When working for a US Bank, I developed a system for predicting customer attrition. As far as such systems go, it was pretty straightforward - construct features, run it through a decision tree, do some pruning, generate predictions, and check the accuracy of the predictions on an on-going basis.
I do remember when time came to apply for the patent that I did insist that I be first inventor as having contributed almost all of the intellectual and programming work for the project. I had done a M.Sc and ABD in machine learning, and published in the area.
There was a program manager on this who had to have cross-validation explained to him.
I see now that my name is at the bottom of the list. Oh well, at least they didn't take me off completely.