Machine learning can sound a bit creepy to the uninitiated. It enables lightning fast analysis of incredible amounts of data to look for patterns that exist. It is then able to apply those patterns to the future and predict behavior. In the card-not-present payments industry, that application is mainly the ability to predict and prevent fraud while enabling transactions that might be rejected by less sophisticated fraud prevention methods.
“It’s real,” said Brian Ducharme, vice president of product and innovation at First Data-STAR Network. “It can help you save money. It can help can help you maintain customer relations.”
Data scientists began this work by creating rules for the computers that manage financial data, scanning for fraud. The trend now is to let the machines learn by themselves, aided by greatly increased storage capacity and faster processing chips. The challenge for merchants is learning about machine learning
“You have to be conscious of not hiding data from your machine,” Robbie Fritts, manager of fraud and risk for OpenTable, said. “You have to be sensitive that one model does not do it all. You have to be very smart about how you are training it.”
When a customer’s credit card is wrongly refused, that shopper may grab another card, which is a problem if you run a network, or may just shop somewhere else, which is a problem if you’re a merchant.
“Right there that customer has gone from friend to foe,” Ducharme said. “Anything we can do to prevent that is great. If machine learning makes that possible I think we should all adopt it.”
Data scientists are working to create programs that scrutinize purchases and separate the ones that are fraudulent from those that simply deviate from normal buying patterns.
“We have created tools that are unifying model building,” said Sandeep Grover, senior vice president of global e-commerce for Feedzai. “It can help you explain what needs to be done next. You’ll see this more and more.”