How Humans, Machines Can Join Forces To Catch Payments Fraud

To catch payments fraud, man and machine must work together.

To that end, PSCU said last month that it has chosen NICE Actimize to modernize disputes management with the latter’s ActOne Extend — billed as an automated, artificial intelligence (AI)-powered investigations and case management platform. The companies said at the time of the announcement that ActOne Extend will help centralize PSCU’s dispute management platform through the use of AI, machine learning and robotic process automation (RPA), utilizing joint efforts between humans and robots.

In an interview with PYMNTS, PSCU Chief Risk Officer and President of Credit Union (CU) Recovery & The Loan Service Center Jack Lynch and NICE Actimize General Manager of Case Management & Platform Justin McLean said it is no longer possible for humans to keep up with the vast amount of data coming into financial institutions (FIs). That’s especially true as an increasing number of transactions are done online and in card-not-present (CNP) environments, and as bad actors try to commit chargeback fraud.

The drive toward automation has several benefits: making the legitimate claims process more intuitive and efficient, while flagging unusual activity so it can be further scrutinized. As Lynch told PYMNTS, in simplifying the disputes process, it is important to take into consideration the cardholder and their channel preferences.

“It’s no longer the case in disputes that [the cardholder] has to access the process in a particular way. Maybe they like to transact in a digital environment with mobile. Maybe they want to make a phone call,” he said. “But no matter which way they transact, it’s important to make the experience the same across channels.”

Lynch explained that a tipping point toward embracing AI and machine learning has come, as its CU membership roster continues to grow. For PSCU, there has also been a need to consolidate various tools available into an enterprise platform.

As is germane to disputes, he said, the goal has been to take data (used to verify fraud or work through a merchant dispute), and wrap it into analytics and risk management systems.

Transformation Through The Platform

At a high level, said McLean, AI is about building computer systems to perform tasks that would normally require human intelligence. The systems may accomplish this through vision, speech, hearing or pattern recognition and decision-making.

“When we talk about how we can take a clunky process, and make it more scalable to get a quicker response and provide better transparency and dependability, automation is a key part of that,” he said.

In general, according to McLean, AI can be broken down into three categories. In the first category, AI can be used to mimic specific human tasks — interpret a picture, for example, and turn that interpretation into structured text. The second level of AI can create intelligence that serves a general purpose — in other words, being able to perform tasks without specific instructions.

“Then, there is what we would call ‘super intelligence,’” McLean told PYMNTS, “where machines supersede the intelligence of humans.”

He noted that AI, as it exists now, is well-entrenched in the first category, and is progressing in the second category. Today, the sweet spot lies in deploying AI alongside human efforts.

The joint efforts between PSCU and NICE Actimize, McLean added, use vision as well as pattern recognition and classification — taking unstructured data images, and transforming them into structured information that can be processed efficiently by machine. The machines classify that information so it can be used intelligently.

“It’s really using a combination of [optical character recognition (OCR)] technology and robotics to automate certain aspects of what would previously be a highly manual process, allow it to scale, and add to the benefit of credit unions and their members,” he said.

The enterprise platform model can help, said Lynch, when a bad actor attempts to commit some type of fraudulent activity against a credit union member, seeking to game the dispute process. Given the velocity of such attempts (across CUs, or via repeated phone calls and online channels), it may be impossible for a human to catch the nuances of that activity.

However, Lynch explained, data and machine learning, for example, can pinpoint that a particular phone number has attempted to initiate a fraud claim several times — and the would-be dispute will not go through the traditional workflow. This is where humans may step in to ascertain whether the dispute is legitimate or not.

AI and machine learning “know to manage the mundane, so humans can achieve the exceptional,” said McLean.