Anti-fraud technology with a human touch

The use of artificial intelligence and machine learning in bank fraud analytics is continuing to move from reactively mitigating fraud that’s already occurred to preventing fraudulent activities from actually happening—but in ways that try not to block legitimate customer transactions.

As anti-fraud technology has become more advanced and scalable, some banks are now investing in a cross-product, omnichannel view of customer behavior, says Philippe Guiral, who leads Accenture’s North America fraud and financial crime practice. This means leveraging customer data across domains within the organization to gain more insights of customer behavior to better assess whether any particular transaction is suspicious.

A growing number of banks are now building cases to show these solutions can not only improve fraud prevention rates, but also enhance the customer experience and be applied across additional functions—including financial crime, ‘Know Your Customer,’ risk and customer intelligence—to uncover hidden risks and discover new opportunities, he says.

Sign up for the free BAI Banking Strategies newsletter and get industry insights delivered to your inbox.

Indeed, it’s critical to have a strong fraud analytics solution that can give banks a comprehensive view of a customer’s identity and real-time insights into application activity, says Kimberly White, senior director of fraud & identity at LexisNexis Risk Solutions in Alpharetta, Georgia.

“We proactively prevent fraud using a multilayered, risk-based approach employing technology solutions that incorporate identity verification and fraud analytical models, combined with multifactor authentication to help prevent fraud and improve the overall customer experience,” White says.

Likewise, FICO has been using been using AI and ML to solve the “precision challenge” in an effort to reduce false positives, says Liz Lasher, Miami-based vice president of portfolio marketing for fraud and financial crimes.

Bank strategic teams are also converging three domains—identity management, fraud and financial crimes—to creat economies of scale and make more contextual decisions that result in consistent customer experiences, operational efficiencies and reduced false positives, she says.

IBM Security is enhancing real-time analysis to prevent customer identities from being stolen. This effort includes employing a combination of ML models with behavioral techniques, says Aarti Borkar, vice president of product and strategy, who is based in San Francisco

“We’re able to recognize patterns of behavior for users who regularly log into an application, such as a banking app, including whether they hold their mobile device horizontally or vertically, the way they type their password in a particular way and mouse movements,” Borkar says. “We combine those behavior patterns with machine-learning models to build a risk model to better assess the identity of the individual using the bank app.”

Mitigating fraud has its place, too

While AI and ML solutions are getting better at proactively stopping fraud, sometimes mitigating fraud is more prudent, the experts say.

“It’s not always the best course of action to prevent a transaction from happening, but rather to allow the transaction to happen and then investigate after the fact,” Borkar says. “That way, they don’t block transactions that are legitimate, and the investigation also gives banks details to figure out how to prevent the fraudster from attacking again.”

Lasher agrees, saying that while prevention is a critical goal, “mitigation is also important, as banks want to strike a balance between preventing fraud and customer experience, by lessening false positives. Banks need to establish trust and consistency with their good customers, so that they can stay the preferred brand.”

While reliance on technology continues to grow, humans still play a critical role in fraud analytics, from designing AI/ML models to determining whether an activity is actually fraudulent or legitimate. One such role is building comprehensive algorithms to ensure there are no blind spots, such as not flagging fraud that has been happening in one geography but not yet in another, Borkar says.

“Humans are also needed to review analytical results to make decisions on what to do next,” she says. “Finally, everyone within an organization needs to be educated on how to support fraud prevention, and follow the rules set up by their security teams.”

Katie Kuehner-Hebert is a BAI contributing writer.

Learn more about how to prevent fraud in the BAI Executive Report “Taking the fight to the fraudsters”