Used structured & unstructured data to identify and reduce fraud risk by 33% for a Top US card issuer

Marketing, Digital Businesses

We helped a major US payments network reduce fraudulent transactions by developing a customized fraud prediction algorithm. The approach involved combining structured and unstructured data from a variety of sources – including Retail PoS data, ATM & other transaction data, geographic information, and historical cardmember data – to build a predictive model that identified high fraud-risk transactions. Simulations indicated a 33% reduction in fraud through the application of this solution.


  • Client was a major payments network.
  • Controlling fraudulent transactions was a key strategic priority for the client.
  • No customer level data was available for traditional fraud analysis.
  • Studied mechanics of fraudulent behavior; Combined structured & unstructured data such as retail purchase data, card reload data, geo data, and transaction data to find patterns of fraudulent behavior




  • Identified patterns of high, medium, and low fraud propensity.
  • Transactions in the riskiest segment were 125 times as likely to be fraudulent as the average.
  • Recommended strategy expected to lead to 33% reduction in fraud