Self-driving car tech takes aim at financial fraud

Register now

Neural network systems — which try to emulate the human brain — will become increasingly important as the stolen identities from the recent Equifax breach and others start surfacing in the coming years on fraudulent payment card or loan applications.

"It's a technology that comes out of the computer vision or self-driving car space, and it is very good at picking up the very small nuances most often associated with images," said Ken Meiser, vice president of identity solutions at ID Analytics, the San Diego-based consumer risk management firm.

ID Analytics has launched ID Score 9.5, the latest version of its fraud scoring system for new account applications. The upgrade brings convolutional neural network technology to the forefront as a way to visualize data to pick up on the subtlest differences and previously confirmed red flags in a bogus application. It can go as deep as providing business or bank clients with "signals" on whether the fraud attempt is coming from a first- or third-party source.

The neural technology essentially creates a model that presents data in a 3D representation, Meiser said. "It has the regular set of behaviors, what the person looks like, the address, phone number, a transaction history and then it examines the application to see if anything is different."

Much like self-driving cars don't rely on reading the body language of other drivers — a concern that has actually led Ford to dress up its driverless car operators as car seats to avoid confusion — ID Score 9.5 doesn't have anything to do with the physical presence of the applicant. Instead, it views the data in a way that creates a "picture" that either equates more to a good outcome or a bad outcome, Meiser added.

In breaking the application down to small pieces, the neural network takes everything it knows about an applicant and compares it to what is on the application — in a matter of seconds.

"It pulls all of that information in and determines how many times those factors have been associated with this applicant and how many times they have not," Meiser said. "Often, the core information on a fraudulent application is correct, but the contact information is related to the fraudster," he added. "The system can see that an unusual address has also been used by three other people in a specific period of time."

It also provides a rescoring service 24 hours later, in case the application was simply the first in a series of fake applications submitted in other places over a short period of time.

In recent Aite Group research, 65% of financial institution executives said their priority for future investment centered on machine-learning analytics for fraud mitigation. At the same time, Aite found that nearly every bank executive it interviewed cited account takeover as a top pain point, with application fraud right behind.

Effective fraud prevention is becoming a competitive issue for banks, Aite said. Early adopters of advanced analytics will have an edge in improving the customer experience, report author Julie Conroy, research director and fraud expert at Aite, noted.

"Financial institutions and merchants that do not invest in advanced technologies will be left behind as their competitors are able to enable better customer experiences with less friction and fewer false positives, while also reducing losses," Conroy said. "The use of advanced analytics is a very important tool for FIs."

The challenge comes on the regulatory side, especially if the analytics are unsupervised, Conroy added.

"Regulators are demanding the same kind of model documentation and transparency from the fraud folks that the AML and credit risk teams have had to do for years," Conroy said.

ID Analytics establishes different risk scoring models for different industries, including one for retail credit card applications and one for bank card applications. In each case, however, a company or bank's underwriting system makes an application programming interface call to ID Analytics and submits the personal identifiable information for the applicant, as well as the location of the transaction, either at a store or a bank.

In getting notices of an application for service "about a million times a day," ID Analytics builds a network that establishes significant insight into the process, Meiser said.

"We get some insight into the internet protocol address, the e-mail address and other factors and are able to deliver back our numerical ranking of the risk and some additional insight as to why we feel that way," Meiser said.

As many as 97% of the applications flow through without any further information exchange, but between a half or two-thirds of the fraud is concentrated in the remaining 3%, Meiser said.

That's where the extra due diligence and the work of a convolutional neural network comes into play, he added.

For reprint and licensing requests for this article, click here.
Payment fraud Fraud detection Machine learning Artificial intelligence