Using technology to analyse payment data, automate workflows and make decisions, for example around fraud detection, is becoming quicker and more efficient. At the same time, AI and machine learning offers greater flexibility and control, allowing banks, merchants, and payment providers to launch new business models quickly. Experts estimate that banks are deploying AI-driven systems in record numbers, with more than $217 billion spent on AI’s applications for use cases like fraud prevention and risk assessment.
The processes involved in fraud detection are an ideal use case for machine learning and artificial intelligence (AI). But as we know, there is more to fighting fraud than technology. The significance of understanding the operational process in combating fraud is fast becoming critical. Historically, many work streams were manual and decentralised, which limited the ability of fraudsters to scale up fraudulent activity. However, as many modern business processes have become centralised, with more and more powerful technology, the ability to scale up fraudulent activity has become a more compelling prize. A fact that has become all too evident by the large-scale data breaches that we see around the world today.
My team and I believe that the current rise in fraud is inextricably linked to highly scalable technical process failures. Fraud prevention processes are no different. Process deference is critical. We need to understand that solving financial crime is not just about products, highly trained people and well-defined processes are also critical. With the increased shift towards digital payments, which has only been accelerated by the COVID-19 pandemic, manual data analysis and threat detection methods have failed to stop the fraudsters. As a result, multiple industries are looking to automation tools to manage, enhance and improve their anti-fraud operations.
Historically, some fraud detection processes that required a mix of machine learning and human subject matter expertise (SME) have been very difficult to automate and replicate. However, new tools are entering the market to enable that process. Tools that can remove the manual effort from fraud detection processes, which had previously required 24/7 management.
Where large scale operations exist, such as with the Internet of Things (IOT) and scalable cloud services, there is a real possibility that security intrusions and threats can go unnoticed. Especially if the right tools are not employed. Forward thinking businesses are now using a set of AI/ML tools known as anomaly detectors. Utilising unsupervised learning methods to understand and identify normal activity. Then creating alerts or processes to follow when abnormal activity is detected.
In this way, whole networks can be effectively monitored and protected, as each individual node will have its own behaviour ‘footprint’, which can be monitored both separately, and as part of the whole system, automatically, in real time. This is extremely powerful, for example, when a web service is targeted by cybercriminals. These tools can detect the abnormal behavior and stop it before it does any damage.
Extremely detailed information about a data set, found using these tools can also be used in other, more positive ways. A deeper understanding of individual customers, which is used to assess and flag if something is wrong, can also be used to identify similar customers and their spending patterns in marketing campaigns and other positive engagement activities.
The benefits of automation
With fraud on the rise, and a more traditional approach creating more rules and models leading to more alerts and higher incident rates, automation of the fraud management process will be paramount to successful fraud detection in 2021.
Many rules-based fraud detection solutions require near round the clock monitoring to ensure effective fraud detection. Even with machine learning doing some of the ‘heavy lifting’, there is still a large amount of monitoring that is required to keep on top of fraud. This, coupled with the rise in electronic payments, means it is becoming far too time consuming to manually discover fraud patterns and write effective rules, or create machine learning models to combat fraud.
Automating this process is not simple, as there are several manual processes, including models and rules performance monitoring, fraud pattern discovery, model and rule generation and fraud alert management processes, to automate. However, the reward for doing so is compelling. Offering huge savings, as any fraud will be detected and stopped much faster, and fraud SMEs can focus on extremely difficult fraud cases, as well as on providing more value for their customers. As a result, automated AI/ML based fraud systems are quickly becoming the standard in the fight against fraud.
However, developing automation tools as products brings its own set of problems. One of the biggest issues my team faced was engineering ways to reduce or entirely remove user interactions, whilst still making our product perform how the user wanted it to and giving them choices when they were required.