Industry leaders in the field of behavioral analysis (Facebook, Apple, Microsoft, Google, and Amazon – FAMGA, as some call them) are “decision architects,” said Joe Mattey, Vice President of Enterprise Risk Analytics, and Chief Risk Officer at USAA.
Typically a FAMGA product designer will put together a solution, such as a screen dialog, and fine-tune it over the course of many randomized, controlled trials, to see if the proposed solution gets the desired result—or whether some components need to be changed.
In this way, behavioral analysis is looking not just at a decision outcome, it is looking at how that decision is made. Behavior might change when small environmental conditions differ, such as a change in the layout of a screen dialog.
“Consumers may be using several devices simultaneously,” Mattey said. “We find the leaders have digital breadcrumbs that leave a trail of information behind.” Cohort classification is a useful tool of behavioral analysis because it holistically presents the typical consumer, instead of perceiving them as just a sequence of events.
Mattey gave two instances of the intersection of behavioral analysis and machine learning: neural networks and pattern recognition analysis. Deep learning is a multi-layered form of neural network. “People change over time” he said, and deep learning can even capture the behavioral evolution of this change as the cohort ages.
In the field of pattern recognition analysis, data with multiple attributions can be classified into groups having similar characteristics.
Uses in Banking
Behavioral analysis has been used for several years to detect fraud such as identity theft. “The world is increasingly digital. The true customer uses a few devices, which can lead to multiple-point verifications,” Mattey said.
Financial technology companies (fintechs) are “leading this space” on credit risk management and underwriting, he noted.
In the financial advisory capacity, designers implementing “nudge theory” consider default settings as a way to help customers make more appropriate choices.
Clifford Rossi, Professor of Finance at University of Maryland, gave a deeper look at the behavioral analysis in credit card fraud detection. Fraud detection tends to be based on quantitative algorithms “that ingest enormous amounts of customer payment data.” The data will “train” the algorithm on a customer’s purchasing patterns.
For example, the algorithm can pick out anomalies such as “if my bike equipment is purchased in a different city that does not fit in with any of my other purchases,” he said, and thus it would signal “fraud.”
Joe Mattey and Cliff Rossi were two panellists speaking about how banks can leverage behavioral analytics. The webcast on March 13, 2018, was organized by the Global Association of Risk Professionals (GARP). ª
The breadcrumbs visual is from the website Usersnap.