As a fraud practitioner using data mining techniques to detect fraud, anomalies, outliers or other indicators of potential problems I use a combination of data mining and data matching techniques. The volumes of data in a client assignment can vary from 15 million records of company directors, 60,000 employees, accounts payables data of suppliers 900,000 and invoice transaction 11,million. I'm not a great fan of predictive technologies as the disparate data sets don't seem to fit with the techniques, but I'm open to alternative methodologies. I've recently tested a single fraud profile using "Receiver Operating Characteristic" to evaluate the sensitivity and specificity of the profile. The results fell within the ROC space.