I most of this thesis can be summarized with a few points: (1) A perfect ROC (100% AUC) on fraud detection is impossible, (2) false positives have costs in both lost revenue and customer insults, and (3) the operating point with 100% fraud capture has an unacceptable false positive cost.
