How to reduce false positives with machine learning (5 tips)

Compliance software enables financial institutions (FIs) to detect a range of criminal threats, including increasingly sophisticated money laundering and terrorism financing methodologies. However, in broadening the scope and capabilities of anti-money laundering and countering the financing of terrorism (AML/CFT) programs, software solutions run the risk of increasing the false positive alerts they generate, misidentifying customers, and incorrectly flagging transactions as money laundering or terrorism financing risks. 

A false positive occurs when a customer profile or transaction appears suspicious under the parameters of a firm’s compliance solution, triggering an AML alert. False positives can occur for a number of reasons, including inaccurate data, inadequate context, excessive rule sensitivity, and human error. False positive alerts represent a significant time and cost drain for FIs since they must remediate each case to meet compliance obligations. At the same time, the associated administrative friction creates negative customer experiences. 

One of the most effective ways to address this is to reduce false positives using machine learning (ML) within a compliance solution. ML tools enable firms to interpret data more clearly and efficiently to reduce false positive rates. Firms should understand how to use ML for this purpose and enhance their AML programs.

How does machine learning work in AML compliance?

The data collection requirements for reducing false positives make machine learning-powered tools particularly valuable for compliance teams. ML systems analyze historical data to produce insights on customer behavior over time. They then use that data to make intuitive decisions about emergent alerts and predictions about future outcomes. 

It’s important to note that ML is distinct from AI, both in general and specifically within AML compliance. AI is an umbrella term referring to a number of technologies that perform functions traditionally associated with human intelligence. ML is a category within AI, and refers to the ability of a machine to improve its capabilities by analyzing and drawing patterns from data. Within AML, machine learning systems can be programmed to adjust their outputs automatically, generating new data points without needing further input or direction from compliance teams.

With the benefit of machine learning, firms can train their AML screening solutions to identify and collate relevant sources and structure the data more effectively for the alert remediation process. Historical information can be referenced quickly and efficiently to inform decisions about incoming alerts, while newly released data can be used to enrich customer risk profiles in real-time. ML is a significant aid to screening measures because firms can train their systems to improve their accuracy over time by learning to notice things like:

• Duplicate names.
• Similar spellings. 
• Nicknames and aliases.
• Non-Latin scripts.
• Differences between global naming conventions.

5 ways to reduce false positives with machine learning

ML tools expedite the remediation process for AML alerts, identifying false positives faster than other types of analysis and escalating true positives where necessary. Specifically, firms can use ML systems to help reduce false positive rates in the following ways:

1. Tailor matching algorithms to AML requirements

Matching algorithms are key to any firm’s ability to conduct effective AML screening. To avoid false positives and maximize efficiency, organizations should customize their algorithms according to factors such as jurisdictional regulatory requirements, the products or services they offer, and their own risk appetite as defined by an initial business-wide risk assessment. 

2. Continuously monitor information sources 

The data FIs use for essential compliance processes is dynamic, not static. Consider, for example, the frequency with which sanctions lists or politically exposed person (PEP) lists are updated by the authorities who maintain them. Firms should ensure they can capture these changes as swiftly as possible by using ML to continuously monitor and update their data rather than manually and on a fixed schedule, which otherwise creates a risk of using old information. 

3. Use the cleanest data possible

Similarly, many false positive alerts are generated by redundant data, often involving outdated information or improperly matched names. Machine learning systems can be trained to recognize redundant data by semantic context to streamline alert remediation. Similarly, machine learning systems can be programmed to perform statistical analysis on both historical and emergent transaction data to help establish the likelihood of a false positive alert classification. 

4. Structure the data 

False positive remediation involves analyzing vast amounts of unstructured data drawn from external sources such as media outlets, social networks, and other public and private records. Machine learning systems can help firms better structure that data, learning to prioritize and categorize information based on its relevance to particular alerts. 

5. Implement intuitive screening 

False positives often occur during PEP and adverse media checks or checks of international sanctions lists due to misidentification of names or misinterpretation of data. In this context, ML can enrich customer risk profiles by intuitively providing additional identifying information or clarification over naming conventions to help compliance teams distinguish false positives.

Machine learning solutions for AML compliance

ComplyAdvantage’s software solutions adopt cutting-edge AI to power matching algorithms for FIs to use in their AML compliance programs. Drawing on dynamically updated databases to minimize false positive alerts, these algorithms can quickly and accurately screen customers against sanctions lists, watchlists, adverse media, enforcement data, PEP lists, and data on their relatives and close associates (RCAs) data. 

ComplyAdvantage has also developed ML models to recognize and eliminate irrelevant data by performing semantic and statistical analyses on new alerts, reducing the false positives that often spring from duplicated or redundant data. Once true positives have been identified, compliance teams benefit from automated risk assessments to tackle the highest-priority cases first.