Optimizing the ability to sell new features to existing customers, and determining what features to sell to prospects, is a difficult challenge for the banking industry. This paper proposes solutions to those problems.
The assumption is that the company doing the study already has statistics to predict the probability that a given customer will elect a given option, and also to predict what the company will yield in profit should they elect to take the offer. It is also assumed that the company can only offer a selected number of its options to any given customer, and that other constraints complicate the issue.
This begins to sound very much like a simple integer programming problem, but given millions of customers and prospects to deal with, and dozens of solutions, the problem becomes unwieldy. The proposed solution is to aggregate the customer list into groups, which are much more manageable.
I think this clearly is a good idea, and the example given demonstrates what looks like a good solution, but the author does not explain how to aggregate. There would seem to be many potential criteria to use, and some of those criteria would seem to lead to failure. This sounds like interesting research, but the aggregation process remains an unsolved mystery. Thus, this is an interesting step, but a long way from a solution to the problem.