optimizing customization: beyond a combinatorial approach?

Say I have a generic non-linear model (ANN, Random Forest, Gradient Boosting, etc) that wants, based on a set of features (price of a product, service duration, age, etc), to give me a prediction of whether a client will buy my product and clearly I can change my "offer" (say a combination of price and service duration).

I want to find the combinations of such factors, that is, the best compination per user, of my offer so that I can maximize the chance of each of my clients will buy my product/service. of course, I can do this in a "combinatorial" approach (trying each combination and see which one yields me the highest implied probability of success) but combinations could be infinite depending on how I set it up or would not scale when users increase considerably.

Since it’s impossible that this problem was not encountered before – and I have neither the intention nor the need to re-invent the wheel, could someone point to some algorithms/topics that adress this issue?