In principle, the best method of mitigating fraud is to set up a strong detection system. A system perceived as unbeatable will mean that potential fraudsters are discouraged from attempting anything and move on to easier targets.

The premise of deterrence is that a system that detects fraud and abuse (even the complex cases involving collusions and undeclared corporate interests) will also signal potential fraudsters to move on. This system must then be combined with targeted communication about the solution's success in identifying loopholes and culprits.

However, the issue is that you need to be very good at detecting fraud to have this deterrent effect. The only way to achieve that is by harnessing multiple different analytical techniques and using them together—and then telling the world about your success.

A better understanding of fraud

We might add to Louis Freeh’s quote below that greediness, in the end, often brings down fraudsters. On average, fraud that is detected has been going on for around 16 months by the time it is detected, usually using manual and semi-manual systems.

Data analytics systems can continuously adapt to changing trends for more efficient investigations, audits, compliance, and forensics. Analytical systems can be built to detect anomalies in customer payments, procurement spending, travel, and expenses. This type of solution surfaces real-time scored alerts for investigation teams to determine if the abnormalities are related to error, process breaches or intentional.

The key is to detect genuine problems without generating too many false alerts. This requires combining different types of analytical solutions.

Analytical solutions for fraud detection

The range of analytical solutions that are often used for fraud prevention and detection includes:

  • Predictive analytics to identify likely areas where problems may occur or potential vulnerabilities in the system. Predictive analytics systems can examine vast quantities of data on previous frauds to predict which people or situations are most likely to be an issue.
  • Machine learning and text mining can harness data that has been challenging to analyse, such as text descriptions on invoices, purchase orders, bid documentation and emails. Sentiment analysis techniques such as word clustering can identify words or phrases that may be suspicious, such as characteristic spelling mistakes or the use of particular terms.
  • Anomaly detection is part of unsupervised learning and looks for ‘unknown unknowns.’ It looks at patterns to identify outliers. This technique is also powerful because it is difficult to simulate the ‘right’ level of anomalies in a fraud, making frauds easier to detect.
  • Entity resolution and link analysis to highlight links between people, places and times and flag up individuals who represent a higher level of risk. Entity resolution is useful in combining records about the same person and cleansing data from multiple sources. The real challenge here is to identify problematic links. For example, do two people have the same address because they use the location for fraud or because they live in the same tower block?
  • Business rules are often an excellent place to start in fraud detection because they are usually put in place post-hoc to prevent future frauds. They tend to generate too many false positives to be used alone and are most helpful when ranked in order of importance. However, they also provide business logic. They often surface data issues and process breaches, which can be beneficial in identifying opportunistic fraud.

Overall, the system works by combining signals and context to determine the severity of an alert. For example, a rapid change of account number could be because the company is a start-up or someone has made a mistake. However, if the account is changed and a payment is made before it is returned to the original number, there may be an issue. Machine learning algorithms can help weight scenarios and improve scoring.

Analytics as a fraud deterrent

How does this relate to deterrence? Deterring wannabe fraudsters can only be done by sharing facts about successful fraud detection stories. The more stories we share of how analytics can be—and has been— successfully used to eliminate all the usual loopholes fraudsters try to exploit, the more likely we are to deter attacks. At the same time, we still need to improve fraud detection systems to stay ahead and reduce the likelihood that attempted fraud will be successful.

Join our virtual roundtable on September 19, where we will discuss this topic in more detail with experts from a major UK infrastructure project, KPMG and more. Any ACFE members joining us will be credited with CPE.


About Author

Laurent Colombant

Laurent has been helping customers tackle financial crime using NLP, ML and analytics since the year 2000. After focusing on sanctions screening, anti-money laundering, payment fraud and terrorist cell financing he is now working to address Continuous Compliance monitoring for SAS customers. This includes P2P, T&E, Know Your Supplier and Insider fraud modus operandi. He’s the NEMEA pre-sale lead for the solution and believes it’s the next to hottest fraud detection solution in the market.

Leave A Reply

Back to Top