporting
- Budget for holdout: 10-15% of spend paused in control regions
- Test duration: 4-8 weeks minimum
Start with your highest-spend channel. If it shows low incrementality, you've found budget to reallocate. If it shows high incrementality, you've validated your investment with causal proof.
The Triangulation Approach
No single measurement method is perfect. The most sophisticated marketing teams now "triangulate" using three methods together:
- MTA (Multi-Touch Attribution): Day-to-day optimization decisions
- MMM (Marketing Mix Modeling): Long-term budget allocation
- Incrementality Testing: Causal validation of both
When all three agree, you have confidence. When they disagree, incrementality is the tiebreaker—because it's the only one measuring causation, not correlation.