Algorithm to predict the best time to recall a client

Let's immagine I have a dataset of calls from a call center to clients. Each call has a lot of information like at what time it was made, duration, if it was answered or not, if the client purchased something (the call center is selling stuff), and of course the client that has been called (I have an average of ten calls per client). The calls are just made by the call center, it's never the client calling. What would be the best way to do an algorithm that gives as an output the probability that the client buys something AND when to call the client, based on the previous called made to that client (Of course it's impossibile to make it time continuos, so it would be something that after the call says: the client will buy something with probability x%, call the client in 1 hours, or 2 hours, or 3 hours (putting a limited amount of classes). Any Ideas? Thanks