Incrementality is the extra result a marketing activity causes, compared with what would have happened without it. If a campaign is credited with 100 sales but 70 of those customers would have bought anyway, only 30 sales are incremental. Incrementality testing measures that difference by comparing people who saw the marketing with similar people who did not.
How incrementality works
Ad platforms and analytics tools use attribution to hand out credit for conversions. Attribution tells you which ads a buyer touched, but not whether those ads changed the decision. Brand search ads are the classic example: someone who types your company name into Google was probably going to find you, ad or no ad. Remarketing to people who already have items in their basket raises the same question.
Incrementality testing creates a fair comparison. The main methods are:
- Holdout or lift tests. A random share of the audience is held back from seeing ads, and conversions in the two groups are compared. Google Ads and Meta both offer conversion lift studies to eligible advertisers; eligibility and minimum spend vary by account.
- Geo experiments. Ads are switched off or changed in some regions and left running in similar regions, and sales are compared. In the UK this might mean using postcode areas or TV regions as test cells.
- On-off tests. Pausing a channel for a period and watching total sales. These are cheap but weaker, because seasonality and other changes muddy the result.
The arithmetic is straightforward. Suppose a test group converts at 2.4% and the matched control group at 2.0%. The lift is 0.4 points, or 20% relative to control, and only one in six of the test group’s conversions (0.4 ÷ 2.4) was caused by the ads. If the platform credited all of them, the true cost per incremental sale is six times the cost per acquisition it reported.
Why it matters
Budgets follow reported results, and reported results are produced by the platforms selling you the ads. Incrementality is the check on that. It often shows that brand search and retargeting are less valuable than they look, and that prospecting campaigns reaching new people are more valuable than last-click reports suggest.
It has become more important in the UK as consent rules under PECR and UK GDPR mean fewer users can be tracked. With gaps in tracking, attribution becomes more of an estimate, and a controlled test is often the most reliable evidence available.
Common mistakes
- Treating platform ROAS as incremental. Platform-reported conversions include sales that would have happened anyway.
- Running tests that are too small or short. A test needs enough conversions, and long enough to cover your sales cycle, to separate a real effect from noise.
- Changing other things mid-test. A new offer or price during the test spoils the comparison.
- Choosing unlike regions. Comparing London with rural Wales measures the regions, not the ads.
How to act on it
Start with the spend you are least sure of, usually brand search or retargeting. Decide the single metric you will judge it on, such as total sales or total brand-name clicks from paid and organic combined. Choose matched groups or regions with similar past performance, run the test for at least one full sales cycle, and change nothing else while it runs. Then calculate incremental cost per acquisition and move budget accordingly. For a whole-business view across channels, marketing mix modelling complements individual tests.
Designing and reading these tests is part of how I run performance marketing.