I have several Gradient Boosting models in production. I have gotten requests from users of the final score\model of beeing able to get insight into why a customer has a high probability of an outcome, say churn or buying a product.
Showing which variables are important in a model, along with the customers variables could of course bring some insight for a trained statistician\data scientist, but that hardly helps an untrained eye.
Is there any smart way of explaining why one customer gets a high score? I picture a stored process that takes a given customer as input, and outputs some report that shows some of the most important drivers of the customers score. This can of course be exceedingly complex in such a model, but have you seen any attempts at it?
