A predictive model learns to predict outcomes by being trained on data with known outcomes. The value of a predictive model is that it does better at predicting an outcome than you would do just by using the averages in the training data. There are various metrics for assessing how well a model does, and one favored by marketers is lift, which is particularly relevant for the portion of the records predicted to be most profitable, most likely to buy, etc. For example, you might say that the lift is 2.0 for the top 15% of the records, meaning that selecting the 15% of the records deemed by the predictive model to be the most profitable, or most likely to buy, you will double your profit, or purchases (compared to not using a model and just selecting randomly).
Marketers like to divide lists up into deciles and calculate lift by decile; a chart based on these calculations is a gains chart. Marketers also like to test different treatments applied to customers (pricing, messaging, web page formats, etc.). The response improvement from a given treatment, compared to the standard treatment, is called uplift. Uplift predictive models, based on a prior A/B test, can predict this uplift on an individual basis, which allows “microtargeting” of individuals with messaging predicted to be most successful for them.