Skip to content

The Untapped Opportunity of Machine Learning for Real-Time Payments Fraud Prevention

machine learning realtime payments

Artificial Intelligence (AI) is among the buzzwords of the moment, but when it comes to tangible innovations that have the potential to drive rapid ROI, machine learning should be part of every bank or processor’s strategy. No matter the size of the institution.

A recent report from Aite Group, ‘Artificial Intelligence in Wholesale Payments: Five Use Cases,’ highlights the immediate need for payments players to implement machine learning within their fraud prevention strategies and solutions. The emergence of the New Payments Ecosystem has brought both opportunities and challenges for banks and processors, particularly in the form of real-time payments (RTP).

Cleber Martins, Head of Payments Intelligence at ACI Worldwide, explains how financial institutions can leverage machine learning for fraud prevention, compliance and customer experience.

RT: Why does fraud increase when real-time payments are introduced?

CM: Because it's faster time to money! Banks are not the only organizations concerned with Return on Investment (ROI). Fraudsters also run their own calculations on the effort and risk of fraudulent activity against the potential payoff. To achieve the same ROI on other payment types would require more manual effort, potentially involving thousands of transactions, the use of money mules and so on.

Successful real-time payments fraud accelerates access to the actual funds. Irrevocability is a key element of RTP; it’s a great uplift in customer experience for payees, but it’s also good for fraudsters. As an originating bank, you cannot revert a payment once it is sent. However, this should not be misconstrued; real-time payments themselves are not more vulnerable and the fraud is no more difficult to detect than fraud in other channels.

There are lessons to be learnt from the eCommerce space. As European regulators attempt to ‘fix’ Card Not Present (CNP) fraud with Secure Customer Authentication (SCA) mandates, it’s clear that it’s much better to have the right technology in place ahead of launching new payments and services, rather than to try and regulate after the fact. CNP fraud is spread across the volume of transactions and is not particular to any one issuer. In this scenario, fraudsters have structured their operations to deliver maximum effectiveness. But with RTP, the danger is that an individual institution becomes the target of fraud efforts because they did not keep pace with competitors’ innovations in fraud prevention. Fraudsters work on fraud full time! Financial institutions must get ahead of fraud trends before bringing services to market – and they must become faster in their reactions to changing fraud.

RT: What new technologies should financial institutions consider in their fraud prevention modernization strategy?

CM: There’s a common misconception that AI will be a silver bullet to ‘detect more fraud,’ but this is categorically false. The challenge is not in detecting the fraud, it’s in creating operational efficiencies to do this. With legacy fraud prevention strategies, you will generate too many false positives that result in costly manual investigations. The implementation of machine learning for fraud prevention supports organizations in reducing their false positive rate, which drives operational efficiencies and customer satisfaction.

It also increases speed in decision-making, which is critical for real-time payments. With true RTP, the originating bank cannot delay the payment, so it must be confident that it can analyze all relevant data sources and generate a risk score in the pre-transaction window.

Machine learning means that institutions can correlate much more data in a shorter timeframe than they possibly could without this technology. This allows them to be more precise on cases created and actions taken for fraud and financial crime prevention. Actions taken could be pushing a transaction to a risk analyst for manual confirmation with the customer, or flagging transactions for Anti Money Laundering (AML) compliance.

RT: How does machine learning support AML?

CM: With AML, banks are looking to report activity in alignment with regulators’ requests. But they don’t want to report their good customers based on false positive AML alerts. This would be a disservice to both the customer and the regulator. Machine learning allows them to analyze cases and make smarter decisions – and be more precise about which cases require human review. By reducing false alerts, compliance staff can be allocated to operate only on relevant alerts for AML, expending effort in a more efficient way whilst maintaining compliance.

As cross-border transactions get faster and the messaging formats evolve to contain rich data, it’s also critical that banks consider how they are going to process this against Service Level Agreements (SLAs) and compliance requirements such as sanctions screening. Machine learning for data correlation, in combination with strong rules that trigger actions, is the modern strategy for compliance.

The potential fines and business repercussions for AML failures are huge. Confirmed failures in AML can cause severe stock value loss for banks. Although it’s key to remember that banks are not expected to ‘fight financial crime’ simply to comply with the regulation for money laundering and terrorism financing. AML is primarily about the traceability of payments. The increasing speed of transactions – including for cross border – presents a challenge, because it expands the boundaries for any potential fraud implications. Because of the increased number of touchpoints in a correspondent banking chain, it’s a challenge to remain compliant without increasing false positive rates, and without negatively impacting operational efficiency.

RT: What lessons can banks learn from domestic real-time payments to get ahead of fraud threats in cross-border RTP?

CM: The biggest lesson is that organizations must launch payment capabilities in combination with fraud prevention strategies and controls. We know that the speed of access to funds makes RTP attractive to fraudsters, and history shows that the bad guys always target new payment types in the hope that financial institutions have not yet implemented robust fraud prevention. Of course, it’s a balance. Fraud prevention implementations cannot be at the expense of time to market, or compliance with regulatory deadlines. Therefore, it’s critical to have solutions that accelerate project timelines.

The right machine learning solution can fit these requirements by correlating data directly from the payments stream, and using intuitive interfacing to allow fraud experts to combine initial features in models. Banks must prioritize the launch of functioning models against specific payment types to begin training them. And they must also integrate an omnichannel view of the customer into payments intelligence. When it comes to a real-time payments fraud prevention strategy, the online banking channel itself is also a critical piece of the puzzle.

 

Payments Risk Management eBook Machine Learning 

Want to improve customer service and reduce fraud? Download our guide: The Six-Step Guide to Leveraging Machine Learning for Payments Intelligence