I Built CausalLens — A Free, Open-Source Causal Impact Calculator for Time Series (5 Methods, Zero Setup)

BrainWire Posted on May 30 I Built CausalLens — A Free, Open-Source Causal Impact Calculator for Time Series (5 Methods, Zero Setup) # python # opensource # datascience # statistics I want to show you a tool I just open-sourced. It's called CausalLens, and it answers one specific question that most analytics stacks get completely wrong: did this intervention actually cause the change in my metric? The problem with standard before/after analysis Before/after comparisons are everywhere. They're also almost always misleading. When you compare a metric before and after an intervention, you're implicitly assuming that the only thing that changed was your intervention. In practice, seasonality changes, external trends shift, unrelated events happen. The "improvement" you're seeing might have occurred anyway. The right answer is to build a counterfactual: a statistical estimate of what would have happened if you had never intervened. The gap between that counterfactual and your observed data is your causal estimate. What CausalLens does You provide a CSV with a time series and an intervention date. The app fits a pre-intervention model, projects it forward as the counterfactual, and reports: Estimated effect size (absolute and percentage) p-value for statistical significance 95% confidence interval Plain-English interpretation Downloadable PDF and interactive HTML reports The 5 methods and when to use each ARIMA ITS (Interrupted Time Series) Best for: single series, no obvious seasonality, straightforward before/after structure. The ITS framework is well-validated in public health and economics literature for exactly this use case. SARIMAX Best for: data with strong seasonal patterns (weekly cycles, monthly cycles, etc.). Ignoring seasonality inflates or deflates your effect estimate badly, so this matters more than people expect. Bayesian Structural Time Series Best for: when you want probabilistic output and explicit uncertainty quantification rather than a point estimat