Blog article
See all stories »

How machine learning can cut costs on transaction monitoring: a new design for a new world!

The need for transaction monitoring

The world today is a very different place than even ten years ago; criminal activity is increasingly sophisticated and terrorist financing uses cutting edge technology. Financial institutions, together with governments and regulatory authorities must use every technological advance to fight it.

In the modern world there is more and more emphasis on fighting financial crime, money laundering and terrorist financing. Regulators are more demanding and are enforcing a higher level of scrutiny on companies in the transaction monitoring space. Digitalisation is a global phenomenon and the amount of ‘wire’ activity is growing significantly, increasing the pressure on companies and making it harder for them to monitor and detect suspicious activity.

IT solutions companies are creating tools to increase robustness in the transaction monitoring process and the detection of unusual financial activity. These systems are based on standard typologies of money laundering:

  • Identifying spikes in value or volume of transactions
  • Monitoring high risk jurisdictions
  • Identifying rapid movement of funds
  • Screening against sanctioned individuals and politically exposed persons (PEPs)
  • Monitoring enlisted terrorist organisations
  • Etc.

These systems are designed to flag potentially suspicious transactions and produce alerts based on established thresholds. Tracking the size of day-to-day operations, the usual data flow and created thresholds allows the system to create standardised alerts. The companies with a high level of wire activity may face a significant amount of alerts, all of which have to be captured, analysed and designated as either false positive or reported to the financial intelligence unit as suspicious activity.

Challenges faced with transaction monitoring

The need to monitor and track transactions does create a series of challenges:

1)      Setting up the correct threshold levels and parameters

2)      Identifying ‘false positives’ quickly and accurately

3)      Streamlining operations to minimise costs

4)      Complying with global and regional laws and regulations

5)      Accurate and timely reporting

6)      Accurate data sources

When thresholds are set too low the system will populate a high number of alerts that require analysis. If thresholds are set too high the amount of alerts will decrease, but the company may not detect all suspicious activities and will fail to meet regulatory requirements, which in turn risks both reputation and exposure to fines.

Analysis of the alerts may be time consuming, but must be completed with a sufficient level of scrutiny to ensure compliance with existing governance processes. False positives, which are likely to be the biggest challenge, should be identified and removed as quickly as possible.

Depending on the cycle to which systems are operating and the amount of data they are examining, the same type of activity may recur in every cycle. It is also possible that the same transaction(s) will trigger alerts across multiple scenarios (typologies). Identifying any recurring false positives and providing operational solutions in order to minimise their impact on the process is a crucial factor for organisations in order to minimise workload and duplication of effort.

Possible solutions?

If the IT system was able to learn from previous cycles and identify false positives before an alert was generated, it would be a ‘game-changing’ factor in transaction monitoring, speeding up and increasing the accuracy of identifying the truly suspicious activity. We are seeing increasing examples of machine learning in many areas of technology and financial institutions should grasp the opportunity to use it for repetitive analysis.

There are also alerts that are generated which carry a relatively low money laundering risk, e.g. an alert based on a counterparty moving funds between their own accounts (possibly held in different jurisdictions or in different banks). This type of false positive is common, but still requires analysis. If the system could perform a simplified up-front analysis to exclude these types of alert, or move them to a specially quarantined area, then allied with self-learning, the alert landscape becomes cleaner and only the truly suspicious transactions require costly human intervention.

To ensure accurate data and downstream reporting the data collection for transaction monitoring, especially when using multiple systems for different financial products, should be streamlined and cleansed to ensure that the monitoring results are consistent across products.

Compliance operations may be called a ‘necessary evil’ since these areas do not generate revenue and exist mainly to satisfy the regulators. However, these areas protect the organisation against multiple risks and prevent any potential fines for not complying with the regulations of the markets in which they operate. Reducing costs for compliance and transaction monitoring teams, whilst increasing the robustness of the surveillance and reporting, are therefore clear objectives for any organisation; which can be attained by increasing the intelligence of the technical solution, whilst streamlining processes and data.

When organisations implement transaction monitoring operations, they must ensure that the systems are sufficiently robust and able to provide accurate proof to the regulator of both historic activity and current processes, from data capture to reporting.


The transaction monitoring operation has its challenges and is perceived by many organisations as a necessary cost which they need to incur to meet regulatory requirements. However, the area has great potential for increased automation and smart technical solutions, in order to minimise human intervention which ultimately leads to increased costs and the risk of additional error.

There are opportunities to create an operating model which links applications and data sources, to create a system that will ‘self-learn’ with the ability to identify only those transactions which pose genuine risks and should be reported.

Different types of organisation require different levels of integration and sophistication; therefore various rules are required to be applied for different types of money laundering and financial crime prevention.

Nowadays, there are many IT solutions for this process, however, only a system which can evolve and understand the challenges of the process will be successful in the longer term.



Comments: (1)

Graham Seel
Graham Seel - BankTech Consulting - Concord 10 August, 2016, 17:12Be the first to give this comment the thumbs up 0 likes

Well explained. Rules-based monitoring systems have to be set with very conservative thresholds, resulting in massive falst positive volumes. This creates very high operational costs, and also risk associated with "fault fatigue". As suggested in my blog How Industrial Operations AI Can Help with Banking Risk Management, banking can learn from industrial process management about anomaly detection in a machine learning environment. Several companies are working on this already including behemoths like IBM, and small startups like Amberoon.

Now hiring