Key points
- AI-powered system combines robotic process automation and machine learning to detect suspicious transactions more effectively.
- The new system automates preliminary checks and can temporarily freeze high-risk accounts, while providing analysts with risk assessments and recommended actions.
- While banks report improved detection rates and fewer false positives, challenges remain around data quality and maintaining regulatory compliance through explainable AI.
In 2020, the bank successfully merged two emerging technologies: robotic process automation (RPA) and machine learning. We needed detection methods that could adapt as quickly as the criminals.
Before automation, the bank’s compliance team faced an avalanche of alerts.
Most were false positives, yet each required manual verification against multiple databases, a task that was time-consuming and slower. Critical warnings were buried under hundreds of benign transactions.
The new system has streamlined this process, with software robots managing the preliminary work, extracting and standardising transaction data and checking it against international watchlists. Machine learning algorithms analyse hundreds of data points per transaction, from payment patterns to counterparty behaviours, having been trained on thousands of confirmed money laundering cases.
The system operates more like an experienced investigator than a conventional detection tool. It develops pattern recognition capabilities through continued exposure to real cases.
When it identifies suspicious activity, the system automatically assigns the case to an appropriate analyst that includes risk assessment and recommended actions. In high-risk situations, it can temporarily freeze accounts pending human review.
New system encountered challenges
However, there were some hiccups along the way. Early versions of the system generated false positives and data quality was an issue. Machine learning algorithms require clean, consistent data to function effectively, which many financial institutions struggle to provide.
Regulatory compliance presented another challenge. Banking supervisors require transparency in decision-making, yet complex AI models often operate as “black boxes”, making it difficult to explain their determinations.
Meeting regulatory requirements means demonstrating how conclusions are reached. Simply being correct isn’t sufficient.
Banks are addressing this through more interpretable AI models that can provide clear reasoning for their decisions. Some employ straightforward techniques such as decision trees and others use sophisticated tools to generate explanations for complex AI determinations.
The scale of the challenge is significant. UN estimates indicate criminals launder between £800 billion (A$1.2 trillion) and £2 trillion (A$3.9 trillion) annually, representing 2-5 per cent of global GDP. These funds support terrorism, human trafficking and drug cartels.
Money launderers increasingly use cryptocurrencies, complex corporate structures and sophisticated transaction patterns, making it a challenge for traditional monitoring systems to keep up.
Machine learning enhances detection
This is where machine learning can help. Unlike rigid rule-based systems, AI can identify new patterns as they emerge. Each confirmed case strengthens the system’s detection capabilities.
Anti-money laundering systems process highly sensitive information, from personal data to investigation details. In response, banks have implemented strict data access controls, encryption and secure transmission protocols.
Every piece of data is encrypted, every access is logged, and every user’s identity is verified. The system assumes attempted breaches will occur and is designed accordingly.
While precise figures remain confidential, banks using these systems report improvements in detection rates and reductions in false positives. This allows compliance teams to focus resources on genuine threats.
There is potential for further development. Emerging technologies such as computer vision, natural language processing and biometric recognition could boost detection capabilities. These tools could analyse surveillance footage, process handwritten documents and verify identities through voice patterns.
Current applications represent only initial implementations. As artificial intelligence capabilities expand, our ability to detect financial crime will increase correspondingly.
The Financial Action Task Force, which sets global anti-money laundering standards, reinforces risk-based approaches to combat financial crime. Advanced automation helps banks deploy resources more effectively.
The underlying dynamic remains unchanged: as detection systems improve, criminals adapt their methods. However, banks have the tools to sharpen their response. These AI systems are fundamentally altering the landscape of financial crime detection.
Note: ‘H International Bank’ is an alias.
Dr Xiaoxin Chen is an IPA member and a certified anti-money laundering specialist who has studied the bank’s digital transformation.
More information on the IPA’s Joint Submission – Modernising Australia’s Anti-Money Laundering and Counter-Terrorism Financing Regime can be found HERE.
