Causal Impact uses Bayesian structural time series to forecast a counterfactual outcome trajectory and compare it to observed data after an intervention.

When to Use

  • Single global change (pricing, policy, homepage) without a clean control group
  • Enough pre‑period history to learn trend/seasonality
  • Optional covariates available to improve fit

Workflow

  1. Model pre‑period structure (trend, seasonality, regressors)
  2. Forecast counterfactual for the post‑period
  3. Compute pointwise and cumulative impact with credible intervals

Diagnostics

  • Posterior predictive checks and residual diagnostics
  • Sensitivity to covariate inclusion and priors
  • Stability of cumulative impact over time

Interpretation

Report lift with uncertainty and show the counterfactual vs observed plot. Discuss driver contributions (if regressors used) and any caveats about extrapolation or confounding events.

Run Causal Impact Back to Service Page