← Glossary
GLOSSARY · Measurement & Methodology

Incrementality

The pure additional performance an ad actually caused

Check this with your own data · Incrementality

Incremental meaning versus incrementality

Incremental generally means increasing in steps or increments. In advertising, incremental conversions are additional conversions that would not have happened without the ads. More conversions than last month alone do not establish incremental ad impact.

Incrementality asks what advertising added compared with its absence. A hundred conversions does not mean advertising caused a hundred conversions. The aim is to estimate the additional outcomes after allowing for those who would have converted anyway.

Incremental conversions = observed conversions with ads − expected conversions without ads

For the same population and period, 100 conversions with ads and an estimated 80 without ads give an estimate of 20 incremental conversions. This is a hypothetical arithmetic example; the baseline of 80 must come from a control or an appropriate model.

You cannot observe the same person both with and without advertising at the same time. Measuring incrementality therefore requires a baseline for "without ads". Holdouts, geo splits and pre/post comparisons are different ways to construct it.

Concept Question it answers Typical output
Attribution Who gets credit for a conversion that happened? Attributed conversions by channel
Uplift How far apart are exposed and control outcomes? Rate difference in pp or %
Incrementality What outcome would disappear without the ad? Incremental conversions, iCPA, iROAS

How it's measured

The most reliable method is a holdout test: randomly split users into an exposed group and a holdout group, then compare conversion rates. The difference is uplift. When randomization is not feasible, consider a geo control, a new launch, or a shutdown with difference-in-differences, and weaken causal claims as the design gets weaker.

Why CPA/ROAS alone isn't enough

CPA and ROAS relate costs to outcomes under attribution rules; they do not establish causation. A near-zero incrementality point estimate is not evidence of no effect until uncertainty and design quality have been assessed.

CPA and iCPA use different denominators

In a hypothetical experiment, $1,000 in cost and 100 attributed conversions give a $10 CPA. If incremental conversions for the same scope are estimated at 20, iCPA is $50. This is an arithmetic example, not an observed incremental share for a campaign.

Check Why it matters
Aligned cost, conversion and experiment scope Do not divide account-wide cost by lift from a subset
Group sizes and observation windows Adjust for different sizes and align conversion maturity
Estimated interval and design conditions If lift is nonpositive or its direction is unresolved, do not base a budget on a single iCPA

In the Incrementality Analysis tool, check the data shape and design before reading results and withholding reasons. More conversion rows cannot replace a missing control.

Go deeper

The three practical ways to measure incrementality (holdout, ramp-up, ramp-down) are covered in Measuring Incrementality. When randomization is impractical and only a time series exists — brand campaigns are the usual case — Brand Campaign Incrementality reads the intervention against its own pre-trend.

REVIEW NOTES

Frequently asked questions

What does incremental mean, and how is it different from incrementality?
Incremental generally means increasing in steps or increments. In advertising, incremental conversions are additional conversions caused by the ads. Incrementality asks what advertising added compared with its absence. An increase over the previous month alone does not establish incremental ad impact.
How is incrementality different from attribution?
Attribution allocates credit for observable, attributable conversions under a rule. Incrementality asks what advertising added relative to its absence. Missingness, duplication and reporting scope vary, so do not assume every report allocates 100% of all conversions.
How do you measure incrementality?
A randomised holdout is the strongest design. Without a control group you can use before-and-after comparisons around turning a campaign on or off, geo splits, or a time-series baseline. Each step down that list lets more competing explanations in.
If incrementality is low, should I turn the ads off?
Check sample size and window before cutting. Incremental estimates often carry wide intervals, and campaign types whose effect arrives late — brand campaigns especially — are understated over short observation periods.
Related:Incrementality Measurement: Validate Ad Lift With Holdouts and DiD