.svg)
We recently asked a room of financial crime compliance experts: which use cases do you believe will be most suitable for artificial intelligence (AI) in ongoing monitoring?
Overwhelmingly, the response was ‘transaction monitoring’. But when asked how far they have traveled on the AI implementation journey, the majority were at an early stage.
Challenges in AML transaction monitoring
Financial institutions looking to implement AI in AML have several common challenges to solve.
- Reducing false positive rates through improved accuracy
- Tuning to the varying risks by geographies, segments, and clients
- Tuning to a risk-based appetite
Accuracy is ultimately driven by a robust risk-based assessment, and then applying it to transaction monitoring models and rulesets. Layering in AI can drive improvements, when the underlying customer risk is well understood and the rules are well configured.
Use cases for AI for AML transaction monitoring
Successfully implementing AI for anti money-laundering transaction monitoring means meeting key regulatory compliance requirements: transparency, auditability, fairness, and legality. When the underlying AI meets these AI ethics requirements, potentially compliant uses of AI in AML transaction monitoring include:
- Auto alert discounting
- GenAI decision explanations
- AI driven insights for missing information, specifically typology detection
Typology detection is one of the most well-developed areas, with initiatives such as the United Kingdom’s Financial Conduct Authority (FCA) launching its synthetic data initiative which partners with Napier AI and others to improve money laundering detection. Using real anonymised financial transactions from high street banks augmented with a wide range of money laundering typologies, the partnership was able to create a new fully-synthetic dataset that now underpins other initiatives such as the Supercharged Sandbox.
The Napier AI / AML Index 2025-2026 identified the top typologies by market, with the UK reporting highest losses to financial crime from:
- Human exploitation and trafficking
- Cyber & Financial Crimes
- Illicit trade and counterfeiting
These typologies are included in the synthetic dataset and available for financial institutions to test and optimise their transaction monitoring, which shows the benefits of regulatory collaboration in the fight against financial crime.
AI for AML Typologies
When AML typologies are built into transaction monitoring systems, AI can identify subtle patterns in alerts that align with known financial crime typologies, even when no single rule was triggered, offering deeper insight into alert quality.
The right approach to AI blends the transparency and familiarity of traditional rules with the adaptability and pattern-detection capabilities of AI, providing firms with a practical and explainable route to innovation that meets compliance requirements for transparency and auditability.
Complex typologies pose real challenges in transaction monitoring, and these could be addressed with AI.
Complex customer relationships
High-net-worth clients often operate through a web of entities - trusts, special-purpose vehicles (SPVs), and offshore holdings - making it difficult to establish a clear picture of beneficial ownership or control.
Fragmented data across intermediaries
Each holds partial data, leading to siloed insights and gaps in transaction visibility.
Evolving risk typologies
Exposure to digital assets and cross-border private placements creates new financial crime vectors that static rules struggle to detect.
Documenting the typologies that transaction monitoring systems should be identifying is a top priority for the FCA – and great strides have been made. But transaction monitoring is a different beast from sanctions screening, with much more grey area. Financial institutions, regulators and law enforcement need greater accuracy in detection to avoid an avalanche of false positives that make it easier to hide financial crime trails.
Explainable AI for AML transaction monitoring
Regulators require that suspected money laundering be flagged with Suspicious Activity Reports (SARs), but there is increasing focus on the quality of SARs, even approaching a ‘crackdown’ on overreporting or incorrect form population. To combat financial crime effectively, regulators and law enforcement need more accurate and detailed information from financial institutions. To successfully improve SAR quality, financial institutions need timely feedback from regulators.
There is an industry need for continuous feedback and outcome-based tuning to quantify:
- SARs quality for to law enforcement
- Conversion rates from alerts to meaningful internal investigations
- False negative rates that resulted in missed risks only surfaced after regulatory inquiries
Firms should establish regular feedback loops to feed these insights back into scenario tuning, model retraining, and risk scoring. Learning based on experiences, models can adapt linear thresholds across time or geography or segment vectors, applying a heuristic learning approach to identifying AML typologies.
This is where the role of centralised data sources for testing become so crucial, such as AI Live Testing from the FCA. A shared repository strengthens the market via secure data sharing and synchonised learnings, leading towards more robust defences against brute force attacks which target financial institutions in sequence.
For individual financial institutions, the key to accuracy is to better understand the shifting ‘normal’ range for their individual customers, for their segments, and for the wider market. This requires AI-powered dynamic customer segmentation to continuously triangulate these shifting normal ranges against historical data and each other.
Fighting financial crime with AI
AI certainly has a role to play in AML transaction monitoring where it can be used to help tackle complex financial crime schemes as well as support operational efficiencies for financial institutions.
Intelligent applications of AI can support proportionality in operational approaches, with humans-in-the-loop analysing data and applying AI to easier tasks so as to release experienced resources to make strategic decisions. AI can tackle the volume and capacity challenges, while expert humans shape the alert review.
Discover the AI regulations shaping AML transaction monitoring in your region, download the Napier AI / AML Index
Photo by and machines on Unsplash
