Machine Learning in fraud prevention: why it’s time for a level playing field

Fraud-management professionals at every level of industry, including banking and financial services, retail and healthcare, are facing increased risk levels amid a perfect storm of threats. Multiple factors, from the increasing digitalisation of society to rocketing inflation and the associated squeeze on personal finances, are conspiring to magnify the risk of fraud for businesses of all sizes.

Experian’s UK Fraud Index for the first quarter of 2022 shows fraud is on the rise in areas such as cards, asset finance and particularly loans, where fraud has reached a height last seen in Q3 2021, which itself was the highest rate for three years. And our 2022 UK&I Identity and Fraud Report highlights the fact that fraud losses in the UK increased 22% in 2021, with 90% of those cases originating online.

But exposure to these risks is by no means spread equally. Larger, more sophisticated organisations may already have the internal expertise, the technology resources and the access to data they need to minimise fraud risk. Younger and mid-market organisations, such as start-ups and scale-ups, may well have less protection in place and are therefore more exposed to risk.

The risks of inadequate protection

This is particularly the case when it comes to the capabilities around data and analytics, based on highly sophisticated machine learning (ML) fraud solutions. This is a branch of artificial intelligence (AI) in which a computer programme analyses vast datasets, identifying trends and patterns that larger players are increasingly taking for granted.

In an ideal world, all new and small-to-mid-market players would have access from day one to an effective and affordable ML-based solution that enables more accurate decision-making within a shorter timeframe. But they don’t – and for such companies, this is an evident danger. Without the budget to invest in a team of analysts and costly technology, the threat is threefold:

• First, fraudsters will be aware that the protection a young business has in place is likely to be rudimentary when compared with the defences surrounding an established leader, making it an obvious target. In addition, once they have identified a weakness, fraudsters will continue to target the business until it is adequately protected, potentially slowing its growth before it has had the opportunity to build economies of scale.
• Second, a business of this scale is at heightened risk of reputational damage among consumers. There are two aspects to this danger. On the one hand, if a time-consuming security check makes it too slow to process an application, there is a chance that the prospect may not only drop out of the process but also talk in negative terms about the company before it has had the opportunity to get properly established. Worse yet, an ineffective security regime may expose that customer to fraud, resulting in even more damaging word of mouth and loss of credibility at a critical moment in the company’s development.
• And third, a young organisation without specialist fraud-management expertise might not fully understand the nature of the threats facing it. For example, while it may be capable of addressing instances of first-party fraud, where applicants for a service distort the truth when filling in an application form, it might not be aware that behind an apparently unsophisticated front there may be a highly motivated and experienced financial-crime network.

Machine learning is essential for every player

As our 2022 UK&I Identity and Fraud report states, “AI and Machine Learning are now key technologies for online customer identification, authentication, and fraud prevention”. This is because the sheer quantity of data points involved in establishing a consolidated view of the overall fraud risk of an individual requiring fewer referrals for assessment, and producing better outcomes for good customers.

As a result, today’s optimal real-time fraud-detection techniques are invariably based on ML.

An application doesn’t even have to be evidently fraudulent for ML to pick it up. It can spot anomalies which an applicant, keen to be seen in the best possible light, would not regard as fraud so much as slightly blurring a detail or two. When traditional risking techniques are used, it may also not appear to be fraudulent. But if this results in an applicant being accepted who cannot afford to repay a loan or service credit-card debt, it can be every bit as damaging to that applicant as third-party fraud is to its consumer victims.

However, using ML that has learned from previous applications and their outcomes enables a more accurate risk profile to be created, covering the likelihood of the application being revealed as fraudulent at some point in the future. In other words, ML is as much about predicting the future as it is about preventing clear ‘black-and-white’ fraud at the point of application.

The barriers hindering ML for all

Three immediate barriers are currently preventing a ‘machine-learning-for-all’ vision from becoming reality:

• Resource and knowledge: younger businesses are less likely to have access to human resources such as data scientists who are highly familiar with the technologies involved;
• Expense: the technology is also highly sophisticated and can be correspondingly expensive; and
• Lack of historic data: a small start-up will not have the breadth and depth of data that big players can draw upon, and may not be part of consortia such as National Hunter and CIFAS in which members share industry data.

So, is there no such thing as a usable and effective solution for non-corporate organisations?

Whilst manual assessments are still widely in play, more accessible, automated solutions are now coming to the fore. After all, for as long as more established businesses have access to resources that are largely inaccessible to emerging competitors, the playing field will never be level.

Quite simply, those companies that aren’t using ML-based management solutions might be leaving themselves open to further exposure.

And that isn’t right.

At Experian, we believe passionately that through automated reviews of data, we can reduce the exposure to fraud and allow more “good” customers to experience a positive user journey – without interruption. By enabling smaller organisations and scale-ups with the capability to integrate Machine Learnt solutions into fraud detection, fraudsters will have less and less opportunity to slip through the gaps.

Find out more about our solutions that can help you meet regulatory obligations and protect against fraud.