Attribution, incrementality or MMM: which method for which decision?

Attribution, controlled experiments and marketing mix modelling answer different questions. Start with the decision to support: describe a journey, estimate an incremental effect, explain aggregate variation or compare constrained budget scenarios. A platform-attributed contribution is not automatically a causal effect.

Attribution for execution

Attribution describes observed touchpoints under a chosen convention. It can support operational reporting, but its result depends on identity resolution, observation windows, conversion definitions and the allocation rule. It does not create the unexposed counterfactual required for a causal claim.

Experiments for incrementality

A controlled test compares a treatment with a credible counterfactual. Specify the population, assignment, primary outcome, exposure integrity and stopping rule before launch. The estimate remains tied to the tested context, time window and implementation; it is not automatically transferable.

MMM for aggregate decisions

MMM relates aggregate variation to media and contextual variables over time. It requires data definitions, enough variation, diagnostics and sensitivity checks. Triangulate methods when possible. This page provides educational orientation only: it does not replace estimand design, a data audit or scientific review.