Abstract: An automated purchase transaction analyzes a purchase transaction using a probability model to determine a probability that the transaction is fraudulent. If the probability exceeds a threshold, the transaction may be manually reviewed to determine whether to freeze the account associated with the transaction. When introducing the model, the threshold may be set to a relatively high value so that a small number of transactions are submitted for manual review. After a period of time, the observed freeze rate resulting from manual reviews is compared to a target freeze rate. The threshold is then adjusted upwardly or downwardly to decrease the difference between the observed and target freeze rates.

Abstract: An automated purchase transaction service implements a two-phase analysis to identify suspect transactions and to freeze merchant accounts associated with certain of the suspect transactions. In a first analysis phase, a transaction is analyzed using a first predictive model to determine a probability that the transaction is fraudulent. If the probability exceeds a first threshold, the transaction is further analyzed in a second phase. In the second analysis phase, the transaction is analyzed using a second predictive model to determine a probability that manual review by a human analyst will result in freezing the associated account. If the probability exceeds a second threshold, the transaction is automatically frozen. If the probability does not exceed the second threshold, the transaction is submitted to a human analyst for manual review.