When transaction volume goes up, banks and other financial institutions have turned to machine learning and AI to help streamline services. In the UK alone, over 72,000 payments are made by consumers and businesses every minute and this will grow to 79,000 payments by 2025.
If there are now more than 38 billion payments annually, and a plethora of products on the international market, how can banks detect the illegal and fraudulent transactions amid all that traffic? Asks Vikram Gupta, Vice President at Oracle Financial Services.
Early detection is the best strategy for catching out the criminals. Hyper-specific detection rules may stop one route of money laundering, but unless banks use a system that can read and react swiftly to fraudulent transactions, criminals will continue to find new ways to circumvent bank monitoring.
Enter machine learning and AI, which have the potential to reshape the compliance landscape using graph analytics and predictive modeling in both front and middle-office banking operations.
It is an exciting time for businesses, with disruptive opportunities in virtually every market sector. We see three key factors driving AI applications for payments and transaction processing. First, online fraud is becoming more prevalent and damaging. Second, financial services and e-commerce companies are especially vulnerable to sophisticated new attacks. And finally, machine learning is helping organisations combat fraud in ways that were unachievable in the past.
Yet many financial institutions are still undecided as to whether these innovations are worth the risk of overhauling their current systems. The technology makes too much sense to ignore any longer. It could save banks millions of pounds and clean up errant funds allocation at a rapid pace.
The Case for Machine Learning
The argument for integration of these innovations is easily won. The current systems for financial crime identification and reporting are both antiquated and inefficient. Whenever a bank identifies a suspicious transaction, they need to develop an extensive, hyper-specific report for regulators to review. What if AI auto-populated these documents, saving an inordinate volume of man-hours every day? Add this to the time already saved from a non-rule-based architecture and we’re looking at a significantly different financial landscape.
The process of pinpointing money laundering attempts should be straightforward but it tends to be an exhaustive, time-consuming exercise. Today, banks utilise rules-based systems to identify threats. For instance, if a client executes several £10,000+ transactions only then are they flagged as a money laundering risk. If somebody breaches a rule, a non-compliant flag is raised and the customer is investigated further to determine intent.
But one problem with this rules-based approach is that banks end up catching far more good guys than bad because they’re casting such a wide net. Entire corporate divisions, often exceeding 1,000 employees must dedicate their time to separating the wheat from the chaff. This is without mentioning that by the time a rule is perfected, the worst offenders have already found a way to circumvent the identification procedures.
Roughly 80 per cent of compliance costs are spent on people, not technology. This is exactly why AI and machine learning adoption will result in significant savings on revenue and time for key banking departments. These systems produce models that can quickly comprehend user behaviour while utilising predictive analytics to outpace fraudsters at a remarkable pace.
The Road to Integration
Machine learning can correct inefficient systems, but it’s not without its hurdles. The key component here is a tremendous pressure on banks to work with regulators in identifying an amicable solution for both parties that catches more fraudsters and fewer bystanders.
For a bank, these new systems make sense. There is less money spent on evaluating threats and faster execution, but they pose a significant regulatory challenge. When money laundering attempts are reported, the regulatory authority needs to be informed of every pertinent detail of the activity. Utilising a rule-based system allows for full visibility of the audit trail, but machine learning tends to come as a black box solution. In other words, you can see the input and output of a bank’s vetting process, but you can’t see how they got to the final resolution.
While it may seem like widespread adoption of machine learning systems is moving slowly, things are changing rapidly. The benefits are clear and they extend beyond just identification of potential malicious activity.
We’re closer to a new solution every day. But it is imperative that we transition from traditionally black box machine learning solutions to a white box approach. We can’t feasibly expect to rapidly implement cost-effective machine learning initiatives without taking into consideration the granular needs of regulators.
The test of the machine
Could 2017 be the year where unintelligent automation disappears? Banks in the UK (RBS, NatWest and SEB) have already begun to adopt AI to speed up certain customer-facing processes, such as chatbots that help with customer interactions, as well as various back-end applications, like compliance and AML monitoring.
Machine learning tools need to be thoroughly tested to ensure we are reaching an agreeable outcome for both financial institutions and regulatory agencies. The onus is on both regulators and banks to ensure that these technologies and their intrinsic risks are managed.
Soon machine learning in compliance will be a necessity rather than just another option. Over the next year, we predict that machine learning and other branches of AI will present an unprecedented opportunity for banks and regulators to come together in reducing regulatory burden but at the same time proactively preventing more illegal activity.