What is marketing attribution?

A model is one specific rule for splitting the credit. A last-click attribution model gives all of it to the last channel the user clicked. Data-driven attribution is Google's model that uses machine learning on converting and non-converting paths. Multi-touch attribution splits credit across several touchpoints.

Models change over time. Google Analytics Help says the first click, linear, time decay and position-based models are no longer available in GA4 as of November 2023¹. Google Ads Help says Google Ads no longer supports them either: it offers last click and data-driven, and data-driven is the default for most conversion actions². So say which model produced a number.

Credit also depends on the attribution window: how far back the rule looks.

Attribution looks at who touched the path. It does not ask whether the sale would have happened anyway. An academic study of 15 US advertising experiments at Facebook found that observational methods often failed to produce the same effects as the randomised experiments³.

Cause is the job of incrementality: the extra sales an activity brings beyond what would have happened without it. A test, such as a holdout, is one way to measure it. How the methods compare is in MMM vs MTA vs incrementality tests.

For ZFD: We use attribution for day-to-day steering and never as proof. We ask what the numbers look like under another model, and where the money is large, under a test.