The good, the bad and the fraudulent - Sean Neary investigates – Featurespace

More news

March 02, 2017

The good, the bad and the fraudulent

Understanding behaviour with machine learning - Sean Neary, Financial Services Subject Matter Expert at Featurespace, explores.

For the last 10 years, I was managing fraud and risk prevention within the banking industry, seeing first-hand the increase in fraud attack types and frequency that is constantly taking place. Criminals are continually improving the sophistication of their methods, and are taking advantage of increased online activity to find the weakest links in prevention methods and commit large-scale fraud.

On top of the challenge of identifying evolving fraud types, the number of fraud alerts produced for analysts to manage is often far too many for a team to deal with on a daily basis.

So, faced with these challenges, how can businesses truly understand their customers to reduce both fraud risk, and the number of false alerts that risk management teams have to deal with, while maintaining acceptable fraud detection rates and customer experience?

Fraud predictions for 2017

The answer lies in advanced machine learning and adaptive behavioural analytics, which gives organisations unique insight into informed business decision making and control over managing individual customer risk.

This is more important than ever for 2017, because criminals are starting to identify new vulnerabilities that impact payment processors upstream of where  fraud systems usually spot an attack at the transaction stage.

It’s a type of attack that I witnessed increasing within financial services over the last 10 years working within fraud operations teams. Criminals  manipulate the standard authorisation message to push systems into Stand In Processing (STiP) or to mislead the rulebase into thinking it was a secure transaction. This happens upstream of where many traditional fraud systems look for vulnerabilities. So organisations involved in protecting their consumers during the authorisation and payments process need to be alert to the need to monitor behavioural activity holistically, throughout the entire authorisation stream. 

This is one example of where new fraud attacks are increasing. Areas that we can expect to see fraud attacks evolving particularly fast in 2017 include:

Social engineering attacks – commonly targeted at vulnerable and elderly banking customers – are increasing year on year. This takes place when a genuine account holder is manipulated by a criminal impersonating the bank via phone or email. Banks and financial institutions are starting to counter these attacks by tightening authentication with biometrics and two-factor authentication. However, the criminals perpetuating these methods are increasing their physiological manipulation techniques, in an attempt to target a wider population. This is already causing concern within the fraud prevention services – you can have all the prevention tools and techniques in your arsenal, but you cannot control what information your customer gives away. Tackling these social engineering attacks requires a sophisticated adaptive behavioural analytics approach to understand the significance of every subtle behaviour change, in real time.

Authorisation stream fraud attacks – as discussed earlier – will dramatically increase as criminals look for fraud opportunities throughout the entire authorisation stream, including attacks on the host/processing systems.

Risk of increasing fraud operations costs because existing fraud systems cannot take the strain of these new threats, which results in heavy
recruitment into fraud operation teams to try and keep up with spotting fraud and dealing with vast volumes of false alerts.

The evolution of fraud prevention – machine learning and AI

What I’ve learnt from managing fraud technology  over the years, is that it is essential that fraud systems keep one step ahead. To do this, a modern fraud solution requires three core capabilities:

1.    The right data:  Gone are the days where each fraud solution requires a specific data set. To enable true customer profiling and anomaly detection, a modern fraud management system requires end-to-end information, including both monetary and non-monetary. The good news is that the majority of businesses are already capturing this type of data about their customers – now organisations need modern fraud systems which can help them put this data to good use in making informed business risk decisions.

2.    Real-time flexibility:  Businesses change and evolve, and so does their product offering and architecture. This is why it’s important to ensure that an organisation’s fraud management system offers easy integration and open APIs.

3.    Self-learning capabilities:  With the rate that fraud evolves, it’s no longer enough to have a fraud management system that only looks at behaviour from one point in time – data becomes out of date too quickly.

To truly understand behaviour and spot the anomalies, organisations need to understand each individual customer’s behaviour in real-time.  people, likes businesses, change over time – and their needs (and consequent behavioural activity) can evolve throughout the year. Self-learning fraud systems constantly learn and adapt to new information to meet these challenges. This is why adopting machine learning is critical to successful fraud prevention and customer management.

Identifying good behaviour from bad

To spot and block these fraud attacks quickly and efficiently, organisations need to be identifying anomalies at the level of accounts, merchants,
individual cardholders and multiple locations, as well as holistically within their host/processing systems. 

Luckily, there is good news for organisations trying to tackle these challenges. New machine learning systems – which use adaptive behavioural analytics to build individual statistical profiles in real time automate the process of viewing events in context, building a deep understanding of every single customer.

By monitoring every event and transaction taking place in real-time and from multiple channels, fraud attacks stand out and genuine customers are easy to recognise. All this takes place within an automated fraud prevention system, which automatically updates itself as new fraud types are identified.

Balancing genuine customers with fraud prevention

For every fraud and risk operative, there is always a balance to be made between stopping fraud without blocking genuine customers. Reducing customer friction at every interaction has a huge impact on both revenue streams and customer experience – more important now than ever, with 24/7 customer demand for services.

To succeed, organisations need to focus on finding automated ways to fill the gaps which currently make your customers vulnerable to fraud.
It’s an approach being adopted by TSYS, the largest payment processor in the United States. TSYS wanted to strengthen its position in faster payments using machine learning to provide clients with actionable insights in real-time. They’re implementing an advanced machine learning engine to protect their clients from fraud while providing a seamless customer experience.

What can organisations do next?

Faced with the challenge of identifying ‘good’ customer behaviour from the bad behaviour of criminals, why wait for incidents to occur before finding the vulnerability in your fraud management systems?

Using advanced machine learning systems with adaptive behavioural analytics to prevent fraud is enabling organisations to balance efficient, robust fraud checks with a frictionless customer experience.

It’s a great opportunity for organisations to take control of tackling financial crime, while accepting business revenue from their genuine customers – all made possible by the power of advanced machine learning. 

Sean Neary will be speaking at the eCrime Fraud Forum in London on 8th March 2017. Get in touch to meet us there.