How is artificial intelligence (AI) changing the world? For many of us, the last few years have seen massive upheaval and change, as employers and employees rush to adapt to new and emerging technologies that support their work.x
While much of the focus on AI innovation has been in the realm of custom service support – using large language models to augment and assist in workflows, there’s a realm of financial services that is often overlooked – that, of course, being the cross-functional teams that work together to deal with business risks, such as fraud.
Fraud can, by design, be difficult to track, with the behaviours of bad actors constantly changing to leverage design gaps in complex systems. Professionals, such as those who have completed a graduate certificate in data science online, are working hard to detect fraud. Could the rapid proliferation of machine learning in the financial services sector begin to do what ordinary workers may struggle to do: detect fraud at scale?
Big Business = Big Data
Let’s face it, banking is a massive part of our day-to-day lives. Money is an indelible part of the systems in which we operate; gone are the days of bartering; instead, the humble pineapple does the talking.
No matter whether you’re paying the mortgage, spending money on a card, buying the vital things that you need to support your life, everything, ultimately, can be boiled down to transactions. In 2023, the Reserve Bank of Australia issued a report that found the average person made approximately 730 digital transactions during the 2022/23 financial year, about two per day. For any institution, the transactions that an individual makes represent a massive amount of data to sort through, interpret, and detect when problems arise.
Consider this: if the average person makes two digital transactions per day, that means over a year, there are upwards of ten billion payments processed by financial institutions around the nation, from card operators, to banking institutions, to transaction terminals alike. That’s an immense amount of data to interpret, understand, and report on, even for experienced financial professionals.
Fraud poses a significant challenge for financial institutions, with an estimated $677.5 million in detected fraud over the same period. While this represents a small amount of the overall amount of money flowing through the financial network (accounting for approximately 64.2 cents for every $1,000 spent), it represents a problem worth hundreds of millions of dollars to the financial sector, a challenge that continues to grow each year.
Overcoming Scale Problems with Machine Learning
In the realm of fraud detection, there are a number of methods that are traditionally used to detect and respond to suspect transactions. The most common of these is known as the business rules system, wherein banks establish a number of distinct rules that identify suspicious transactions.
While the rules-based system may have been effective in the past, in modern times, it’s becoming less relevant to fraud detection. This is because bad actors are always looking for ways to exploit loopholes, backdoors, and vulnerabilities, and as a result, are constantly refining their strategies to defraud banks and customers alike.
Constantly adding and changing fraud rules can pose challenges, too; for example, if said rules leak, a fraudster may be able to understand precisely what they can or can’t do, reducing the effectiveness of fraud rules and increasing a bank’s susceptibility to fraud.
In an ideal world, having a system that can create highly refined rules (incorporating individual customer behaviour) and work with millions, or even billions, of transactions, while running efficiently and accurately, is ultimately the best practice goal of many organisations. That’s where a subset of AI models, known as machine learning, comes in.
Think of machine learning like having an army of analysts in the palm of your hand, being able to rapidly assess and make decisions, not only on an individual level, but also being able to review and compare payments at a much larger scale. By incorporating not only transactional data but also activity records from a customer, a machine learning model can gain a very good sense of a customer at the individual level.
For example, a machine learning model could leverage a customer’s transactional records to identify their average spend, basket size, and number of refunds, thereby determining whether a transaction appears legitimate or fraudulent. A customer who orders frequently and records a high rate of refunds, for example, may be considered suspicious, compared to a customer who orders infrequently.
Machine learning models can also look beyond transactions, flagging suspected fraudulent addresses or contact details. Where an analyst may only have the capacity to work with a small number of datapoints, a machine learning model can augment and support their work, acting as an early warning beacon for suspicious transactional clusters.
Fraud Detection in Practice
How does machine learning work in practice? Depending on the type of model executed, machine learning algorithms can work in many ways to identify fraud. For example, supervised machine learning models may utilise an input of agent-declared legitimate and fraudulent transactions to help start to identify suspicious transactions.
For example, banks such as ANZ are exploring the role of transaction scoring – scanning clusters of financial data and assigning a risk score, assisting in identifying customers that may be in the early stages of financial stress. Other algorithms are also being used to understand customer behaviour, so that if a password is compromised, a bank can identify where login activity may be suspicious and needs to be locked down.
Fraud detection can happen in many ways, and for operational reasons, many banks choose not to publish their strategies. It’s clear, however, that protecting customers from fraud remains a top priority, particularly in the fight against scammers.
The Fight Against Fraud
The fight against fraud is complicated, as one door is closed, another is forced open.
Seasoned professionals often describe the fight against scammers and fraudsters as similar to the battle to protect one’s home during a cyclone or bushfire, as nature rages around the property, and the homeowner does all they can to protect themselves. Sometimes, a bit of persistence and luck can make all the difference.
As new machine learning algorithms make their way into the financial sector, it’s exciting to consider how the next generation of banking professionals will protect customers from fraud. Who knows – perhaps in ten years, banking fraud may be a thing of the past, monitored and managed by the very things we’re developing to protect consumers today?