Consider this real life scenario: A prepaid reloadable card issuer was experiencing transaction fraud in the range of 1 percent of transaction volume, significantly higher than industry norms. A significant portion of the fraud came from criminals who used stolen credit card or checking account numbers or to reload prepaid cards and then cash out at an ATM or retail outlet before the issuer was able to identify that the funding account had been compromised.
The issuer attempted to moderate this trend by applying a rule-based fraud detection strategy that denied seemingly risky transactions or had analysts attempt to contact the customer and verify the funding source prior to releasing the funds to the prepaid card. While this stopped some of the fraud, the issuer found that too much expensive manual review was needed to identify fraud transaction characteristics and create new manual rules. And by the time that new rules were put into production, the criminals had moved on to different techniques. Additionally, many of the rules only caused legitimate transactions to be denied, or inconvenienced good customers with requests to contact customer service and verify the reload source accounts.
The issuer then applied sophisticated machine-learning technology, combined with manual rules, to address the problem. Data analysis identified that a small percentage of accounts with a few common characteristics had a fraud rate approaching 40 percent of their transaction volume. Other clusters with different patterns also exhibited high fraud rates, while “good” account groups that shared some, but not all, of the characteristics of the high-fraud clusters had virtually no reload or transaction fraud. Common signals were used to discern legitimate users and transactions. In turn, that reduced friction for good users and identify fraud cycles in real time.
By leveraging these insights, the issuer was able to develop new strategies to identify nearly 85 percent of the fraudulent transaction volume that had been approved prior to the introduction of machine-learning models. In addition, the new rules and machine-learning models have reduced manual reviews (and friction for legitimate users) by up to 70 percent.
So what are the key takeaways businesses can learn from this example to make customer transactions smoother while combating fraud?
- Prepaid issuers likely are subjecting good customers to needless friction or verifications while attempting to identify and stop transaction and reload fraud.
- Fraudsters continue to innovate and morph attack vectors so a system based solely on manual rules may find it impossible to keep up. By the time new controls have been implemented, the fraudsters have moved on. In other words, the rules are just as likely to create friction for legitimate customers as they are to mitigate future fraud.
- Machine-learning technologies can detect the subtle patterns and connections between data points in real time and identify fraud attacks as they synthesize. Being able to spot these subtle patterns is often the difference that enable issuers to accelerate legitimate transactions while stopping fraud attempts in their tracks.
- These strategies can not only significantly reduce friction for good customers, but stop fraud and contribute to significant operations savings through a reduction in manual reviews.
You need to recognize and stop fraud, but not at the expense of a good user experience for legitimate customers who will enable your business to grow. Fraud morphs as criminals continue to change tactics and your fraud detection technology needs to stay one step ahead. Combining machine-learning technologies with manual rules provides a way for you to leverage the newest artificial intelligence capabilities to recognize patterns and behavior that eliminate friction for your legitimate customers and enable you to recognize and stop fraudsters.
Robert Stock is head of global business development at Simility, a company which helps financial services and e-commerce organizations detect fraud. Stock was most recently head of business development for CA Technologies’ New Business Innovation Division, where he initiated strategic global partnerships while implementing solutions in authentication, mobile payments and security for banks and payment providers globally. He has also held senior management roles at Intuit and Visa USA.
In Viewpoints, payments professionals share their perspectives on the industry. Paybefore presents many points of view to offer readers new insights and information. The opinions expressed in Viewpoints are not necessarily those of Paybefore.