Integrating AI into AML strategic transformation

The rise of Artificial Intelligence (AI) is a mathematical certainty. As companies and individuals move towards a more digitally enabled world, data grows at an exponential rate, and this data fuels AI systems. With increasingly large volumes of data available within financial services institutions (FIs), and third-party data sources now providing a wider range regulatory compliance data, AML systems are ripe for the AI revolution.

Machine Learning and Deep Learning

It’s important to start by clarifying terms which are often misunderstood. Let’s address machine learning first because it’s the least contentious. Machine Learning (ML) describes computational learning using algorithms to make predictions from data. Deep learning is an extension of ML, where multiple layers of machine learning processes are used. The term ‘deep’ refers to the multiple layers. AI is the umbrella term that encompasses both ML and Deep Learning and several other techniques to bring computer systems to a point where they can mimic human behaviour, including making differential decisions on ‘fuzzy data’ - just as humans do. 

Challenges to AI implementation remain

Although many believe AI is the future, it is not universally accepted or welcomed just yet. There are six major challenges related to this: 

1. There is a lack of consistent regulatory support and guidance at a global level.

2. The ‘explainability’ of an AI decision in such a way that a bank or regulator can use, understand and explain is not a simple task, due to AI platforms’ complexity. 

3. The ethics and morality of using AI to make decisions that could be systematically bias makes some potential users of AI within AML nervous - the negative media that could occur if AML decisions are systematically bias is a real and present danger.

4. The implementation experts within this domain are in globally short supply. There are literally hundreds of vendors within this space vying for the same resources, and while rapid training and sponsorship is helping to alleviate this, we are still a few years away from balancing the supply and demand for resources. 

5. It is not uncommon for financial institutions to have systems that are 10 or sometimes 20 years old, and accessing data which sits within decades old technology and is often stored in siloed systems, is a challenging mission.

6. And finally, there are significant cost implications, which are expanded on below. 

The cost of compliance is a key consideration in AML AI

Anti-Money Laundering compliance processes have historically been labour intensive and some financial institutions are challenged by the cost of compliance. High cost processes to onboard customers and maintain AML customer compliant records often exceed profit made for those customers, especially for the SME retail and commercial banking market. On average, banks take 24 days to complete the onboarding process. The cost issue is exacerbated when a Know Your Customer (KYC) remediation is required by regulators. This is because the full update of one KYC file - even for an SME - can easily cost £1,000 or more to process.

However, for simple decision making, AI techniques can help reduce the overall cost of compliance within AML and in particular KYC. A great example is the use of AI natural language processing (NLP) to assess validity of documentation submitted by customers within the KYC process.

This is a booming area of research, product development and implementations. It’s no secret that Transaction Monitoring alert dispositions can result in up to 99% of false positives, with each false positive requiring an analyst to spend 2-3 hours to investigate the alert. AI techniques like ML can reduce false positives by as much as 95%, leaving only those cases which require real human investigation. 

AI is a strategic, long-term solution

Moving into the more strategic implementation of AI, advances in unsupervised ML techniques enables firms to detect AML risk patterns across multiple AML processes which are not predetermined. Alerts can be generated for new issues that have not been previously seen. Using a combination of predetermined risk typologies and unsupervised learning provides a completely different way of approaching AML risks.

Combining predetermined rules and anomalous pattern detection can improve the quality of risk management, beyond just the traditional and often simple risk typologies that financial institutions use to detect issues. 

The power of AI can be more broadly realised when integrated into a strategic AML transformation. It is now possible for AI to correlate multiple sets of complex data and synthesis it into actionable intelligence for an AML operative.

This, coupled with a single view of the customer, gives a much more holistic understanding of both the customer and their risk. The power to augment the human is the next logical step in the use of AI within AML.

With the platforms available now, this ambition to augment the collective intelligence of financial crime operations by embracing the potential of AI to work in the loop with experienced financial crime operators is within the grasp of visionary leaders who want to drive transformative change through technology. 

by Sha Ali, UK Financial Crime & Forensics Director, and Sabah Hussein, UK Forensic Data Analyst, Ernst & Young