How Machine Learning Is Putting An End To Financial Crime

The problem is one of the most pressing challenges faced by the industry, but there is a solution


The finance industry is built on data and banks are fast realizing the potential it has to improve the experience of their consumers, create efficiencies, investigate new markets, and manage risk. International Data Corporation (IDC) estimates that banks spent almost $17 billion on big data and business analytics solutions in 2016, and this number is only set to rise as technologically-driven start-ups like Monzo eat into their market share.

While banks are systematically collecting data, however, analyzing the unfathomable amounts they have at the pace required in today’s fast moving world is a challenge beyond the scope of humans. In order to analyze their data as systematically as it is collected, banks are having to turn to machine learning algorithms across the spectrum of their operations, and it is a technology that will fundamentally change banking, making it unrecognizable to today.

This will happen in myriad ways. In a recent Baker McKenzie survey of senior executives from financial companies, 49% said they expect their firms to use AI for risk assessment within the next three years, 29% that their firms will apply AI to learn more about their clients and to prevent money laundering, and 26% anticipate AI will help with regulatory as well as risk and compliance issues.

One of the most pressing challenges for machine learning to deal with in the industry is financial crime. Banks must deal with numerous different forms of fraud and the technology is a solution for almost all of them.

Firstly, it can help prevent money laundering. Banks have been obligated to assist governments in catching criminals hiding their money since the Bank Secrecy Act of 1970. Those that fail face severe punishments. Deutsche Bank, for one, was earlier this year fined more than $630m for failing to prevent $10bn of Russian money laundering and exposing the UK financial system to the risk of financial crime. The New York Department of Financial Services (DFS) also fined the bank $425m, citing one senior compliance officer who said he had to ‘beg, borrow, and steal’ to get the resources to combat money laundering. While numerous others have faced similar fines. But while it is clear Deutsche was not doing all it could, it is an incredibly difficult task to track laundered money as it spreads throughout the financial system. In a recent Dow Jones survey, nearly half of the 800 anti-money laundering professionals who responded said false positive alerts damaged confidence in the accuracy of the screening process. Analysis from Fortytwo Data, meanwhile, revealed banks’ AML divisions are wasting nearly £3 billion a year chasing these false leads.

Machine learning is helping to better distinguish between these false positives and real cases. According to Rahul Singh, president of financial services at IT services provider HCL Technologies, ‘We are already experiencing use-cases of AI and advance analytics in the anti-money laundering function where technology is able to bring false positives down, allowing focused approaches to risk detection and avoidance.’

Essentially, machine learning is able to learn ‘normal’ behaviour from training data and identify anomalous behaviors that could indicate money laundering, such as when money is moved between suspicious geographies, rapid movement of funds between a number of accounts, or invoicing number sequences have been falsified. Where legacy systems relied on static algorithms and criteria, which basically became redundant as soon as criminals changed their behavior, which they inevitably did, machine learning is constantly learning, which means they can identify when the pattern of laundering changes and adapt rapidly.

It is not just in money laundering that machine learning technology is proving to be a real boon. It can also help prevent consumer fraud. In the UK, financial fraud losses across payment cards, remote banking, and cheques totalled £768.8 million in 2016, an increase of 2% on the previous year. Cybercrimes, meanwhile, are costing the global economy nearly half a trillion dollars a year, according to the insurer Allianz, and finance is one of the most obvious targets. As the February 2016 hack of the Bangladesh Central Bank showed, customer accounts can be the most vulnerable point of entry to a bank's systems. The hackers used stolen privileged credentials to steal $81 million before they were caught.

The threat is constantly mutating as criminals adapt to security measures and find new methods of extracting money illegally. Organizations must adapt their security countermeasures rapidly or risk massive losses, both financial and to their reputation. Machine learning can better our understanding of the transactions that flow through the bank by understanding customer behavior. This means that when any anomalies do occur, they can be automatically flagged in real time and appropriate measures taken, whether this be blocking a card or kicking someone out of a system.

Nikhil Aggarwal, Executive Director and Head of FCC Analytics at Standard Chartered Bank, states that:

‘There is an increasing focus across two broad categories: i) people and ii) technology solutions within financial crime risk analytics. Banks are increasingly recognizing the importance of creating a dedicated financial crime risk analytics team and are investing in talent at all levels – ranging from analysts to senior leadership roles.

In terms of analytics, the focus is on technologies that apply ‘smarter’ and ‘more powerful’ algorithms in risk identification, and tools that improve the efficiency and effectiveness of operational workflows. One example of analytics innovation is the optimization of alert generation, and the subsequent routing of these alerts to specialized investigations teams.

Alerts are generated due to unusual and potentially suspicious transaction behavior and represent the first human touch point in a generic assessment of anti-money laundering risk. With advances in analytics, sophisticated algorithms combine unusual transaction behavior with the full history of client transactions and other behavioral patterns to form an enriched and composite view of aggregate risk. The algorithms then rank the overall riskiness of ‘events’, and channel these events to the workflow queues of specialized investigations teams.’

This is not to say that machine learning is without flaws. It is still a technology in its infancy and there are a number of problems. For a start, it is essentially operating in a black box making decisions that humans are unable to understand. It will itself need regulation, which will need to tred a careful line between helping the technology do its job and preventing it from running amok like some financial Minority Report. At the moment, however, machine learning in banking is still at a fairly nascent stage. Adoption is going to rise rapidly over the next year though, and banks must look to the technology as much as possible.


Read next:

Why Blockchain Hype Must End