Page 27 - GS210901

CoverStory

adding that these efforts were only moderately successful, and can rely on biometric data alone to identify the differ-
he added, because they relied on connecting identities to ences between the two."
devices.
Advanced AI solutions like BionicIDs are difficult to copy,
"Fraud solutions then evolved into predictive analytics and their ability to provide a unique identifier facilitates
based on artificial intelligence (AI) and machine learning accurate user identification and protects customers from
(ML) models which detected fraud at different stages of manipulation and impersonation threats, Renshaw stated.
the payment process, often in silos," Renshaw said. "For
example, know your customer versus transaction fraud What's next?
versus anti-money laundering solutions provide a fraud Berkhahn pointed out that effective AI solutions employ
probability or riskiness score. ML models have evolved federated learning and graph neural networks (GNNs),
over time with more robust algorithms and strategies re- methodologies that enable technologies to operate holis-
placing legacy ones." tically as part of a collective network. He described fed-
The predictive stage erated learning as a way to collaboratively build models
without sharing sensitive data; graph neural networks
Next-gen solutions focus on protecting end-to-end cus- are data structures composed of nodes connected across
tomer journeys across the global payments landscape, a network. "Research in federated learning and GNN has
from enrollment and onboarding to continuous transac- improved in recent years, providing additional opportuni-
tion and account monitoring. These solutions, which have ties for machine learning innovation," he said.
become more sophisticated over time, are embedded into
an organization's fraud and risk management processes, Renshaw summarized AI best practices as "models are ro-
Renshaw stated. bust, models are observable and explainable, model gov-
ernance is enabled, model management is automated to
Over time, advanced predictive analytics became more drive operational efficiency and finally, models are fair."
prescriptive by providing actionable intelligence and en- He urged AI resellers and end-users to incorporate these
hancing business outcomes when they detected fraud, best practices. "Any AI model implemented should be ex-
Renshaw noted, adding that next-gen predictive AI uses plainable," he added. "Blackbox models reduce both cus-
robust ML models that are measurable and explainable. tomer and auditor trust."

"An example of advanced predictive capability is the Renshaw also emphasized monitoring of existing models
Feedzai Genome—a dynamic visualization engine includ- and remediating performance issues. "AI is a powerful
ed in the Feedzai solution to show hidden connections tool and should ultimately benefit the customers by pro-
between transactions," Renshaw said. "This provides an viding fair, inclusive decisions," he said.
intuitive way for risk teams to proactively identify emerg-
ing financial fraud patterns and not just flagged fraudu- Snitkof expects AI to continue to expand in capabilities
lent transactions. Risk teams can use this intelligence to and scope due to its proclivity for continuous process im-
put preventative measures in place." provement. Ocrolus leverages AI with humans in the loop,
he stated, describing the process as attended ML applied
The prescriptive stage to unstructured content. "The model employs predictive

As advanced predictive analytic solutions evolved, they modeling and valuation metrics with a defined feedback
developed the ability to identify users (good or bad) from loop based on human supervision," he said, adding it iden-
individual composite characteristics based on thousands tifies document types, data fields and actual data, contin-
of data points, Renshaw recalled. uously improving as it learns from human input.

They could recognize users based on digital assets such Newlin envisions a new generation of AI solutions in an
as behavioral biometrics (touch/keystroke, mouse move- environment resembling a vast hall of mirrors, as hu-
ments, mobile spatial sensors, etc.); behavioral analytics mans supervise machines that supervise humans. "When
(user journey, fast travel check, date/time of connections, I think about machine learning with a human supervisor,
etc.); and device and network data, he explained. These it makes me think we could be in this loop forever, be-
compiled digital assets create a unique digital fingerprint cause now we would need another sort of AI and machine
(which Feedzai calls a "BionicID") for every user, whether learning solution to check the review of the human who
legitimate or bad actor. supervised."

"BionicIDs can identify legitimate users to speed them Dale S. Laszig, senior staff writer at The Green Sheet and managing
through their journey or prevent bad actors from commit- director at DSL Direct LLC, is a payments industry journalist and content
ting fraud," Renshaw said. "Prescriptive analytic solutions strategist. She can be reached at dale@dsldirectllc.com and on Twitter
utilize BionicID based data to identify good vs. bad users at @DSLdirect.

27