Around the world, combating public sector fraud is a major problem. "Benefits cheats" get the most media attention, but are by no means the biggest loss. Tax fraud is also a huge issue. And some estimates suggest that procurement fraud could dwarf both in terms of both scale and complexity.
Fortunately, however, even as fraud gets more imaginative and more complex, better and more effective analytics techniques, including machine learning and artificial intelligence (AI)-based algorithms, are becoming available. These are helping to save taxpayers’ money at both local and national levels in a number of ways. I asked Colin Gray to weigh in on some options that are already available today.
How has analytics changed the way fraud is being targeted?
Traditional methods for detecting fraud relied on the experience of individuals looking at cases. Fraud detection was invariably reactive and often months out of date because it relied on audits or whistleblowers. New analytics techniques, however, can be used to build a profile of a fraudulent claim. This profile can be used to compare past patterns with the present, and stop transactions for further investigation, based on the probability that they are fraudulent. Preventing, rather than reacting to fraud, should save millions in taxpayers’ money.
What tactics work with complex situations?
One of the issues in detecting fraud, especially in government, is its sheer volume and complexity. Rules-based systems are very good for detecting fraud that follows a recognised pattern. In other words, a fraud that has been used before. In Howard Davia’s book Fraud 101, he estimates that only 20% of fraud is exposed and in the public domain, and it is these cases on which rules-based systems are developed. These systems simply cannot detect new forms or patterns of fraud. An analytical approach needs to be hybrid, using the latest AI techniques and combining rules with anomaly detection and outlier investigation, as well as predictive methods, to detect both existing and new forms of fraud with more accuracy.
How have social networks come into play?
Social network analysis is a way of uncovering links between individuals or groups. It looks at data like names, addresses, telephone numbers, bank accounts, internet access and user devices, and then draws on analytics, including text-based techniques, to find out when and how people are connected. This can be used to identify organised fraud networks involving large numbers of players – and big money. One such case, for example, found that over 200 people were all using the same phone number – and most of those people turned out not to exist.
How important are data sources?
One of the ways in which analytics is helping to combat public sector fraud is that it is easier to analyse large volumes of data. Governments have no shortage of data, but previously struggled to use it effectively for fraud detection. Analytical capacity in data centres or in the cloud means that large volumes of data can be analysed, but also that data can be brought together from multiple sources: government records, health records, social media information and other third-party data. Bringing together data from several sources often results in more insights – and more fraud detection.
Why is procurement fraud attracting more attention?
Procurement fraud is big business in government because of the large sums of money involved. It is also a problem because of the reputational damage it can cause. Just as with benefits or tax fraud, a hybrid approach to detection can uncover potential procurement frauds by looking at patterns of behaviour and identifying anomalies. This can stop payments being made, or prevent fraudulent suppliers being set up. Social network analysis can be particularly important.
How is the overall risk of fraud being handled?
Fraud prevention has traditionally been a case-by-case issue. Detection traditionally has been about investigating individual cases or anomalies, and preventing or stopping that one case. However, analytics does not just allow a preventive approach. It is also enabling governments, in particular, to move to a more risk management-based approach by looking at trends in waste, abuse and fraud, and managing all three more effectively as potential risks.
Changing to a risk-based, analytical approach may mean changing how data are collected and held. It is all very well saying that data can be taken from a range of sources and used for fraud prevention purposes. However, a more strategic risk management-based approach to fraud detection may require a strategic look at how data are collected and held. It may, for example, make more sense to break down organisational data silos to bring data together more systematically than simply combine data each time you need to do so.
Help in addressing prevention of fraud in the public domain
I hope you’ll be able to join both Colin and me in sharing insights on preventing fraud in the public domain. Fraud is one theme at the Government & Health Care Executive Experience in Milan on Oct. 21. Executives and domain experts from government and health care across EMEA and APAC will share their experiences of use cases in digital transformation. Find more info here on how to register.
Also join the effort to stop fraud by sharing your voice on Twitter for a #SASChat. Mark your calendar for Thursday, 3rd October, at 14 CEST. We'll be discussing these questions:
- What are the major types of fraud affecting your organisation?
- How much of it has an element of internal collusion?
- What methods do you use to detect fraud?
- What frauds aren’t we finding?
- How can we help fraud SIU (special investigator units) get better at finding and stopping fraud?
Here's everything you need to know to participate in this #SASChat! We hope to see you online!