Incrementality

Attribution credits an ad when a conversion follows it. Some of those people would have bought without the ad: existing customers, people who searched for you by name, people already on their way to checkout. Incrementality strips them out.

Example: if 2.4% of the test group converts and 2.0% of the control group does, the lift is 20%. Most of the test group's conversions would have happened anyway; the ads added the difference.

1. Split the audience randomly into a test group that can see the ads and a control group that cannot.
2. Run the campaign for long enough to collect meaningful conversion counts in both groups.
3. Compare conversion rates. Lift = (test rate − control rate) ÷ control rate.

• Retargeting, which reaches people already close to buying and often looks better in attribution than in lift tests.
• Brand search, where people were already looking for you.
• Any channel where the platform reports more conversions than your sales data supports.

Meta now offers incremental attribution as a model when you create an ad set: it optimises for, and reports, conversions that its models predict were caused by the ad. Meta says results cannot be compared across ad sets using different attribution models.

Meta's A/B testing can also measure strategies on a cost per conversion lift basis, as well as cost per result.