Incrementality Test Planner
Design holdout experiments with statistical rigor
The Insight
Platform-reported ROAS frequently overstates true incrementality. 2025 benchmarks from Stella showed no consistent relationship between platform ROAS and actual incremental lift — some brands showed 4.0x platform ROAS but <1.0x incrementality factor. Without holdout testing, brands are flying blind.
The Opportunity
KLAR's incrementality testing (currently in beta) could become a key differentiator. A self-service test planner that handles power calculations, geo-region selection, and projected outcomes would justify the Core+Attribution tier and give brands confidence that their measurement is grounded in causal evidence, not correlation.
The Proposal
- 1Build a self-service incrementality test planner: select channel, configure holdout %, test duration, and daily spend
- 2Auto-calculate statistical power, minimum detectable effect, and required sample size — remove the need for a data scientist
- 3Add a test timeline visualization showing projected treatment vs control group outcomes
- 4Feed test results back into MTA and MMM calibration — creating a feedback loop that improves all measurement over time
Skill Mapping
- Designed A/B testing and incrementality measurement frameworks at Google for app campaigns
- Built automated analysis pipelines with statistical rigor for 25+ advertiser accounts
- Experience with geo-based testing methodologies and causal inference in advertising
Working MVP Prototype
Incrementality Test Planner
Design a geo-holdout test — configure parameters and check statistical power
14%
Statistical Power
4,480
Holdout Size
17,920
Treatment Size
10%
Min Detectable Effect
Underpowered test (14% power). You need 43+ weeks at 20% holdout, or increase holdout percentage / MDE for reliable results.
Cost of Test
€1,828
Estimated forgone revenue from the 20% holdout group over 4 weeks (assuming 15% true lift).
Estimated Incremental Lift
+15%
(€7,310)
Projected incremental revenue from the treatment group over the control, normalized to equal scale.
Test Readiness
Not ReadySimplified power calculation with synthetic projection data. In production, KLAR's incrementality engine would use historical variance, seasonality, and geo-region selection for precise test design.