ACI Worldwide’s Director of Data Science – Jimmy Hennessy: The tech to fight emerging fraud


Writing for Payment Expert, Jimmy Hennessy of ACI Worldwide details the challenges that await next year when it comes to combating fraud, especially as an intensified period of transactions looms.

There are few constants in life, and one of them is change. From the way businesses operate to how consumers purchase goods and services – since the start of the pandemic we have seen massive changes in the last few years – with much of our personal and business affairs now being conducted online. On the flipside, this change has come with new opportunities for online fraudsters. Indeed, more than one in 10 people in the UK (12.1%) have fallen victim to payments fraud in the last four years. 

With big shopping events, such as Christmas, around the corner, we will start to see new fraud trends emerging – especially around fake promotions. Retailers are expected to spend around USD 9.6 billion annually, between 2021 and 2025, on fraud detection and protection. However, when it comes to monitoring and analysing online fraud trends, the continuous and rapid evolution of fraud ‘vectors’ can often be too much for existing machine learning technologies.

Despite this, there is technology that can help combat these increasingly advanced fraudsters, which additionally promises to lower operational costs and free up resources. A new iteration of machine learning capability, called incremental learning, is specifically suited to tackle rapidly evolving fraud and aid in preventing fraud from happening in the first place.

Out with the old machine learning 

Traditional machine learning makes up a crucial part of a merchant’s fraud strategy, but often it finds it hard to process rapid changes and requires human intervention. As customer preferences, purchasing trends and fraudster behaviour continues to evolve, the harder it is for these machine learning models to keep up. 

Each machine learning model can take a few months to be built and deployed and then it takes another few months to train over the large data sets that represent a single moment in time. The shorter the time between new trends, each requiring a new model, the quicker the old models deteriorate. This results in a perpetual cycle of trying to catch the latest fraud trend and as such, trying to get in front of a fraud trend can be highly unlikely. 

In with the new incremental learning

Put simply, incremental learning models can think for themselves, by continuously updating and refreshing in real-time. While traditional machine learning models need to be completely retrained. Incremental learning models can keep up with the shifting fraud patterns.

By operating in the present, with new data being fed into the model every day, these models can spot new behaviours and trends as they happen, causing a reduction in fraud losses by up to 75% and aiding in improving fraud detection by up to 85%. These newer models can go longer without degrading, by detecting if a change in fraud behaviour is enough to deteriorate the model and if the case, can update using what was recently learned. 

Incremental learning models are more effective because, in essence, they do what needs to be done. As they do not need as much human intervention and do not need complete model rebuilding and retraining every time they degrade, they can free up resources. Further, by using the most recent fraud intelligence and consumer behaviour data it continuously adjusts and maintains a high level of protection against fraud. Allowing merchants to focus their time on other areas of work that may require it. 

With rapidly changing consumer behaviour comes an increase in rapidly changing fraud patterns. . And incremental learning models are part of the answer to combat the new fraud threats.  

They require less human intervention and produce fewer errors whilst protecting merchants from fraud to a larger extent than traditional machine learning. Further, such models can keep up with fraudsters by updating and learning from the latest consumer behaviour data and fraud trends. They operate in the background and allow merchants to focus on other key areas of the business.