Causal Inference

What Is Causal Inference?

Causal inference is the discipline of distinguishing what caused an outcome from what merely correlated with it. Developed in statistics (Rubin, Pearl) and economics (Angrist, Imbens), it provides the mathematical and experimental toolkit for answering counterfactual questions: what would have happened if we had not intervened?

Also Known As

• Marketing team: "causal measurement," "true impact analysis"
• Sales team: "influence analysis"
• Growth team: "causal lift," "counterfactual analysis"
• Data team: "causal inference," "counterfactual modeling," "treatment effect estimation"
• Finance team: "incremental ROI"
• Product team: "feature causal impact"

How It Works

You observe that users who use Feature X retain 2x longer than users who don't. Correlation, not causation. Causal inference asks: would those users have retained longer anyway, even without Feature X? Method 1: randomize access to Feature X (A/B test). Method 2: if randomization is impossible, use propensity score matching to find users who were equally likely to use Feature X but didn't, and compare their retention. Method 3: use a policy change that forced some users to adopt (instrumental variable). Each method estimates the counterfactual.

Best Practices

• Prefer randomization (A/B tests, geo-tests) over observational methods whenever possible.
• Pre-specify your causal question before looking at the data.
• State the counterfactual explicitly — "what would have happened without this?"
• Use multiple methods when the stakes are high; convergence across methods strengthens claims.
• Distinguish average treatment effects from heterogeneous effects across segments.

Common Mistakes

• Calling an observational correlation "the effect" without addressing selection bias.
• Assuming A/B tests need no causal thinking — they're causal inference in the simplest form.
• Confusing "statistically significant" with "causally meaningful."

Industry Context

SaaS and B2B teams increasingly apply causal inference to product analytics — which features cause retention, which onboarding steps cause activation. Ecommerce and DTC teams use causal methods for promotion design and pricing. Lead gen operations use causal inference to distinguish sourcing effects from sales-cycle effects in pipeline conversion.

The Behavioral Science Connection

The human brain is a causation-detection machine that sees causes everywhere, including where none exist. Behavioral economists call this the narrative fallacy — we construct coherent stories to explain random patterns. Causal inference is the formal discipline of resisting that urge. It makes us ask "how would I know?" before we say "this caused that."

Key Takeaway

Before claiming any intervention worked, articulate the counterfactual and the method used to estimate it — otherwise you're reporting correlations in a lab coat.