Did your campaign actually cause the lift?
Most dashboards tell you what happened. This one tells you why. Upload a time series, mark the intervention date, and Bayesian Structural Time-Series will reconstruct the counterfactual — the world where the campaign never ran — and quantify the true incremental lift.
The counterfactual, made rigorous
Instead of comparing "before" vs "after" — which conflates every concurrent trend with your campaign — the platform asks one question: what would have happened without the intervention? Bayesian Structural Time-Series answers it with full posterior uncertainty.
Upload your time series
A CSV with a date column, the target metric (Y), and optional control covariates (X). The model learns the relationship between Y and X in the pre-period.
Fit a Bayesian state-space model
pycausalimpact fits a structural time-series model on the pre-intervention period — capturing trend, seasonality, and regression on covariates with Bayesian posterior uncertainty.
Project the counterfactual forward
The fitted model is rolled forward into the post-period to predict what Y would have been without the intervention. The gap between actual and predicted is the causal effect.
Quantify significance
A Bayesian p-value is computed from the posterior tail-area probability, telling you whether the observed lift is real signal or could plausibly be noise.
The same algorithm, two very different verdicts
Both datasets have an intervention at day 700. Only one has a real causal effect.
Fake Success
True Success
Bring your data
Upload a CSV with a date column, a target metric, and optional covariates. Or load one of the two pre-built samples to see the platform in action.