What is machine learning?

Machine learning is essentially the practice of teaching a computer to make predictions based on historical information, and then adjust its behaviour autonomously to optimize output functions.

Advances in technology have enabled financial institutions to develop innovative new tools to enhance both the services they offer and their internal administrative processes. One of the most significant FinTech trends of the 21st century has been the integration of artificial intelligence (AI) systems: based on complex algorithmic analysis of vast amounts of financial and customer data, AI integration helps financial institutions add speed and efficiency to their financial products and services and to manage their compliance responsibilities more effectively in an increasingly complex regulatory landscape.

While AI has driven the rise of online banking, a facet of the technology, known as machine learning (ML), has brought new possibilities to the digital financial landscape. Powered by AI algorithms, ML systems go beyond the speedy, accurate processing of high volumes of customer data by making observations about customers and financial products over time – and then adjusting their future behaviour based on those data inputs without any need for changes to existing programming.

Machine learning systems hold significant potential for financial institutions, not least in a regulatory context, since they represent a way to manage the increasingly challenging data requirements of the modern financial landscape. With that in mind, in order to get the most out of their technological infrastructure, it is important that financial service providers understand the practical capabilities of ML and how it may be deployed to meet their compliance needs.

How does machine learning work?

Machine learning is essentially the practice of teaching a computer system to make predictions based on historical information, and then adjust its behaviour autonomously to optimize output functions. A computer may, for example, be trained to distinguish an apple from other types of fruit based on previous pictures of apples. The same principles may be applied in financial contexts: a computer may be trained to identify suspicious activities based on previous patterns of behaviour and warn a financial institution that a customer’s risk profile should change.

In order to build a machine learning system, firms must understand what kind of data the system will require and how it will use that data to produce desired outputs. Accordingly, the development of a system should involve the following steps:

• Selecting data sets. In order to make predictions, machine learning systems require relevant input data. That input data must be categorized and refined: the system must be able to autonomously identify data that is useful to its training protocol and data that is irrelevant (and that can be discarded).
  
• Developing an algorithm. Algorithmic analysis enables computers to make decisions about the data they receive, and subsequently make predictions about the future. The algorithms used by machine learning systems vary depending on the data being analysed and the system’s desired outputs, but essentially represent a series of procedural steps, each involving a decision that might be based on statistical analysis, sets of conditions, or other specific data characteristics. At the end of the algorithmic process the system creates an output that can be compared to established data and used to inform future decision making.
  
• Training. An effective machine learning system should be able to optimize its outputs and train itself to deliver more accurate results over time. In practice this means that ML systems must be fed new data sets continuously in order to generate outputs that can be assessed and verified for efficacy.

How is it used?

Machine learning is already used widely in a variety of contexts around the world. Websites and apps recommend content to users based on previous browsing habits, email filters learn to recognize and redirect spam emails based on their content, and self-driving cars learn to recognize and navigate their environments.

Machine learning systems also offer possibilities for financial institutions, especially in regulatory contexts where companies must manage large amounts of customer data to meet risk-based compliance obligations, and must adjust their compliance solutions constantly to account for changes in legislation or new criminal strategies. Some of the practical applications of machine learning in financial compliance include:

• Customer experiences. Machine learning tools can help companies reduce compliance friction for customers. By learning to accurately identify high risk customers, computers can ensure that lower risk customers are onboarded more efficiently and subjected to simplified due diligence measures.
  
• Suspicious activity monitoring. Backed by machine learning, computers can be trained to spot subtle changes or patterns in a customer’s financial behaviour that a conventional monitoring system might miss. ML tools enable computers to gauge whether those changes indicate involvement in criminal activity based on comparison with historical patterns.
  
• False positives. While the speed and efficiency of automated compliance tools enable financial institutions to capture a larger number of suspicious transactions, they also generate a larger number of false positive alerts as a result. Machine learning systems can be trained to spot false positives through algorithmic analysis, reducing alert rates and the associated cost of remediation.
  
• Adaptive compliance. Machine learning systems allow companies to spot and react to emerging criminal methodologies more quickly and deploy a compliance response in a more targeted and effective manner. Similarly, these systems can help companies adjust their compliance solution to new regulatory regimes by anticipating and resolving friction points and emergent blind spots.
  
• Financial intelligence. ML systems can be trained to provide financial institutions and national authorities with valuable, contemporary, and targeted financial intelligence that can in turn be used to shape policy decisions and develop new regulations.

What is the future of machine learning?

Machine learning has the potential to transform the regulatory landscape, reducing the cost of compliance solutions while increasing their effectiveness. Beyond its compliance applications, ML can also be used to shape customer-facing financial products and services, recommending, for example, more suitable accounts or budgeting strategies.

Machine learning represents a relatively novel and unfamiliar technology and its impact on the financial landscape is still being explored. Since they require extensive data collection and analysis, machine learning systems also raise ethical concerns and companies should be aware of their jurisdictional data-handling responsibilities under legislation such as the General Data Protection Regulation (GDPR). Companies that integrate machine learning must also ensure that their technology infrastructure is sufficiently protected from external cyberattacks such as hacking or phishing threats. As the application and use of machine learning systems continues, it will ultimately fall to companies to consider how to integrate their new tools safely and effectively.