Data Analytics for Fraud Detection and Prevention

Using data analytics to keep fraud at bay

In the 1970’s, criminologist, Donald Cressey explained fraud as a combination of three factors known as the Fraud Triangle. This model explains that fraud is likely to occur when pressure, opportunity and rationalisation are present. As businesses evolve and processes become further digitalised, new opportunities are created for fraudsters to exploit. Fraud schemes have become increasingly complicated and thus the traditional approach to fraud must be improved to remain effective. The complexities of modern-day fraud schemes and its mechanisms to conceal such fraud, continue to develop and surpass conventional anti-fraud measures. Data analytics for fraud detection, often referred to as fraud analytics, has become a fundamental solution for both external and internal auditors to prevent and detect fraud.

While data analytics cannot directly identify fraud, it can effectively identify and highlight anomalies and potential indicators of fraud. Data analytics, as part of a risk assessment process, can also recognise vulnerabilities or deficiencies within a system or business process that may need to be addressed. As global data volumes increase exponentially, data analytics can alert users on areas and/or transactions more susceptible to fraud. In analysing an entire population, deeper, more insightful conclusions can be drawn, by performing comparisons, trends and even keyword searches over a complete database. This can facilitate a focused and streamlined approach in investigations resulting in the most efficient use of limited resources.

While data analytical tools primarily concentrate on consuming large transactional datasets, they are equally powerful in analysing master data files. This is a crucial component of fraud analytics in order to detect any manipulations of master data (for example, duplicate bank accounts, duplicate identity numbers, etc.) as well as any unauthorised amendments to master data. Data analysis software also provides the opportunity to apply statistical models, such as Benford’s Law, to datasets. Such models enable users to perform otherwise scientific analyses, in a more user-friendly manner that can contribute significantly to any fraud detection process. Using such data analytic techniques will also ensure that any evidence obtained is based on data that is reliable and complete.

Fraud by its nature, is a relatively slow crime that occurs over a period of time and may often go undetected until after some loss has been suffered. Fraud analyses and physical reviews consume valuable time and company resources which may deter fraud assessments from occurring more frequently. A key benefit of using data analytics is the ability to automate repetitive tasks. Such analytics can continuously analyse big data, without human intervention, notifying users to any exceptions identified, which can then be investigated timeously. Data analytics for fraud detection goes beyond the analysis of accounting data. User profiles, access rights and even login details can be analysed for any inconsistencies. Structured as well as unstructured data can also be analysed to ensure a company wide fraud approach.

Prevention is better than cure. The same logic applies to fraud. Preventing fraud is more effective and economical than detecting fraud. Fraud analytics can be used to proactively identify fraudulent or suspicious transactions and be used to develop anti-fraud controls. According to the Association of Certified Fraud Examiner’s 2020 Report to the Nations, organisations that use proactive data analytics report fraud losses that are 33% lower than organisations that do not use data analytics. In this way, fraud is prevented at the earliest stage possible and stopped from developing into a complex scheme. Exceptions can be investigated promptly, and corrective action can be taken immediately. However, if for any reason, these preventative measures are bypassed, data can still be thoroughly interrogated by various other techniques to detect any occurrence of fraud.

In conclusion, the use of data analytics in fraud detection and prevention has proven to be invaluable. However, the investment in a functional data analysis process requires additional costs to be added on already constrained budgets. With this in mind, the cost of fraud far exceeds monetary value and can be detrimental to one’s reputation as well as the going concern of an entity. With the increased pressure on auditors to identify fraud, new techniques must be embraced to detect ever evolving fraud schemes.

Salim Mohamed
Data Analyst, Johannesburg