Drummond and Holte introduced the theory of cost curves, a graphical technique for visualizing the performance of binary classifiers over the full range of possible class distributions and misclassification costs. In this talk, we discuss a novel way to generate a cost curve directly from a predictive model, using normalized confidence scores. We also introduce the concept of an Improvement Curve, a user-friendly metric based on the cost curve that measures the percentage improvement of replacing one model by another. We conclude the paper by applying cost curves and improvement curves to the process of risk-assessing marine cargo containers. We illustrate how implementing even a basic predictive model would save CBSA both time and money, regardless of class distributions or misclassification costs.