Causal Impact

What Is Causal Impact?

When holdback experiments or conversion lift tests are not feasible, Causal Impact offers an alternative with no revenue sacrifice. Typical use cases: measuring the effect of a new campaign, quantifying the impact of a budget increase, or proving the value of brand campaigns to stakeholders. The open-source package is available for R and Python and requires a sufficiently long pre-period without the intervention being measured.

Causal Impact is a statistical method developed by Google Research that lets you measure the causal effect of a marketing action without setting up a real control group. The method analyzes time series data and constructs a counterfactual scenario: what would have happened if you had not run the campaign? The difference between the projected and actual outcome is the causal effect.

Technically, Causal Impact is built on Bayesian structural models. The open-source package (available for R and Python) uses historical data and control variables to compute a forecast for the intervention period. A prerequisite is a sufficiently long pre-period without the action being measured, plus at least one stable control time series that was unaffected by the intervention.

For Google Ads managers, Causal Impact is especially useful when Holdback Experiments or Conversion Lift are not practical. Typical use cases include measuring the effect of a new campaign, quantifying the impact of a budget increase, or proving the value of brand campaigns. Compared to Incrementality Testing with control groups, Causal Impact is less precise but can be applied at any time without sacrificing revenue.