The banking and payments industries are undergoing rapid transformation, from customers using mobile banking applications to contactless, online and mobile payments. Throughout these changes, fraudsters unfortunately manage to keep pace with attacks on new channels. However, a new machine learning approach to fraud prevention has emerged that can stop fraud in its tracks.
Tactics used by fraudsters are becoming increasingly sophisticated and this has become more apparent with the increase in both the types and frequency of fraud attacks. In 2016 alone, £2 million per day was lost in the UK to financial fraud, according to a report by Financial Fraud Action UK.
The most common denominator in fraud attacks is the illegitimate transfer of money – whether it is through account takeover, fraud attacks during authorisation, malware, or email phishing scams.
Traditional fraud systems have focused on detecting and predicting fraudulent activity at the point of a financial transaction. Outdated and inefficient existing fraud systems largely detect fraud by setting thresholds on customer activity, such as transaction amounts or geographical locations, which fraudsters are increasingly finding ways to exploit.
Simply monitoring and detecting fraud at the point of transaction is no longer enough to protect customers and business revenues. The real key now is to quickly and accurately identify and prevent fraud by looking holistically at all the non-monetary data points leading up to a transaction, as well as the transactional data itself.
What do we mean by non-monetary interactions and why do they matter?
From creating a new account, or signing-in to an existing one, through to interactions with customer services or changing personal details, these are just a few examples of the numerous non-transactional interactions that customers have with their banks.
Fraud prevention systems need to consider all of these interactions as they are key to building an end-to-end lifecycle of customer behaviour, learning each time a customer interacts with their bank. Using adaptive behavioural analytics to spot the subtle changes leading up to a transaction is vital to catching fraud before any money leaves an account.
Changes can be as subtle as a shift in speed or accuracy of typing in a password; a recent update to a telephone number or address, or a change in the sequence of pages a customer looks at when logging-in to their online bank account.
The amount of data involved is vast – a joint report by the BBA (British Bankers’ Assocation) and EY revealed that in 2015 there were 9.6 million log-ins to internet banking per day in the UK, that is over 6,600 log-ins per minute.
So, modern fraud prevention systems now need to take a more complete view of customer interactions and behaviour, analysing all the available data points and scaling to understand vast volumes of data in real time to prevent fraud from taking place.
Distinguishing criminal from customer
Advanced machine learning fraud systems consider all these customer interactions to build individual profiles, a kind of behavioural roadmap, which gives a better understanding of a customer’s normal behaviour. So that even when something subtle changes, such as a slight pause when logging in or an out of character account balance check, the true significance of the change is realised and a red flag is automatically raised of possible fraudulent activity.
So, while technology continues to transform the banking industry, potentially exposing new avenues of attack for fraudsters, it also supplies a trail of breadcrumbs, a collection of subtle clues for banks to monitor and better understand their customers.
These clues are what allow banks to tell a criminal from a customer, so that even when a fraudster does manage to creep through, banks can see them coming.