Why incrementality matters
Attribution distributes credit. It does not test causation. A channel can earn lots of attribution credit and drive zero net new revenue if it’s just intercepting customers who would have converted anyway.
The textbook example is branded search: a customer types your brand name into Google, sees your paid brand-search ad, clicks it, and converts. Attribution credits the paid search ad. Incrementality says: would the customer have converted via the organic result instead? If yes, that spend produced zero incremental revenue.
This pattern shows up everywhere, retargeting, prospecting on warm audiences, brand-keyword bidding. Without incrementality testing, the channels that look most efficient on attribution dashboards are often the channels with the lowest actual lift.
How to measure incrementality
Three common methods:
- Conversion lift studies (holdout tests). Platforms like Meta and Google offer these natively: randomly hold out a control group from seeing your ads, compare conversion rate against the exposed group. The difference is incremental lift.
- Geo experiments. Pause or boost a channel in one set of geographies, leave others as control, measure the revenue gap. Slower, more disruptive, but works for channels without native lift testing.
- Matched-market studies. Pair similar markets, treat one, control the other. Statistically cleaner than naive geo splits but requires careful market matching.
How to use the results
Incrementality results don’t replace attribution, they calibrate it.
A typical workflow:
- Run an incrementality test on a channel (e.g. Meta retargeting)
- Compute the incremental ROAS (revenue lift ÷ test spend)
- Compare it to the attributed ROAS the model reported
- Adjust the attribution model’s weight on that channel to bring the two in line
- Re-test annually as audiences and creative refresh
This is sometimes called a “calibrated MMM” or “incrementality-calibrated attribution”, a hybrid that gets the granularity of attribution and the causal grounding of experiments.
Common mistakes
- Treating attributed ROAS as incremental ROAS. They are different numbers. A channel with 5× attributed ROAS may have a 1.5× incremental ROAS.
- Running too few tests. A single test on a single audience tells you about that audience at that time. Build a cadence.
- Confusing lift with significance. A 10% lift with 200 conversions is noise. A 3% lift on 50,000 conversions is real. Power your tests properly.
FAQ about Incrementality
What is incrementality?
Incrementality measures the portion of conversions that would NOT have happened without a given marketing action. It is the truth-test for attribution.
How do I measure incrementality?
Three common methods: conversion-lift studies (holdout tests offered by ad platforms), geo experiments (pause spend in test geos), and matched-market studies.
Why is incremental ROAS lower than attributed ROAS?
Because many attributed conversions would have happened anyway. Branded search, retargeting, and warm-audience prospecting all get credit for users who were going to convert without seeing the ad.