One score fixes the past (bias). The other predicts the future (impact).
Observational data is riddled with selection bias. Treated and untreated groups differ systematically, so naive comparisons confound the true effect with pre-existing differences.
Simulate a randomized experiment after the fact. Propensity scores estimate each unit's likelihood of treatment, then re-weight or match the data so treated and control groups are balanced on observed covariates.
Explore Propensity Scores →Standard response models predict who will convert, but they can't distinguish genuine incrementality from "Sure Things" who would have converted anyway.
Predict the incremental lift caused by treatment for each individual. Isolate the "Persuadables" — the segment where marketing spend actually moves the needle — and stop wasting budget on everyone else.
Explore Uplift Modeling →