How machine learning can prevent fraud in a real-time payments environment

What are the most recent developments in machine learning and its benefits for banks, from a fraud perspective?

The development of machine learning has exploded in the last few years. The key areas in which we’ve seen development is split into two categories: first, there is computing power as well as distributed computing technologies, ie reducing the time it takes to analyze, build and deploy models. The increased computing power has allowed developers and data scientists to expand on the complexity and types of algorithms, which can be used to build models. Similarly, the second area we have seen a lot of development in is the field of data discovery and visualization. The ability to rapidly characterize and understand the data signals within a data lake or fraud detection system allows for the democratization of the machine learning process, which enables wider adoption and reduces the need for highly specialized resources to administer and manage the machine learning models.

What are the main types of fraud in banking and what areas of financial institutions are the most affected?

The main types of fraud we see are Digital, Card and Application-based fraud. In the current environment, the card and merchant channels would be the most impacted; however, we are seeing a shift away from card-based crime to digital channels, like internet banking, for example. This shift is primarily driven by the introduction of immediate payment networks globally and, of course, improved card controls such as EMV and contactless.

The emergence of the Authorised Push Payment (APP) fraud types has begun to materially impact the loss line for many financial institutions, and there are far wider reputational and customer experience implications from excessive declined payments through to negative press and regulatory intervention. Ultimately, it is the product line owners who are most affected by fraud.

What are some of the challenges you are hearing from banks around the implementation of machine learning for fraud prevention in real-time payments environment? How does machine learning help them fight fraud in real-time?

The main challenges we are hearing about implementing machine learning are the cost, complexity and transparency of such solutions. The investment required in terms of cost for both technology and human resources is still proving to be a significant barrier of entry. The specialized resources required, such as data scientists, can be very costly given the scarcity of the skill with fraud-specific domain expertise, thus still limiting the ability of many institutions to effectively use machine learning. Similarly, data sourcing, cleansing and aggregation are also still proving to be a challenge. Clean reliable data is instrumental in the effective deployment of machine learning and without this, the return on investment will be hindered.

What are the best practices for preventing fraud in a real-time world? What should banks look for in machine learning and fraud prevention and what are ACI’s real-time fraud solutions offers?

Naturally, an investment in the right tools will enable any organisation to provide secure payment rails while protecting the customer experience. It’s generally accepted that a layered approach is best practice for protecting payments in a real-time world. Such a layered approach is characterized by protecting the different customer interaction points in the payment stream; this includes authentication and customer-initiated events.

When looking to invest in machine learning for fraud detection, several considerations will assist in the purchasing decision. Firstly, consider the level of internal domain expertise required to administer such a solution. Moreover, how the solution interprets the machine learning process into meaningful business intelligence, such as data discovery or explainable outcomes. Lastly, what data signals we need and have available within the detection system to develop and deploy effective machine learning-based detection strategies.

At ACI, we have several machine learning-based offerings available for all levels of expertise. We’ve been actively engaging and listening to the challenges many of our customers have with integrating machine learning into their fraud prevention strategies and in response to these challenges have developed our ACI Model Generator. This tool democratizes the machine learning process through visualized feature discovery, allowing business users to comprehensively assess the effectiveness of each candidate feature. Combinations of these candidate features are then used to train and benchmark models against in-solution historical data, to find the best-fit model. This unique process allows the business to rapidly prototype and update their machine learning strategies in an instantly repeatable and explainable manner, responding to threats in a timely manner as and when they arise with minimal effort and lengthy development periods.

What are the risks that PSD2 is introducing? How can machine learning help?

Dynamic authentication is one of the current concerns, although this approach is beneficial to the overall customer experience and usability of the payments network. It has the potential to introduce both risk and friction to the payment stream. As we know, there is almost always a tradeoff between a frictionless experience and payment risk. Reducing the risk usually implies that additional authentication and screening over the payment channel is required, thus having the potential to negatively impact the customers’ ability to instantly make a payment. To counter the risk, we can leverage machine learning to passively authenticate the payment based on history or digital identity features, to provide a seamless experience without the need for excessive complex and cumbersome rigid authentication processes.

Do you want to learn how to leverage machine learning in your payments fraud prevention strategy? Learn from ACI experts in our webinar on Machine Learning for Fraud Prevention.

About Damon Madden

Damon is Principal Fraud consultant at ACI Worldwide. He has over 12 years’ experience in financial crime management specialising in machine learning and analytics. Damon is based in Dubai and leads the fraud consulting practice for the MEASA region, along with being responsible for the development and delivery of ACI’s fraud specific machine learning and analytics products.

About ACI Worldwide

ACI Worldwide, the Universal Payments (UP) company, powers electronic payments for more than 5,100 organisations around the world. More than 1,000 of the largest financial institutions and intermediaries as well as thousands of global merchants rely on ACI to execute USD 14 trillion each day in payments and securities. In addition, a myriad of organizations utilize our electronic bill presentment and payment services. Through our comprehensive suite of software and SaaS-based solutions, we deliver real-time, any-to-any payments capabilities and enable the industry’s most complete omni-channel payments experience. To learn more about ACI, please visit www.aciworldwide.com. You can also find us on Twitter @ACI_Worldwide.