Incrementality
Incrementality is the lift in revenue or conversions that would not have happened without a given campaign. It answers the only question that truly matters in marketing: what would have happened had we not run this ad. It separates campaigns that caused the result from campaigns that merely claimed a conversion that was coming anyway.
In short
| What it is | The difference between the outcome with the campaign and without it |
| How it differs from attribution | Attribution splits credit for measured conversions. Incrementality asks which of them would not exist at all |
| How it is measured | Experiments: geo lift tests, holdout groups, platform conversion lift studies |
| Where it tends to be lowest | Brand search and retargeting: campaigns aimed at people already on their way |
How it works
Test and control group
It is an experiment. The audience is randomly split into two parts: the test group sees the ads; the control group does not and is otherwise treated the same way. Both groups are measured over the same period with the same metric, usually orders or revenue. Since the ad is the only difference between them, the difference in results is its effect. The formula: lift = (test group result minus control group result) / control group result. The groups usually differ in size, so relative values are compared (conversion rate or revenue per thousand users), not absolute counts. Incremental conversions are then the test group's conversions minus the number it would have had at the control group's conversion rate. The control group can be formed from users (holdout), regions (geo lift test) or inside a platform tool (conversion lift); the principle is always the same.
Example calculation
Worked example: an e-commerce retargeting campaign, six-week test, 10% of the audience randomly held out.
| Test group (sees the ads) | 90,000 users, 2,700 orders, conversion rate 3.0% |
| Control group (does not see the ads) | 10,000 users, 250 orders, conversion rate 2.5% |
| Lift = (3.0 minus 2.5) / 2.5 | 20% |
| Incremental orders = 2,700 minus (90,000 × 2.5%) | 2,700 minus 2,250 = 450 |
| Orders credited to the campaign by the platform | 1,200, of which 450 were caused (37.5%) |
| Cost €18,000 | credited CPA €15, incremental CPA €40 |
The campaign works; the 20% lift is real. Yet 750 of the 1,200 credited orders would have come in without the ads: people who had already decided to buy. The true cost of a new order is €40, not the €15 in the report. Whether the campaign pays off depends on comparing incremental CPA with margin per order, not on the credited number.
Difference from attribution and when to test
Attribution works only with measured conversions and splits credit among the channels on the customer's path. All 1,200 orders in the example were credited to some channel, whether the ad had an effect or not. Incrementality, by contrast, needs a control group and answers a different question: how many of those orders would not have happened without the ad. The two complement each other: attribution suits day-to-day management; a test verifies whether a channel causes anything at all. A test has a price: the control group gets no ads for the duration, so some revenue is temporarily lost. It pays off where the budget is large enough for the result to change a decision and where there are enough conversions that the gap between groups is not random noise. Small campaigns with a few dozen conversions a month will not give a reliable result.
Why it matters
Platform reports credit conversions to clicks and views, not to causes. A campaign on your own brand name, or retargeting people with a product in the basket, reports beautiful numbers because it stands at the end of a journey that would have finished anyway. Incrementality removes that optical illusion: it measures the lift against a control group that did not see the ad. Budget then flows away from campaigns that claim and toward campaigns that cause.
From our own practice: TikTok brought 15 percent extra revenue
An incrementality test of TikTok for an e-commerce client showed 15 percent additional revenue that would not have existed without the channel. The important part: pixel attribution undervalued the very same channel, because TikTok-initiated conversions often get finished elsewhere and someone else claims them. Without the test the channel would have looked weak and the budget would have moved away, despite genuinely creating new revenue. We use our own platform DiagnostIQ for incremental impact measurement.
Common mistakes
- Confusing attribution with causation. A credited conversion is not a caused conversion. That is the entire point of the concept.
- Testing during seasonal peaks. Seasonality drowns the campaign effect. Tests belong in stable periods.
- Switching a campaign off everywhere instead of testing. Without a control group there is nothing to compare the effect against.
- Measuring once and forever. Incrementality shifts with budget and season. Large channels get retested.
Related terms
See also geo lift test, holdout test, marketing mix modeling, attribution model and MER.
Frequently asked questions
Which campaigns should be tested first?
The ones with the biggest budgets and the prettiest numbers: brand search and retargeting. That is where the gap between claimed and caused tends to be widest.
How long should a test run?
At least one full purchase cycle, typically four to eight weeks. Short tests capture only noise.
What if incrementality comes out near zero?
That is a valuable result: the budget can move elsewhere without losing revenue. It is exactly why you test.
How we can help
Incrementality testing and DiagnostIQ measurement are part of our Performance marketing agency service.