Democratizing Access

Leveraging the power of shared intelligence

• By Damon Madden
• Sep 04, 2020

The COVID-19 pandemic has put financial institutions under more pressure to stay on top of fraudulent activity— as opportunists are looking for any weakness in a system that can be exploited. Moreover, as consumers turn to eCommerce and digital payments while social distancing, there’s no avoiding the increased levels of associated risk for financial institutions.

As organizations prepare for more commerce to be conducted online during the pandemic, sometimes through quickly transplanted or repositioned business models, payment fraud will proliferate. In fact, research from ACI Worldwide has revealed that merchants have experienced significant increases in COVID-19 related phishing activities and friendly fraud, with nonfraud chargebacks up 25 percent in May this year. While overall fraud attempt rates fell from March (5.3%) to May (3.4%), the research shows that the average ticket price of attempted fraud increased by $18 yearover- year. This indicates that fraudsters are getting more bullish and confident in their pandemic-related methods.

For financial institutions, getting the balance right between identifying genuine payments and creating a frictionless customer experience is key. And, machine learning has emerged as an essential tool for detecting fraudulent payments among the many thousands or millions of genuine ones made every day.

THE DOUBLE-EDGED SWORD OF DIGITAL PAYMENTS

Digital payments offer many benefits including better, faster experiences for customers. However, when payments happen in real-time, the window for fraud detection is reduced to milliseconds and the likelihood of recovering fraudulent payments is far lower than with traditional methods. Essentially, as payments get faster so too does fraud — and when the money is gone, it’s gone.

Further, as the volume and variety of digital payments surges so too does the volume and variety of data generated by those payments. Geolocation information, behavioral clues and biometrics provide a wealth of intelligence for financial institutions — but only if they can make sense of the deluge.

As such, machine learning plays a key role enabling financial institutions to operate at the speed and scale required to authenticate genuine payments, catch fraud as it happens, reduce the volume of false positives, and improve the time it takes to react when they do occur. With machine learning, financial institutions can flag activity that deviates from the norm but isn’t necessarily suspicious. For example, when a customer logs in using a different device, it’s less likely to be unauthorized access and more likely that they’ve upgraded their phone – nevertheless, it needs verifying.

To avoid overwhelming already stretched staff, such as call center and support departments, and introducing more friction for the customer, these non-financial transaction scenarios — and thousands like them — need to be digitized, automated and contextualized wherever possible. This is a complex challenge, but getting it right promises to provide an additional layer of competitive differentiation for financial institutions. It opens the opportunity to provide greater fraud coverage and more seamless experiences that can be applied consistently to an organization’s entire customer base, all without additional headcount in either the fraud department or service centers.

Yet to be truly effective in the fight against fraud, machine learning solutions must be agile enough to be developed, tested, deployed and updated, as either new threats emerge, or as existing ones become better understood. And access to an industry-, region- or market-wide view of possible threats – not just an internal one – is essential. It improves the decision-making performance as the “machine” interacts with more data patterns.

DEMOCRATIZING ACCESS TO MACHINE LEARNING

By enabling non-specialists to build, test and deploy machine learning models in minutes, financial institutions can democratize access to the technology. This can be done by abstracting away the complex math that lies behind these models and replacing it with intuitive interfaces that enable drag and drop model building using the ‘features’ of fraud as building blocks. In bringing machine learning to an organization’s in-house data and in-house fraud expertise – as opposed to taking that data and expertise to a machine learning specialist – financial institutions can accelerate the time to market of fraud-fighting applications.

As financial institutions become aware of new fraud risks, additional features can easily be added to the models and the weight of evidence scoring adjusted accordingly, to ensure banks’ defenses keep pace with emerging risks. This can even take place automatically, through adaptive machine learning, where the technology responds to analyst-applied ‘markers’ for potential fraud.

Transforming the way financial institutions use machine learning (by allowing them to adopt a business-led approach) offers greater ownership and control of their fraud detection strategy. It empowers them to act self-sufficiently without the costs, risks and time associated with the involvement of third parties in artificial intelligence implementations, and – importantly for compliance – it promotes better accountability of the solution’s outcomes.

THE POWER OF SHARED INTELLIGENCE

Individual banks already have access to a wealth of data with which to develop machine learning solutions for fraud detection and prevention.

However, when that data is shared across institutions, it has the potential to create a complex and varied intelligence network that can introduce more context to every machine learning decision, exponentially increasing its effectiveness.

This ‘shared intelligence’ empowers unprecedented collaboration in the fight against fraud. By harnessing the power of the community to increase threat visibility and distributing enhanced detection and prevention capabilities back through the community, it creates a powerful jurisdiction- or network-level deterrent to fraud.

Shared Intelligence takes the features of machine learning models deployed by participating organizations and sends them out to a central repository in metadata format. That could be a central infrastructure (CI) or an organization to which the participating financial institutions belong (either as members or connections), where they can be tested against the community view for their effectiveness.

These features are then made available to the rest of the community for members to aggregate with their own models or to build upon as needed.

Unlike a consortium approach, which over-emphasizes its largest members’ experiences of fraud, members can access the benefits of a Shared Intelligence community on their terms. The biggest contributor doesn’t rule the community models and risk scoring criteria.

Thanks to its power to improve the detection of emerging threats through a scaled-up ‘early-heads-up’ approach to feature calculation and contribution, Shared Intelligence is set to be a game changer in the use of machine learning to fight payments fraud.

SHARED INTELLIGENCE IS SHARED COMPLIANCE

The Shared Intelligence approach has the added benefit of allowing regulators and CI owners to understand the wider fraud environment with precision, empowering them to act on new and emerging threats before clusters become endemic financial crime risks. Trends specific to organizations can be tracked and understood at any level required by a regulator, enhancing efforts to combat fraud beyond payments, such as money laundering or identity theft. Further, CI’s can choose to prescribe both the contributing data and time periods to ensure data consistency across the intelligence network.

This can reduce the costs of compliance too for member organizations. First, it resolves the burden and regulatory risks for attempting to extrapolate and submit data externally. Second, if a CI mandates that organizations deploy a particular model, that model can be easily distributed and then run concurrently with their own models. Indeed, an unlimited number of models can be run and tested side by side, on live data without the risk of hindering performance. Suddenly, intelligence sharing to mitigate fraud using machine learning becomes easy with the use of a democratized method.

THE NEXT FRONTIER IN CUSTOMER CENTRICITY

Machine learning allows banks to truly build their payments risk management strategy around the customer, and not around their own channels or other organizational factors that may have little bearing on the customer’s needs. It also serves to ensure that specialist resources like fraud analysts are free to focus only on the activity that’s deemed the highest risk and therefore the highest priority, which machines cannot — and should not — be left to handle.

Machine learning improves the application of payments risk management frameworks and policies to boost risk mitigation (both in terms of fraud and reputational risks) and enhance compliance. It forces organizations to clearly define what their policies are, and the practical steps required to enforce them. And by removing the need for human intervention in the majority of cases, rules and procedures will rarely be bypassed since machines will always follow rules.

This article originally appeared in the September 2020 issue of Security Today.