As financial institutions work to combat financial crime, their interest in the use of AI technology in anti-money laundering (AML) processes continues to outpace implementation, says data science and analytics company SAS.

Based on a global survey of 850 members of global organisation the Association of Certified Anti-Money Laundering Specialists (ACAMS), the study reveals that the adoption of AI and machine learning (ML) remains modest.

Only 18% of survey respondents report having AI/ML solutions in production. Another 18% are piloting AI/ML solutions, 25% plan to implement AI/ML in the next 12 to 18 months and 40% have no current plans to adopt AI/ML.

The study includes contributions from SAS's partner organisation, audit, tax and advisory services firm KPMG.

“AI and ML aren’t a magic fix for every financial crime challenge. But they are showing to be increasingly effective in certain areas, especially those involving large amounts of data,” says KPMG International Global Fraud and Financial Crime Transformation lead and KPMG in Germany partner Timo Purkott.

“That includes automating alerts from transaction monitoring, generating enterprise-wide risk assessments, reporting suspicious activities, AML checks, striving to reduce false positives and more. It all depends on data. Organisations must invest in their data management infrastructure to maximise the value of AI and ML and stay ahead of financial criminals,” he says.

“The survey indicates that AML practitioners believe regulators have cooled on AI,” notes ACAMS chief analyst and editorial content director Kieran Beer.

“Among AML practitioners, 55% said their regulator promotes or encourages AI/ML innovation, which is a 15-point drop from the 2021 survey. Those who said regulators are apprehensive or cautious about AI/ML adoption rose to 36% from 28%, and those describing regulators as resistant to change more than doubled to 13% from 6%,” he points out.

Further, the study shows that interest in generative AI (GenAI) technology is robust but the industry remains cautious.

Almost half, or 45%, of respondents are considering GenAI technology, with 10% of respondents saying they are currently piloting GenAI solutions and 35% saying that are in the discovery phase. However, 55% have no plans to adopt GenAI, SAS says.

The survey produced a number of insights on how AI technology is being used in AML and why companies may be slow to fully integrate it into their operations, including that organisations are identifying more uses for AI/ML.

In the first edition of the survey in 2021, 78% of respondents cited either improving the quality of investigations and regulatory findings, at 40%, or reducing false positives, at 38%, as their primary reason for AI/ML adoption.

However, the answers to this question in this year's survey were more diverse. While improving investigations and findings and reducing false positives remained the top two answers, their combined percentage dropped by 11 percentage points to 67%.

Meanwhile, adopting AI/ML for detecting complex risks rose from 17% to 21%.

Further, in ranking the three technologies based on their impact, 58% of respondents ranked ML as their top choice, up 6% since the 2021 survey, 28% ranked robotic process automation as their top technology choice, and 14% ranked natural language processing as their top choice.

While ML’s ability to identify patterns in large amounts of data is impactful, the low response for natural language processing might indicate that compliance teams are missing early warning signs owing to underdeveloped capabilities, SAS notes.

Additionally, the reasons for not adopting AI/ML have also evolved. In 2021, the top obstacle for passing on AI was budget constraints, at 39%, SAS says.

This figure slipped to 34% in this survey and was overtaken by the lack of a regulatory imperative, up slightly to 37%. Lack of available skills is also becoming less of a concern, with the percentage falling by nearly half to 11%.

However, the study shows that AI and ML are producing value when they are fully implemented.

Reducing false positives is a growing priority among respondents. When asked about their priorities for AI/ML deployment, AML experts cited the reduction of false positives in existing surveillance systems at 38%, which is an 8% increase since 2021.

Further, 25% said their priorities for AI/ML deployment were automating data enrichment for investigations and due diligence, and 23% said it was for detecting new risks with advanced modelling techniques. However, both of these reasons dropped by several points from the previous survey.

The remaining 13% of respondents cited customer segmentation for behavioural analysis as their main priority for AI/ML deployment.

Reducing false positives and negatives was also the top answer for which area offers the most value from AI/ML, at 38%. However, better and faster investigations, at 34% of respondents, and triaging high- and low-risk alerts, at 28%, were not far behind in areas that offer the most value, the study says.

“The key to unlocking the full potential of AI and ML is integration of data sources, teams and technology,” says SAS Risk, Fraud and Compliance Solutions senior VP Stu Bradley.

“In this ACAMS survey, 86% of respondents reported doing some form of integration between AML, fraud and information security processes. Nearly one-third have a fully integrated case management capability across those functions. Another one-third collaborate through cross-functional teams to deploy controls to prevent financial crimes exposure.

“Some organisations may be waiting on regulatory guidance, but firms that press ahead with integrating data and operations with governance in mind are laying the groundwork for responsible innovation in AI and ML and will enjoy a competitive advantage over those who hesitate,” he adds.