The new baseline for financial crime

Many credit unions turn to AI and machine learning to solve issues.

April 12, 2021


Credit unions must continually evolve to protect against the growing sophistication of financial crime. Traditionally, rule-based solutions have been used to fight financial crime threats. But using rule-based models alone are no longer sufficient. That’s because there are three main problems with rules:

  1. Rules get stale quickly. Stale rules lead to a lot of false positives, meaning you’re likely to block a lot of good customers and may lose their business.
  2. Rules are not dynamic. Threat vectors are constantly evolving, meaning the library of rules must expand with it. Beyond that, you need to reach out to vendors to continuously update and fine-tune the rules, leading to a costly and time-consuming process.
  3. Rules tend to be inefficient from an investigation perspective. False positives create a lot of noise. So, in practice, only a fraction of anti-money laundering (AML) alerts progress to becoming suspicious activity reports (SARs)—yet teams spend the majority of their days reviewing them. This is a heightened issue for credit unions because they usually don’t have large teams of analysts and investigators.

Credit unions should make the most of their resources and become more proactive in improving efficiency. Many are turning to AI and machine learning to solve issues.

It’s not Skynet—AI can truly improve efficiency

In the past, the idea of a machine autonomously learning and adapting was quite intimidating. To better understand how AI can help, it’s best to look at the two main areas where it most drastically improves efficiency: operations and threat detection.

  • Operational efficiency. AI automates operational processes. Looking at AML and SARs as examples, AI prepopulates these reports with relevant information and case details, meaning staff doesn’t get bogged down by administrative tasks and can focus on meaningful initiatives, prevent losses, and catch criminals.
  • Detection efficiency. AI reveals richer insight into customers and transaction patterns and detects suspicious activity earlier on to stay ahead of evolving threats. These insights can lead to more granular analyses, without requiring more resources.

AI (and machine learning by extension) enable autonomous analytics at scale for every member account to ensure that alerts are only generated for true, risky behavior.

Machine learning: A crucial threat-detecting tool

Machine learning enables monitoring and threat detection with as few resources as possible, which is crucial for managing day-to-day threats. Understand the types of machine learning, the differences between them, and the effort required to ensure these models work effectively:

  • Supervised machine learning. This type of machine learning requires labeled data. The models require training with examples and patterns indicative of the "good," or lower risk, and the "bad," or higher risk. This style focuses on a specific type of threat—meaning it will miss unknown threats not accounted for. Supervised machine learning models are fine when you know what type of financial crime you want to look for. New threat vectors will not be detected because your models aren’t trained to look for them.
  • Unsupervised machine learning. This type doesn’t require labeled data to train it. Rather, it self-learns a user’s behavior and identifies and groups patterns together. By understanding user behavior, it can effectively identify anomalies unique to each user—reducing unnecessary alerts and false-positives and detecting unknown threats.

Fincrime will continue to increase, but the same can’t necessarily be said for the size of a credit union’s staff or budget. Make sure tools and technology empower you to do more with less. Credit unions are quickly adopting AI and machine learning to work smarter—not harder—and stay one step ahead.