The crucial role of real-time analytics in fraud prevention

Dan Dica, chief executive officer at Lynx

Picture a world where fraudsters have nowhere to hide, their every malicious move is monitored, and their swindles are halted at their source. 

We’ve become accustomed to crimes going undetected for a shockingly long time. In past eras, news of fraud would travel only as fast as the fastest horse. Even today, fraud can continue unchecked for days (or longer) before patterns emerge due to the delay in reporting or victims being unaware of the loss. Equally, victims may feel ashamed of the loss, and be unwilling to report it for these reasons. Regardless of the reasons, fraud has a hugely limiting impact on economic growth, with more than one in three companies reportedly using reactive, manual methods to detect fraud.  

As the world continues to digitise, we must fight back harder. Criminals are finding it easier than ever to steal banking and credit card information, or even purchase packages of personal data on the dark web. In turn, this has led to a sharp rise in the amount of money stolen globally every year. The audacity of fraudsters is growing, as are the advancements in the channels and methods they can use to target their victims. Automated synthetic identity fraud, for example, results in mass mule accounts following the compromise of identities. These are being harnessed to wash money, and according to insights from Experian, 42% of first-party current account fraud in the UK is now mule-related, with the fraud rate for current accounts rising by 13% in the first three months in 2023. 

Anyone who has fallen victim to fraud knows that it can destroy lives and livelihoods, Victims may face damaged credit scores, loan defaults, bankruptcy, and long-lasting financial instability. For these reasons, both consumers and businesses are becoming increasingly selective about the financial institutions they use and trust. Those that offer top-tier security are more likely to retain their core customers for longer. 

Real-time detection in action 

Traditionally, fraud detection has relied on analysing past attacks and creating rules to address them. However, this approach suffers from critical flaws. For instance, looking at past patterns leaves companies vulnerable to emerging fraud tactics. Furthermore, every new attack needs to be studied to identify how it beats the current rule set. This is not an adequate way to deal with today’s advanced and rapidly evolving attacks. 

Financial institutions may then build rules and a static machine learning model that will try to protect future transactions and events. The problem is that the past is not a good prediction of the future, which is termed ‘data drift’. Customer behaviour changes with new products, trends and behaviours, while attackers change their modus operandi to bypass rules and get paid. 

Ultimately, the attackers have an advantage as they get a response for every transaction, allowing them to learn the behaviour of the machine learning models and rules. Once this is known then they can change their attack to bypass the defence. Those at the forefront of fraud prevention utilise self-learning machine learning models that are retrained every day to keep up to speed with customer behaviour and attacks. 

In the modern era, it’s imperative to move beyond hindsight and unlock 20:20 real-time decisions and vision to succeed with predictive machine learning models that learn every day. 

By analysing every event in real time, financial institutions are equipped with immediate identification of suspicious patterns and potential fraud. Real-time decisions flip the script by analysing data in milliseconds and providing a meaningful response, i.e. declining transactions synonymous with fraud. This is realised by self-learning machine learning algorithms that monitor real-time transactions, communications, and account activity. This enables the instant identification of suspicious behaviour based on up-to-the-moment patterns, helping institutions to block potentially fraudulent transactions before they are approved. 

The enabling technologies of real-time detection 

Attackers can very quickly identify a vulnerable channel and focus their attention on increasing fraud, for example, telephone banking. To maximise protection and prevention, vision is required across all channels with a unified view of all events and transactions. This includes pinpointing fraudulent transactions during authorisation across payment networks, detecting suspicious login attempts on mobile banking, changing address information via the telephone, visiting a branch to request a new card, and spotting irregular purchasing patterns indicative of theft via credit card usage and more.  

To unlock the full potential of real-time prevention platforms, solutions like stream processing, in-memory computing, self-learning machine learning algorithms, and a real time decision engine are required.  

Most importantly, these models must be adaptive. Best-in-class solutions are trained daily, which includes data from the day, the relevant features and confirmed fraud–hence, daily adaptive models. This ensures that the machine learning models continue to adapt to changes in evolving customer behaviour, changes in attacks and new types of fraud. With models updating daily, strong governance and expertise are crucial to ensure rigorous testing and responsible oversight. 

As fraud evolves, so too must our defences. With real-time decision making, financial institutions can transform fraud loss from an inevitability to an exception. 

This new method of fraud detection brings finance into a new era, where transactions can be secured in real-time against digital threats. Put simply, daily adaptive models are a critical asset in any financial institution’s toolbox to halt fraud attempts before they succeed. 

Image: Lynx  

Guest Editorial This article was produced specially for Fintech Intel by an expert guest contributor.