*By Alessandro Chimera
Banking has gone through several evolutionary cycles over the centuries. Leaving ancient history aside, we pass from the vaults of the Wild West to the Victorian era, with its grand halls and marble facades, to the present day.
In the contemporary era, we have progressed rapidly from cashiers to the first ATMs and, in some countries more than others, drive-thru services that come with optional parking. But even these modern conventions are becoming outdated and outdated.
Ubiquitous, frictionless, seamless
Boosted by their popularity among Gen Z youth, electronic payment systems and the use of banking apps have skyrocketed. Now, we look forward to being able to transact, access money and interact with banking services anywhere, anytime in a secure and frictionless way on a device of our choice.
With the new generation of so-called “challenging banks” not even investing in physical facilities and the specter of cryptocurrencies on the horizon, we need a new approach to protecting digital banking systems and the increasingly digitized economy from the risk of fraudulent transactions.
There are several risks that emanate from the current use of digital transaction channels that are still evolving. As we adopt more use of mobile wallets and banking apps, we naturally increase the speed at which transactions actually occur even further. This inherently makes it more difficult to examine and analyze all transactions to assess their fraud potential.
All of this is happening in a scenario where banks themselves are sitting on years or decades of poorly managed, poorly maintained, haphazardly organized and often unstable data. As disconnected silos of incongruous data are brought together, antiquated organizational barriers can further hamper any chance for an agile, connected business.
Elimination of false positives
A major challenge for organizations operating modern banking systems is the fight against false positives. When transactions are flagged as potentially fraudulent when they are not, systems slow down, services are delayed, customers are frustrated, and eventually they change banks.
False positives are everywhere. The analysis website Global Investigations Review (GIR) estimates that up to 98% of alerts from digital banking systems never result in a formalized Suspicious Transaction Report (STR). This type of activity can lead to fines and reputational damage for banks.
In a world of near-instant payment processing, old-school rule-based fraud detection systems are only as smart as the code they're composed of, which, in this case, isn't smart enough.
There has to be a better way—and there is; we can use anomaly detection.
anomaly detection
As a formal definition, we can say that an anomaly itself is an unforeseen variation or deviation from an expected pattern in a given data set. The branch tells us that one or more input conditions have changed; and this movement outside of what is defined as “normal” can be used to trigger an appropriate response and thus take action against fraud, security breaches or perhaps even operational performance issues.
Old and new financial institutions process large amounts of data in a wide variety of different datasets, databases and data repositories; By using anomaly detection, these organizations can identify transactions that break expected patterns or deviate from previously observed behaviors.
Anomaly detection takes three basic forms: visual detection, supervised learning, and unsupervised learning.
- Visual sensing requires a data analyst, data scientist, or industry expert to study dashboards composed of tables, graphs, gauges, and other data visualizations to look for variations in data. Limited by any specialist's domains of industry knowledge, visual detection is useful, but it depends on human skills and capabilities that are susceptible to failure.
- Supervised learning also involves humans who will work to label a defined relationship of data sets as normal or abnormal. In this case, a data scientist uses data labeled “normal” to create machine learning models that can detect anomalies in unlabeled data. This technique is also useful, but it fails because of the constantly evolving nature of fraudulent threats.
- The most “machine-driven”, that is, the most machine-driven of the three types of anomaly detection – unsupervised learning – analyzes unstructured data in real time using automatic encoders and machine learning algorithms to identify anomalies without human intervention. In modern banking systems, where payment approvals need to be instantaneous, unsupervised learning is especially useful in detecting unknown patterns from massive data sets.
By enhancing anomaly detection with AI, ML, event processing and advanced analytics, financial institutions can detect emerging fraud patterns appearing in real-time data streams. They can then analyze these patterns in the context of the overall transaction history and instantly flag potential fraud indicators in real-time for manual review.
One company that leverages these technologies is Asurion, a leading provider of device insurance, warranty and support services for cell phones, electronics and home appliances. Using an enterprise-class analytics platform, Asurion has estimated a reduction in its fraud dispute rate by up to 50%. In addition, advanced analytics help prevent fraud and risk in your systems and provide a better customer experience.
A combined holistic approach
With so many factors to consider here—and with fraudulent attack vectors multiplying and transforming by the day—it may come as no surprise to find that the most prudent tactics for operational security in this space boil down to a combined holistic approach. What this means is a combination of supervised and unsupervised models brought together through AI/ML, event processing and analytics to provide the most accurate indication of potential fraud.
Adopting an anomaly detection platform to combat fraud in digital banking and financial services means selecting technology that is adaptable, modular and flexible. In this way, organizations can adapt to new fraud scenarios as they arise.
By integrating data discovery and statistical modeling into one solution, finance organizations can create visual tools that collect real-time data from multiple sources and transform it in multiple ways to drill down into an alert.
From the gold of the wild west to the bitcoin of tomorrow, we can now use anomaly detection to keep our vaults safe and our economies more secure. Just remember that while the sheriff is out of town, we still need someone to keep an eye on.
*Alessandro Chimera is Director of Digitization Strategy at TIBCO Software
Notice: The opinion presented in this article is the responsibility of its author and not of ABES - Brazilian Association of Software Companies
EN 
PT 
