To outpace bad actors, every degree of precision in financial crime intelligence counts. In 2024 cyber threats escalated significantly, with cybercriminals increasingly leveraging artificial intelligence (AI) to enhance the sophistication and scale of their attacks. As noted by James Rundle in the WSJ on November 21, 2024, “Amazon reported detecting nearly 1 billion potential cyber threats daily, a tenfold increase from earlier this year. This surge, attributed to attackers’ adoption of AI, allows them to innovate faster than defenders can respond.”
As bad actors exploit AI to outpace traditional defenses, organizations in the public and private sectors face mounting pressure to adopt advanced tools that provide precise, actionable intelligence and scalable automation. Quantifind has made significant advancements to its Risk Relevance Models to improve the accuracy of identifying relevant financial crime risks while significantly reducing false positives.
Below, we outline the key changes driving these improvements and how they deliver value to our customers.
The Risk Relevance Enhancements and How They Benefit Risk Organizations
- Enhanced Accuracy: The model’s improved accuracy (10% increase in AUC while enhancing both precision and recall) means customers spend less time investigating irrelevant alerts and focusing more on genuine risks. This boosts operational efficiency and reduces costs associated with manual reviews.
- Increased Control Through Better Calibration: The improved calibration allows customers to fine-tune score thresholds with predictable outcomes. This means tailored decision-making based on their specific operational priorities, whether they prioritize broad coverage (recall) or precise targeting (precision).
- Targeted Risk Typologies: By training a model focused on a narrower selection of financial crimes-oriented Risk Factors such as Fraud, Money Laundering, and Bribery, the model is more targeted and effective at identifying nuanced risk signals within this domain.
- Efficiency Gains from Reduced Alerts: A 10% drop in strong confidence, high-risk alerts leads to direct time and resource savings, enabling compliance teams to allocate efforts more strategically.
- Scalability and Generalization: The use of GPT-4o for machine labeling dramatically increases training data, enabling the model to generalize better to new cases. This reduces reliance on constant retraining and improves the adaptability of the system to emerging threats.
- Better Domain-Specific Understanding: Custom word embeddings and the inclusion of 19 new features enable the model to capture the nuances of financial crime-related language and behavior. This provides a richer context and deeper understanding, which translates into higher relevance and reliability of alerts.
Want to learn more about the risk relevance model? Email contact@quantifind.com and let’s talk!