Counterfactual

A counterfactual is the outcome a unit, a user, a market, a company, would have experienced under a different condition than the one it actually experienced, most commonly what would have happened without a treatment that was in fact applied. It is never directly observed: a single user either saw a feature or didn't, and there is no way to rerun history for that same user under the other condition.

Causal inference is, in essence, the problem of estimating a counterfactual well enough to compare against what was actually observed. A randomized A/B test solves this by using a comparable group, the control, as a stand-in for the counterfactual of the treatment group, relying on random assignment to make the two groups alike on average. Quasi-experimental methods such as difference-in-differences, synthetic control, and propensity score matching build an approximate counterfactual from observational data when randomization isn't available.

The concept matters because it defines what a causal effect even means: the difference between what was observed and what the well-estimated counterfactual would have shown. Confusing a merely correlated group, users who didn't opt in, say, with a valid counterfactual is one of the most common sources of biased causal claims.