How Automated Alert Hibernation Frees AML Investigators to Pursue Fraudulent Transactions
05/25/2021 by Calvin Crase Financial Crimes
Over the past several years, financial institutions have grown increasingly vulnerable to fraud and financial crimes. Online and cross-border international transactions have grown exponentially, which introduce complications as regulations shift toward anti-terror tactics. Plus, the COVID-19 pandemic accelerated the shift toward digital banking. Users have flocked to online transactions and non-cash internet payments, which are executed fast and without in-person interaction.
As a result, your institution’s anti-money laundering (AML) and regulatory compliance efforts are squarely in the spotlight. In response, banks and payment processing organizations have dedicated new resources toward fraud detection and regulatory compliance. But the vast majority of the results generated by your AML solution are false positives. Consequently, instead of investigating and preventing illegal activity, analysts are mired in an avalanche of busy work with transaction data.
However, by implementing best practices in your AML solution, you can reduce your bottleneck of false positives through automation. Rather than asking analysts to sort through a haystack to uncover each productive alert, machine learning provides the triage capabilities your team needs.
Your AML solution is configured to flag any transactions that fall outside of your institution’s acceptable risk level. If one of your bank’s customers attempts to deposit $10,000 spread across multiple branches in a given day, that activity will generate an alert.
Your analysts consult the AML solution’s findings and determine whether the transaction merits opening a suspicious activity report (SAR) or is a false positive. But in addition to applying a basic, binary conclusion, many banks’ best practices include a third category that sets aside or “hibernates” the alert for future review.
Hibernated transactions may show suspicious behavior but fall within an acceptable risk category based on the customer’s prior behavior. When a transaction has been hibernated, the incident remains open for response if the same customer generates further alerts. If the number of suspicious transactions reaches a number defined by an organization as above its risk threshold, then an analyst opens a report.
Unfortunately, even after a transaction has been hibernated, each subsequent incident tied with a customer requires manual review. After your AML solution reviews a weekend of transactions, your analysts spend the better part of Monday clicking through a bunch of bulk alerts instead of conducting actual investigations.
To keep up with all the data, institutions wind up hiring more analysts rather than finding a way to maximize their time and effort.
Instead of asking your analysts to robotically process a bank of false positives, you can enlist AI and machine learning to streamline the hibernation process.
Recent technological advances have inspired increasingly sophisticated applications for data analysis, algorithms, and modeling. But the banking industry has long relied on human judgment for fraud prevention.
Through proper segmentation of their customers and applying appropriate thresholds, financial institutions can ease the burden on their analysts. First, a bank needs to group its customers into specific behavior patterns and risk profiles. You can define these segments by account status (business versus personal). Or, you can apply customer details such as income level, occupation, or average number of monthly transactions.
Once your institution has segmented its customers, you can then apply specific risk thresholds to each group. If a segment is a low risk, you can allow a higher number of hibernated alerts for each customer. Then, once that threshold is exceeded, the customer merits investigation.
With these conditions in place, you can apply patterns in your customer behavior to automate the decision to hibernate alerts.
Machine learning is built upon using trends and patterns within your data to generate conclusions. With a model in place composed of segmented customer populations with defined risk thresholds, the right data partner can incorporate an algorithm into your AML solution to automate the hibernation decision.
Rather than requiring your analysts to review individual records, an algorithm applies a risk assessment for each of your institution’s new transactions. By leveraging the patterns in your analysts’ historical responses to alerts, the algorithm establishes a baseline that informs its decision to hibernate or escalate an alert.
After review, the algorithm then hibernates or escalates a transaction in accordance with your institutional requirements. If a transaction indicates a strong likelihood of fraudulent behavior, the algorithm sends it to your analysts as an actionable alert. Or, if your customer is within a threshold of risk — such as 5 hibernated alerts — then the transaction is automatically hibernated.
With machine learning incorporated into your business process, analysts no longer need to be stuck manually dismissing false positives. Plus, with a predictive model in place, your team only responds to a hibernated alert once it reaches a high enough priority to merit their time.
Your organization’s ability to harness the capabilities of advanced analytics depends upon a trustworthy data source. Many institutions must manage fragmented, inconsistent, or error-prone data as a result of mergers and acquisitions or information siloed between departments.
If your business has not undergone a digital transformation to centralize its data, the prospect of automated alert hibernation provides further incentive. Along with reducing false positives, you also save money as your AML investigation team grows more efficient and cost-effective. Plus, machine learning increases your customer risk scoring capabilities by automating your effort to triage transactions.
Automation can reduce your investigators’ workload by reducing false-positive alerts. However, human review is still essential to accurately determine whether a transaction is fraudulent. Even then, efforts to prevent illegal activity will still be imperfect.
Once your alert hibernation model is in place, you can adjust its performance to ensure its results are consistent. Given the realities of the industry, AML is sure to remain a work in progress. However, with the addition of automation and predictive analytics to your arsenal, that work grows far more manageable.