First National Bank reimagine risk management using AI

South Africa’s First National Bank is harnessing artificial intelligence (AI) to mitigate financial and regulatory risks – from insider trading and fraud to tax evasion and money laundering. Called “Manila,” the AI application is designed to make it possible to meet regulatory requirements and flag risks more efficiently and accurately.

“As the world changes, the world of financial crimes changes. This demands that we rethink our risk management processes,” explains Mark Nasila, chief analytics officer for FNB Risk. “My responsibility is to develop processes that leverage data and technology to enable proactive and efficient risk management and that keep up with evolving risks.”

In the past, analysts would spend most of their days gathering and reviewing large amounts of financial information in order to make decisions about potential or perceived risk. Since Manila’s launch in August 2019, the time it takes to produce reports has reduced by 70 percent, on average. Similarly, the time taken to generate a forensic synopsis ready for a human analyst to review has decreased from several hours to as little as 8 seconds.

Manila works much like a human analyst would – identifying and flagging any seemingly nefarious financial activities using rules, models and algorithms that were developed in-house at FNB using Python. Once a customer is red-flagged, Manila runs an enhanced due diligence process. The AI software collects and combines a wide range of customer data – from up to 50 sources and including all information around customer spending and other transactional activities within the FNB banking environment.

AI creates more inclusive reports than humans

Manila creates a forensic synopsis based on the gathered financial data and this is used to guide a human analyst’s review, Nasila explains. But Manila doesn’t just bring together data and presents it; this AI platform actually generates a natural language written analysis of findings. And because Manila sources information from hundreds of thousands of data sources, the reports are more comprehensive than anything a human could produce. “Manila looks at financial crime from a holistic perspective. Something that would take a human being a very long time to do,” Nasila says.

This written report outlines cash flows in to and out of accounts, identifies any anomalies and offers behavioural comparisons. It is used to help analysts determine whether or not a customer is high-risk. “Our aim is to make sure that our risk assessments are more robust, consistent and that they keep up with how the market is evolving over time, Nasila says.

The use of this AI system has dramatically improved accuracy and overall efficiency, according to Nasila. But Manila doesn’t remove the need for analysts, he stresses.

Analysts required to train AI on evolving risks

In fact, Manila has created a new role for the bank’s analysts. These individuals are now required so that Manila is constantly being trained around how financial crimes are evolving. This is an example of machine learning: artificial intelligence programs that create models, evolve and become more accurate — in other words, “learn” — as data gets fed to them. During the quality assurance process, analysts make any adjustments or add additional information to the synopsis first produced by Manila. The changes are then fed back into Manila to help it learn new trends or insights that it did not have before.

According to Dr Nasila, the name Manila was actually inspired by boxing. The “Thrilla in Manila” was the third and final boxing match between Muhammad Ali and Joe Frazier. Held in the Philippines’ capital city, the 1975 boxing bout for the heavyweight championship of the world is widely remembered as the greatest fight of all time. Dr Nasila explains that global interest in the bout meant that the famous match acted as a platform to bring the world together.

For FNB, their AI-enhanced forensic due diligence solution is also a unifier. This platform brings together partners, regulators, serrvice providers and customers to make it easier to detect and mitigate fraud.

“Manila started out as a proof of concept but we quickly realised that if we want to keep up with the way the world is changing we have to come up with ways to use technology to reimagine our services and improve as we go along,” Nasila says. “We are already scaling our experiences with Manila across other use cases.”