Best AI/ML approaches for fraud detection and prevention

By Anna Lykourina, EMEA Fraud Analytics Expert at SAS

In the past, the fight against fraud has been a bit hit-and-miss. It has relied on auditors to identify patterns of behaviour that just didn’t quite fit. They often only detected problems months after the event. And then organisations had to claw back stolen funds through legal processes.

In a world where transactions happen in under a second, however, this is no longer acceptable. We need to be able to detect fraud immediately, if not before it happens. Customers want safe and protected data that is not vulnerable to identity theft through company systems. But they still want to be able to pay online and in seconds. The stakes are high, but fortunately new tools and techniques in fraud analytics are enabling companies to stay ahead of fraud.

Trusting machines to do the work

Machines are much better than humans at processing large data sets. They are able to examine large numbers of transactions and recognise thousands of fraud patterns instead of the few captured by creating rules. On the other hand, fraudsters have become adept at finding loopholes. Whatever rules you set, it is likely that they will be able to get ahead of them. But what if your system was able to think for itself, at least to a certain extent?

New approaches to fraud prevention combine rules-based systems with machine learning and artificial intelligence-based fraud detection systems. These hybrid systems are able to detect and recognise thousands of fraud patterns and learn from the data. Automated analytical-based fraud detection systems can reveal novel fraud patterns and identify organised crime more consistently, efficiently and quickly. This makes them a good investment for businesses across a wide range of sectors, including public sector, insurance, banking, and even healthcare or telecommunications.

How, though, can you harness analytics as a tool in your fight against fraud?

Identifying needs and solutions

The first step is to identify which options you need. Probably the best way to do this is through a series of company-wide workshops with the fraud analytics experts to determine what analytics you need, which data to include and techniques to use, and what results to report. They can also identify the ideal combination of rules-based and AI/ML approaches to detect fraud as early as possible.

Companies looking towards advanced analytics for fraud detection will need to make a number of decisions. They will need to optimise existing scenario threshold tuning, explore big data, develop and interpret machine learning models for fraud, discover relevant information in text data, and prioritise and auto-route alerts. There may be industry-specific decisions to make, too, such as automating damage analysis through image recognition in the insurance sector. By automating these areas, companies can both significantly reduce human effort – reducing costs – and improve their fraud detection and prevention.

Benefits of an analytical approach to fraud detection and prevention

Companies that are already using an analytical approach for fraud prevention have reported several important benefits. First, the quality of referrals for further investigation is better. Investigators also have a much clearer idea of why the referral has been made, which improves the efficiency of investigation. Analytics also improves investigation efficiency by reducing the number of both false positives (that is, alerts that turn out not to be fraud) and false negatives (failure to spot actual frauds). This improves customer experience and reduces risk to the company.

Analytics makes it possible to uncover complex or organised fraud that rules-based systems would miss. Companies can group together customers and accounts with similar behaviors, and then set risk-based thresholds appropriate for each scenario.

There are several sector-specific benefits too. For example, insurance firms can identify fraudulent claims faster to prevent improper payments from going out. Claims investigation is likely to be more consistent because claims are scored through technology, algorithms and analytics, rather than by people. Finally, it becomes possible to shorten the claims process through automated damage analysis. It is no wonder that organizations across a wide range of sectors are placing analytics at the heart of their anti-fraud strategy.