The Role of Artificial Intelligence and Machine Learning in Transaction Monitoring

It’s no secret that digital fraud is becoming a huge threat. Last year, US citizens alone lost close to $10 billion due to fraud, while fraudsters in the UK stole £1.168 billion.

According to Sumsub’s 2023 Identity Fraud Report, 70% of fraud takes place after initial verification. This means that companies need to stay alert throughout the entire user journey, constantly monitoring behavioral patterns and transaction history. While there are automated solutions that help businesses detect suspicious patterns,criminals are improving their tactics to include fraud networks and AI.

However traditional transaction monitoring systems can’t always spot complex patterns and are often tied to outdated technology. On top of that, many require significant human intervention to function properly. This presents significant vulnerabilities in the face of advancing fraud methods, including AI-driven attacks.

That’s why I believe that the best way to confront fraudsters now is to fight fire with fire. That means employing AI and machine learning (ML) technologies.

AI/ML tools can easily spot complex transaction patterns. This enables businesses to proactively monitor customer behavior while interrogating large data sets and producing court-ready reports.

Now let’s dive deeper into the benefits of AI/ML in transaction monitoring and how to properly integrate them.

What is transaction monitoring?

Transaction monitoring is an ongoing process employed to spot suspicious activities with digital or fiat currencies. The goal of transaction monitoring is to identify illegal activity by analyzing financial data (e.g., withdrawals, deposits, receiving and sending money). Without a proper transaction monitoring solution, companies run the risk of fraud and money laundering (smurfing, integration, placement, money muling) taking place on their platforms.

If you want to learn more about transaction monitoring, download our in-depth guide on the topic. In addition, you can also check our KYC/AML and fraud prevention guide for fintechs.

The challenges of traditional transaction monitoring

Traditional solutions can’t keep up with the scamming techniques employed by criminals today. Here’s why:

• Inability to spot of complex behavior patterns
• High numbers of false positives
• The need for large teams of analysts to review flagged transactions
• Time consuming and error-prone processes
• Reliance on old tech and manual intervention
• Inability to develop new rules as AML regulations evolve

These issues arise, in part, as a result of using predefined rules when analyzing transactions. If a criminal figures out how to bypass these rules, then the solution becomes ineffective. To confront that, companies then tend to use more human power to review transactions and adjust their system.

How AI/ML can improve transaction monitoring

AI/ML have the potential to transform transaction monitoring. Since these algorithms learn as they go, they can detect hidden relationships, anomalies, historical patterns, and non-linear patterns that indicate illicit activity for any type of transaction. This includes fraud that’s often difficult to spot, such as smurfing or structuring.

AI/ML can also improve transaction monitoring by:

• Reducing false positives
• Lowering costs
• Automating the review of flagged transactions, freeing up analysts to focus on more complex cases

If implemented properly, AI/ML can reduce human intervention, reserving it for corner cases — rather than each time a criminal bypasses a pre-made set of rules.

AI/ML-driven transaction monitoring is also well-adapted to handle changing AML regulations and fraud threats, including account takeovers, buy-now-pay-later schemes, card-not-present-attacks, and much more.

How to use AI/ML in transaction monitoring

When considering an AI/ML solution, companies should consider the following parameters:

• Ability to adhere to robust security standards
• Risk-based alerts
• Ability to properly assign risk scores to customers and their historic activity (e.g.,login attempts, typical withdrawal methods, IP address, geolocation, device fingerprint, etc.) and transactions
• Ability to identify complex network patterns and hidden relationships
• Flexibility and scalability for different volumes
• Regulatory compliance support
• Real-time monitoring capabilities
• Embedded analytics to get a bird’s eye view on what’s happening across all applicants within a single-dashboard
• Ability to integrate with new systems while keeping data coming in from other transaction monitoring or KYC systems intact
• KYC data inputs which are crucial for building a holistic customer profile for effective monitoring
• Convenient UI and UX

Companies first need to understand the threats that criminals pose to their transaction systems by establishing a risk-governance matrix that can be used to determine loopholes. Based on that, they can identify red flags and set up the AI-driven transaction monitoring system to spot them.

Companies should also collaborate with fraud investigators and law enforcement to maximize the potential of their AI-driven solutions

We need to remember AI/ML isn’t a one-and-done solution against fraud. Rather, it’s a tool that needs to be adapted to be used effectively.

Challenges of adapting AI/ML to transaction monitoring

There are several main challenges to AI/ML in transaction monitoring:

• Over-reliance. Despite the fact that AI-driven transaction monitoring is constantly evolving, it still requires a human professional that can monitor the system. Moreover, the solution must be adjusted and updated on a regular basis to ensure that the algorithms are evolving in the right direction.
• Adaptation to the regulatory system. Just like with any new technology, regulators need time to adapt to AI-driven solutions. At least at this point of time, AI algorithms are complex and opaque, which makes it challenging to understand their decision-making process.
• Complex cases. AI can be trained well to spot anomalies and flag them, but it still needs to be monitored for complex cases. It’s necessary to implement stringent AI rules and parameters and update them regularly. Otherwise, you’ll likely run into false positives/negatives.

Still, companies can fix these challenges by adapting the solution to their specific needs. This will enable them to monitor large volumes of transactional data in real-time.

If you want to learn more about the ways to overcome these issues, check out our Ask Sumsubers bi-weekly series.

Conclusion

In today’s world, businesses won’t be able to survive without integrating AI/ML into their checks. To learn more about the technical aspects of an efficient transaction monitoring system, check out Sumsub’s advanced solution.

When choosing a verification vendor, it’s essential today to pay extra attention to the quality of the services and solutions it provides. Among other things, companies can benefit from all-in-one platforms that take a holistic approach to transaction monitoring.