Swift’s Avalon Ingram says payment security relies on AI and collaboration to detect anomalies, strengthen risk management, and anticipate financial threats.
As the financial ecosystem continues to evolve rapidly, the importance of adopting AI-driven solutions becomes increasingly clear. For financial crime especially – where threats are reaching new levels of sophistication – leveraging AI to analyse vast datasets helps financial institutions pinpoint irregularities in payments, significantly bolstering fraud prevention and transaction security. The focus for the community is not just improving efficiency, but constructing a future where the financial landscape is both secure and innovative, and speed of payments doesn’t coincide with a lack of protection from financial crime.
As new players emerge in the payments space, this focus only accelerates. While new entrants and the diversification of payment options is widely celebrated, it broadens the attack surface for financial criminals, necessitating a proactive approach to security. As the industry continues to change, it’s clear that AI and pattern recognition will not just be tools – but the foundation of a more secure and resilient financial ecosystem.
Navigating the Complexities of Anomalies with Flexible Detection Models
Anomalies are deviations from typical behaviour and represent a small subset of information within massive datasets, therefore leading to the risk of imbalance. For instance, institutional fraud makes up a very small fraction of fraud cases — the question is, can we leverage an imbalanced dataset to extrapolate and contrast between normal and abnormal transactional behaviour? While unsupervised AI can determine learning patterns from datasets without human intervention, the lack of insight opens the door to unmonitored AI hallucinations, where businesses cannot explain their system’s decision making. This harms financial institutions’ credibility and reputation.
To overcome this challenge, financial institutions can train AI models on specific transaction data so it can then determine threshold values, and identify anomalies in payments such as what should be analysed and what is considered abnormal based on available data. This approach is dynamic and overcomes technical challenges that rules-based approaches and models face. However, it requires more technical knowledge, and data quality issues like potential false positives and unforeseen patterns of fraudulent behaviour must still be addressed and monitored regularly.
Combining Detection Approaches to Overcome Blind Spots
Enhancing detection capabilities can be achieved by combining rules-based and supervised algorithms. The rules-based approach identifies relevant datasets, while supervised AI can be focused solely on detecting outliers. Both offer explainability as they require human intervention and expertise to be designed. However, several challenges and hurdles remain in place, preventing financial institutions from conducting effective anomaly detection.
One significant challenge in anomaly detection is the siloed nature of data within financial institutions. Financial institutions often have a limited view of transaction data, restricted to what they send and receive, hindering the detection of anomalies that might be apparent in a broader dataset.
Furthermore, certain anomalies can only be revealed in federated datasets, which combine information from multiple institutions. For instance, if an account is suspected of fraud, sharing data across institutions can provide a more comprehensive view of the account’s activity, revealing potential risks that might not be visible to a single institution. Currently, there are limited legal frameworks that allow for the sharing of such data, which presents a significant hurdle.
Existing frameworks are often limited to single jurisdictions, where federated datasets hampered by geographical and legal boundaries cannot offer their full value. Without the ability to share data about potential fraud, financial institutions are left to tackle financial crime with incomplete information.
Contextualising Fraud for Stronger Risk Management
With access to billions of transaction insights, Swift is working with the financial community to see how raw data can be transformed into powerful tools and actionable insights for financial institutions. This empowers institutions with broader datasets that significantly enhance their fraud detection systems, uncovering anomalies that might be missed from previous levels of visibility.
Swift has developed five markers of anomaly which can be used independently or in combination. These include identifying abnormal volumes and values of payments sent or received by an account or exchanged between a pair of accounts that may warrant deeper analysis. This provides financial institutions a crucial insight into account numbers linked to payments that are those rejected, recalled, returned or aborted on the Swift network. To further enrich the context of anomaly markers, anomaly scoring is also employed for financial institutions. The anomaly scoring detects patterns by combining the markers together, bringing an additional insight to financial institutions.
This pre-emptive step integrates seamlessly into financial institutions’ existing detection processes. Even while a payment is in motion, institutions can monitor transactions in real-time, potentially alerting or blocking abnormal activities. The insights received post-transaction is used to aid trend analysis, risk monitoring, and post-fact examinations, providing a layered defence mechanism for precise risk management.
The Future of Anomaly Detection
The future of anomaly detection in payments hinges on the integration of advanced AI models within collaborative frameworks. Innovative anomaly detection frameworks should strengthen risk management, and enable financial institutions to anticipate potential threats with greater precision. However, achieving the next level of effectiveness will require industry-wide collaboration to develop common standards and frameworks for interoperability. The path ahead is clear: sophisticated AI and industry collaboration will be the bedrock of the future in anomaly detection, ensuring a safer and more secure financial landscape.
—
Avalon Ingram is Head of FCC Experts at Swift, where she oversees an expert group dedicated to delivering innovative data products and services that enable clients to manage their financial crime compliance risks and obligations.