Uplift modeling predicts how much a specific action, such as a discount, changes an individual customer's behavior, separating the persuadable from the people who would buy anyway. A propensity model asks who will buy; an uplift model asks who will buy because of this.
Updated
Category
AI, Metrics
Also known as
uplift model; incremental response model
In practice
It needs a randomized control group in the training data. A model built on past campaign results without one learns who buys, not who buys because of the campaign.
Something out of date?