Blog

Künstliche Intelligenz und Regeln für intelligentere Warnmeldungen kombinieren

12.20.2021 | SymphonyAI team
 

The most common tactics used by financial institutions to detect illicit activity are rules-based detection scenarios applied by transaction monitoring systems. These rules are straightforward and not complex in nature, with some being necessitated by OCC regulations. When a transaction triggers one or multiple rules, an alert is created and passed to an investigator who must decide if it is a false positive, worth investigating, or escalating to the authorities. Since rules are simplistic in nature, they tend to create a lot of noise (false positives) and investigators are required by law to investigate every alert that gets generated. This causes high operational cost and investigator fatigue as they have to sift through 97% of noise to find that 3% of truly suspicious behavior. And unfortunately, the rules in place today do not cover all the risk typologies or nuances used by today’s sophisticated criminals.

Advances in artificial intelligence (AI), machine learning (ML), and scalable technology now make it possible for financial institutions (FIs) to implement advanced detection capabilities versus rules alone. The success and capabilities of this advanced technology have been analyzed extensively by research institutions like Aite-Novarica Group, and their report can be read here. According to this report, “Legacy transaction monitoring systems often generate excessive false-positive alerts, leading to high operational costs and wasted investigative effort.” They go on to state, “Many current AML (Anti Money Laundering) practices and technologies are becoming outdated. FIs must modernize their AML control frameworks to deliver more actionable intelligence to law enforcement.” Given the stringent regulatory requirements in the financial services industry, AI and ML are only as good as their explainability to support investigators in deriving meaning and process for their investigations. Without cogent and understandable alert risk factors and a means by which investigators can logically conduct their investigations, advanced AI is just a black box that will never pass regulatory scrutiny. 

The solution then is a blended approach of combining rules-based detection and machine learning to provide improved risk coverage, reduced alert volumes, and full explainability of resulting alerts. Curious how that works? Here are a few samples.  

  1. Risk Similarity: The model looks at prior L3 reviews and SAR (suspicious activity reports) filings to determine risky behavioral profiles based on current behaviors. 
  2. Behavioral Models: Significantly improve data yield, enabling FIs to identify suspicious behavior up to 12 months earlier than their existing legacy systems. 
  3. Alert Disposition: By distinguishing between low-and high-risk customer behavior, dynamic risk profiles can drive down false-positive alerts, improving operational efficiency. 
  4. Anomaly Detection: unsupervised learning techniques scrutinize over 500 behavioral features to identify clusters of customers who share similar behavioral, transactional, and other characteristics. Within these clusters, AI spots those customers with unusual activity and risk-scores them against other parties within their specific behavioral group. Identified anomalous behavior supports the discovery of previously uncovered risks. 

All these tactics are working in conjunction with the rules put in place by FIs to detect illicit activity. The proper combination of rules and explainable AI models provides human-readable explanations as to why certain behaviors or activities are suspicious and what risk factors were combined to create an alert. This level of detail expedites the investigation process and provides the authorities with the detail and accuracy these institutions need to catch financial criminals.  

Our years of experience at some of the largest global banks in the world tell us that the best move forward when evaluating a new detection system is to ensure that it can incorporate rules-based detection as well as AI, ML, and advanced analytics in an explainable and transparent fashion. It is safe to assume that rules are not going away in the foreseeable future and financial criminals are only getting smarter by the day. Next-generation detection systems like AI and ML are imperative in evening the odds for FI’s to catch illicit activity taking place within their organization. As the adoption of such techniques becomes more widespread, FIs will be empowered to achieve compliance and drastically improve the effectiveness of their fraud and anti-money laundering discovery capabilities.  

The most common tactics used by financial institutions to detect illicit activity are rules-based detection scenarios applied by transaction monitoring systems. These rules are straightforward and not complex in nature, with some being necessitated by OCC regulations. When a transaction triggers one or multiple rules, an alert is created and passed to an investigator who must decide if it is a false positive, worth investigating, or escalating to the authorities. Since rules are simplistic in nature, they tend to create a lot of noise (false positives) and investigators are required by law to investigate every alert that gets generated. This causes high operational cost and investigator fatigue as they have to sift through 97% of noise to find that 3% of truly suspicious behavior. And unfortunately, the rules in place today do not cover all the risk typologies or nuances used by today’s sophisticated criminals.

Advances in artificial intelligence (AI), machine learning (ML), and scalable technology now make it possible for financial institutions (FIs) to implement advanced detection capabilities versus rules alone. The success and capabilities of this advanced technology have been analyzed extensively by research institutions like Aite-Novarica Group, and their report can be read here. According to this report, “Legacy transaction monitoring systems often generate excessive false-positive alerts, leading to high operational costs and wasted investigative effort.” They go on to state, “Many current AML (Anti Money Laundering) practices and technologies are becoming outdated. FIs must modernize their AML control frameworks to deliver more actionable intelligence to law enforcement.” Given the stringent regulatory requirements in the financial services industry, AI and ML are only as good as their explainability to support investigators in deriving meaning and process for their investigations. Without cogent and understandable alert risk factors and a means by which investigators can logically conduct their investigations, advanced AI is just a black box that will never pass regulatory scrutiny. 

The solution then is a blended approach of combining rules-based detection and machine learning to provide improved risk coverage, reduced alert volumes, and full explainability of resulting alerts. Curious how that works? Here are a few samples.  

  1. Risk Similarity: The model looks at prior L3 reviews and SAR (suspicious activity reports) filings to determine risky behavioral profiles based on current behaviors. 
  2. Behavioral Models: Significantly improve data yield, enabling FIs to identify suspicious behavior up to 12 months earlier than their existing legacy systems. 
  3. Alert Disposition: By distinguishing between low-and high-risk customer behavior, dynamic risk profiles can drive down false-positive alerts, improving operational efficiency. 
  4. Anomaly Detection: unsupervised learning techniques scrutinize over 500 behavioral features to identify clusters of customers who share similar behavioral, transactional, and other characteristics. Within these clusters, AI spots those customers with unusual activity and risk-scores them against other parties within their specific behavioral group. Identified anomalous behavior supports the discovery of previously uncovered risks. 

All these tactics are working in conjunction with the rules put in place by FIs to detect illicit activity. The proper combination of rules and explainable AI models provides human-readable explanations as to why certain behaviors or activities are suspicious and what risk factors were combined to create an alert. This level of detail expedites the investigation process and provides the authorities with the detail and accuracy these institutions need to catch financial criminals.  

Our years of experience at some of the largest global banks in the world tell us that the best move forward when evaluating a new detection system is to ensure that it can incorporate rules-based detection as well as AI, ML, and advanced analytics in an explainable and transparent fashion. It is safe to assume that rules are not going away in the foreseeable future and financial criminals are only getting smarter by the day. Next-generation detection systems like AI and ML are imperative in evening the odds for FI’s to catch illicit activity taking place within their organization. As the adoption of such techniques becomes more widespread, FIs will be empowered to achieve compliance and drastically improve the effectiveness of their fraud and anti-money laundering discovery capabilities.  

Neueste Einblicke

ACAMS Vegas 2024
 
10.30.2024 Blog

ACAMS Las Vegas 2024: AI Takes Center Stage in Financial Crime Prevention

Finanzdienstleistungen Quadratische Ikone Svg
 
10.29.2024 Webinar

Untapped potential: advancing model maturity, predictive AI, and generative AI deployment to transform financial crime compliance in Asia

Finanzdienstleistungen Quadratische Ikone Svg
2024: A Year of Pivotal U.S. AML Regulatory Change
 
10.23.2024 Webinar

2024: A Year of Pivotal U.S. AML Regulatory Change

Finanzdienstleistungen Quadratische Ikone Svg