'The perpetual evolution of financial crime necessitates an equally dynamic response from anti-financial crime and identity technologies.' 

This paper examines the three evolutionary stages of artificial intelligence (“AI”) integration into the anti-financial crime software market. It explores the initial adoption of Natural Language Processing (NLP), the subsequent incorporation of machine learning models, and the emerging potential of generative AI capabilities.

1/  Introduction

Financial crime poses a significant threat to the global financial system, requiring continuous innovation in prevention and detection methodologies and capabilities. The integration of artificial intelligence (“AI”) into anti-financial crime (“AFC”) software has undergone distinct evolutionary stages, each ushering in transformative capabilities. Understanding these stages is crucial to eliminating the allergic reaction to AI by financial institutions and regulatory bodies.  It also provides a narrative with which software providers can take their audience on a journey dispelling the fear of AI adoption.  What most haven’t realized is that they’re already using it today.

2/  The Three Evolutionary Phases

Phase One:  The introduction of Natural Language Processing and Robotic Process Automation

The first stage in the AI revolution involved the adoption of Natural Language Processing (“NLP”) technology.  If we follow IBM’s logic, NLP is a subfield of AI.  And for those more endeared to Amazon Web Services (“AWS”), NLP is a machine-learning technology.   

We first start to see NLP deployed in the context of watchlist and sanctions screening facilitation.  The use of ‘facilitation’ is intentional, given the additional components of a front-end platform and data being necessities in delivering screening outcomes. NLP allowed for fuzzy matching algorithms for approximate string matching, enabling name detection or approximations within degrees of confidence.  We also begin to address different languages, scripts, and transliteration needs.  There are some who make the leap of faith that his ushered in contextual analysis, but the evidence is lacking at best.  What we do now is that this was the first breakthrough in leveraging AI in the anti-financial crime (“AFC”) fight. 

One common misconception is the inclusion of Robotic Process Automation (“RPA”) in the AI lexicon.  I have been guilty of this presumption as well.  The more committed could debate the merits of why/why not while we move on to the second evolutionary phase.

Phase Two:  Adoption of Machine Learning Models

The second evolutionary stage leads to the proper introduction of machine learning models into AFC applications. They enable the prevention (fraud) and detection (AML) of patterns and anomalies in transactional data that traditional rule-based workflows may not take action on. Particularly in the transaction fraud realm, a massive opportunity opens up to improve PnL figures at a time when payments were beginning to move digital.  In real-time transaction fraud prevention and batch processing AML transaction monitoring, machine learning provided a glimpse into the future and an opportunity to get on the front foot in the AFC fight for the first time.

Despite the evident benefits, industry adoption has been gradual. Many financial institutions are only beginning to integrate machine learning models into their compliance and AFC frameworks. The adoption amongst fraud programs has become an expectation vs. their AML colleagues’ more cautious approach.  Regulatory skepticism and explainability played a large role in that.  We also see challenges such as data quality, model interpretability, and reliable technical delivery slowing widespread implementation. However, institutions who were early to embrace machine learning models in their fraud programs and explore capabilities in their AML function are seeing significant benefits.  They also are advanced in their organizational comfort level, ahead of their peers in their preparedness to embrace the current stage of our AFC AI evolution.

Phase Three:  Emergence of Generative AI and LLMs

The current and third stage involves the emergence of generative AI, particularly Large Language Models (“LLM”s), in the AFC domain. LLMs have the capability to understand and generate human-like text and context, offering advanced solutions for analyzing unstructured data sources such as emails, contracts, and communications that may indicate fraudulent and/or illicit activities.  Some of the more adventurous start-ups in the AFC fight are exploring large transaction models, vertically specific LLMs to the various discipline within AFC and compliance, and analyzing the data gathering and categorization opportunities.

There is an accelerated interest and adoption of generative AI capabilities within the industry, spurred heavily by adoption in other areas of the bank.  Where machine learning model adoption was slow in the non-AFC areas of banks, hampering second phase adoption, AFC teams can benefit from the institutional (read ‘procurement’ or ‘approved vendor’) embrace of this nascent technology to improve their own programs.  In particular, vertical-specific generative AI models address complex compliance challenges with enhanced transparency and efficiency. Early applications also include intelligent document processing, automated report writing, and advanced customer due diligence.

Looking ahead

New entrants in the software market are leveraging modern technical stacks devoid of legacy technical debt, enabling rapid innovation and deployment of generative AI solutions. However, these providers often lack deep domain expertise in financial crime compliance, posing challenges in meeting the nuanced requirements of regulatory frameworks. Conversely, legacy software providers are attempting to reinvent themselves by integrating generative AI capabilities into their existing platforms. They bring extensive domain knowledge but may struggle with outdated architectures and technical debt that hinder agility and innovation.

3/  Landscapes

The dichotomy between legacy providers and new entrants highlights a critical tension in the industry. Legacy providers possess valuable domain expertise and established client relationships but must overcome the limitations of outdated technical infrastructures. Their efforts to integrate generative AI involve significant reengineering and investment to modernize platforms while maintaining compliance integrity.

New entrants, unencumbered by legacy systems, can adopt the latest technologies and architectures, offering innovative solutions with potentially greater efficiency and scalability. However, their lack of domain expertise may result in solutions that do not fully address the complex regulatory requirements or fail to gain trust from conservative financial institutions.

Financial institutions must navigate these options carefully, balancing the allure of advanced technological capabilities with the necessity of compliance accuracy and regulatory acceptance. Collaborative approaches, such as partnerships between legacy providers and new entrants or the incorporation of domain experts into new tech firms, may offer pathways to combine technical innovation with compliance expertise.

4/  Search for Value

The substantial capital investment awaiting deployment in this third evolutionary stage reflects the market's anticipation of generative AI's transformative potential. Venture capital and private equity firms are actively seeking opportunities to invest in AI-driven compliance technologies, recognizing the significant ROI potential.  Banks are dedicating significant R&D resources and budget across all functions in search of generational transformation in operations.

5/  Conclusion

The evolution of AI in the anti-financial crime (“AFC”) software market encapsulates a journey from machine learning techniques met with excitement, to sophisticated analytical capabilities seen through a skeptical adoption lens, and now a fever around advanced generative intelligence. Each stage has offered tools to prevent and detect financial crime, yet also presented challenges in their own right.

As the industry stands on the threshold of widespread generative AI integration, financial institutions have the opportunity to significantly enhance their compliance functions (not just anti-financial crime). Success will depend on balancing innovation with domain expertise, navigating the complexities of integrating the most advanced technology in a highly regulated and ROI sensitive domain.  Sprinkly significant investment capital into the mix and we’ve set the stage for a transformative era in anti-financial crime efforts. Stakeholders who proactively engage with these developments will be better positioned to mitigate risks, improve operational efficiencies, and contribute to the integrity of the global financial system.