In predictive modeling, the goal is to make predictions about outcomes on a case-by-case basis: an insurance claim will be fraudulent or not, a tax return will be correct or in error, a subscriber will terminate a subscription or not, a customer will purchase $X, etc. Lift is a measure of how much better the statistical model does than not using a model at all. Decile lift is this measure applied to deciles of the target records ranked by predicted probability (for a binary outcome) or predicted amount (for a continuous variable). For the top decile lift the steps, for a 0/1 classification problem, are
1. Split records into training and validation samples
2. Train a model on the training data, apply it to the validation data
3. Rank the validation data in order of predicted probability of being a "1"
4. Count the number of actual 1's in the top decile of the validation data
5. The lift is the ratio of #4 to the average number of 1's per decile across the entire validation set