How Artificial Intelligence is Transforming Fraud Detection

Fraud Is Growing

Deceit, crime and corruption are rapidly infiltrating every day society. Fraud, in all its forms, can be categorised as an anti-social behaviour, and this behaviour can be categorised even further into two distinct categories. On the one hand, there are those crimes perpetrated directly against victims, such as theft of an individual’s identity and taking over of their bank account, or tricksters cheating pensioners out of their life savings. On the other hand, there are so-called ‘victimless frauds’, typically perpetrated against corporates or the state, such as tax evasion, obtaining credit with no intention of paying it back, and insurance fraud claims.

In this modern society, characterised by an increasing awareness of the social divide between the ‘haves’ and ‘have nots,’ fraud is becoming a safer and more profitable form of criminality. Subsequently, it is hardly surprising that two to five per cent of transactions are fraudulent. However, this doesn’t represent the amount of fraudulent people. Rather, criminals are making fraud their business and carrying it out it at an industrial level.

Without a doubt, fraudsters are getting smarter and digital channels are making it easier for them to conduct this heinous activity from the comfort of their bedrooms; yet, businesses and governments can begin to take on these fraudsters and prevent this. The answer is Artificial Intelligence (AI) and network analytics.

Traditional Monitoring Approaches Will Not Work

Traditionally, the approach taken to tackle fraud has been to embed a set of simplistic rules to alert potential fraudulent behaviour. This can encompass anything from a payment of a certain amount, to purchase of conspicuous goods or consistent round number of payments.

Indeed, it can’t be denied that basic AI techniques have been used to catch criminal activity, but it’s clear more sophisticated AI must come into force. Constantly, fraudsters are testing the boundaries, notably breaking up their fraudulent transactions into small packages to go below the radar and remain undetected. Second guessing the rule sets and thresholds designed to detect them makes the criminals increasingly dangerous as they can become invisible to all detection capabilities. The result is a frustrating game of cat and mouse, adjusting thresholds and changing rule sets.

The Solution: Combining AI and Network Analytics Technologies

Global AML spending is predicted to rise from US $5.9 billion in 2013 to US $8.2 billion in 2017

In order to understand how artificial intelligence can outsmart fraudsters, businesses must understand more about artificial intelligence. There is a myriad of misconceptions about AI, the role of AI within a business, and what lies at the heart of AI: data. When businesses think of AI, many automatically think of machine learning or deep learning, which for some is the perfect solution; however, for fraud prevention, it isn’t appropriate. For AI to autonomously process data and learn how to detect fraud, it would generate huge volumes of false positives and in fact provide investigators with no real insight as to why something has been flagged as fraudulent. There isn’t a set methodology to commit fraud, and the machines just can’t keep up.

To outsmart the fraudster, the system must understand the fraudster, how they scale their operation to make money and then they will be able to adapt their detection methods to avoid reverse engineering by the bad guys. This can be achieved through a hybrid model, beginning with humans teaching the AI the basics – just as a human would learn – with a point to start and then develop a complete picture. They will learn from frauds they have caught and look for any comparable variables in order to identify frauds of a very similar strategy that have not yet been caught.

To use AI to its full capacity, banks need to understand the network and its wider context. Understanding the network and its wider context is the first step into reducing false positives and becoming more efficient and effective in the fight against criminal activity. Contextual monitoring uses entity and network analysis techniques, in combination with advances analytical methods, to detect anomalous and suspect activity.

When understanding a customer, whether fraudulent or not, it is important to take all entities into consideration: shared identities, bank accounts, addresses, transactions, time correlated events, relationships, the list goes on. When fraudsters industrialise their activity to increase profits, they regularly re-use part of identities, or leave parts behind to trick the system and miss any connections to related fraud. By replicating your best investigators’ approach across a networked data, the AI will alert far more accurately and then provide the investigator team with a picture and wider, detailed explanation of the fraud.

However, this does beg the question – how does one track frauds that are completely new, with no trail and understanding of the scale? The beauty of networks is that the network continues to remain as the same starting point and then apply a range of techniques to detect abnormal behaviour. Peer group analysis, clustering and outlier analysis must be widened beyond purely transactional activity. It’s important for investigators to remove their blinkers and focus more widely on a range of entities and objects such as people, addresses and mobile numbers, for example. Businesses then must then take this a step further and overlay a complex set of techniques that make it nearly impossible for a fraudster to cheat the system and subsequently fly below the radar. This industrial level of networking is what will ensure businesses outsmart the fraudster.

Is Fraud just a Cost of Doing Business?

Historically businesses have often adopted the approach of putting in some basic measures to tackle fraud. Banks across sectors have installed current AML systems as a reaction to increasing regulatory pressure, rather than building bespoke systems. Over 25 per cent of financial services haven’t conducted AML/CFT (Combating the Financing of Terrorism) risk assessments across their global footprint. This practice is no longer acceptable for several reasons, most notably the cost of fraud is still being passed back to the consumer and money laundering is becoming a key means to financing terrorism. However, according to Wealth Insight, global AML spending is predicted to rise from US $5.9 billion in 2013 to US $8.2 billion in 2017, promising new opportunities for banks to create stronger barriers to fight criminals through the integration of AI and network analytics. Taking action will safeguard your reputation, save money and avoid regulatory action.