Incrementality: estimate what would not have happened without the action.
Answer:
Incrementality is the difference between the observed result with an action and the result that would have occurred without it for the same population and horizon. The second result is counterfactual: it must be constructed through randomisation, a credible comparison or an explicitly falsifiable model.
Method and decision:
Define the estimand before choosing a method: outcome, population, treatment, comparator and horizon. Describe concretely what happens without the action, choose an assignment unit based on contamination risk, calculate the minimum detectable effect and ensure the decision remains useful with a wide interval. Pre-register the metric, exclusions, duration, analysis and decision rule before observing results. Choose the least demanding method able to create a counterfactual compatible with the decision’s grain, mechanism, delay and risk. A local test can support a similar rollout, not automatically an annual allocation; document extrapolation separately.
Worked example:
A campaign spends €100,000 and the platform attributes €400,000 sales. A comparable unexposed group would have generated €270,000 for the same population. Incremental sales are 400,000 − 270,000 = €130,000. Incremental ROAS is 1.3. At 45% gross margin, contribution before creative costs is only €58,500, below spend. The campaign captures existing sales. Reduce it or redefine targeting rather than scale it on attributed ROAS.
Limits:
A badly executed experiment can be less credible than a good quasi-experiment. Rare or small effects can require samples incompatible with the calendar. An average effect can conceal important segment or area heterogeneity.