Incrementality analysis
Subtract what would have happened anyway to estimate the outcomes advertising truly added.
End of analysisNext steps and reference material below. Not part of the analysis result.
Next step
Proof
Full journey
How much would I lose if I turned ads off?
Estimate only the advertising-created share with a holdout or pre/post design.
Method and FAQOpen only when you need the methodology
See the three designs and their conditions
When to use it
Use it when attribution numbers are hard to trust, when evaluating brand or retargeting spend that overlaps organic demand, or when checking the real impact of turning a channel on or off. Data requirements differ by design.
What it calculates
The holdout design compares an exposed group with a suppressed group over the same period. Pre/post designs estimate a counterfactual baseline around the switch point, and adding a control group removes shared trend through difference-in-differences.
How to act
A randomized holdout gives the strongest evidence; pre/post comparison is vulnerable to anything that changed at the same time. The tool states that difference on the result screen, so report conclusions at the strength your design supports.
Frequently asked questions
Can pre/post comparison prove causality?
No. Seasonality, promotions, and competitive shifts at the same moment are absorbed into the estimate. A difference-in-differences control is the minimum; randomized holdout is strongest.
How large should a holdout be?
It depends on the effect size you want to detect. Too small and the test ends undecidable, so size it first with the experiment analysis tool.
Can the result be negative?
Yes. Sampling variation or concurrent changes can produce it. Read the interval alongside the point estimate and never shut a channel on this evidence alone.