You don’t very often hear about procurement fraud. It’s not really discussed, perhaps because it is often an internal problem, and therefore involves trusted employees and breaches of that trust. In terms of losses, however, it is the second-most important economic crime in the world. The average amount lost in a procurement fraud is 103,000 EUR. In other words, it is a real problem – and it is only likely to get worse.
The increasing complexity of procurement processes means that existing audit processes may no longer be sufficient to protect against procurement fraud. Common approaches to monitoring and auditing the procurement cycle rely heavily upon auditors’ skills to detect fraudulent behaviour – and this, of course, only happens at particular points in time. This can in no way be described as providing continuous prevention.
Procurement fraud is not usually a one-off crime. Perpetrators tend to start small, but their activities escalate over time as they discover what is possible. The average time before ongoing procurement fraud is identified is around 18 months, and many frauds go on for years. What’s more, the average time it takes to bring a case to court for prosecution is around four years. This is a crime with a long memory, and long-lasting effects.
Identifying and preventing fraud using data analytics
Fortunately, new ways of discovering and even preventing procurement fraud are available. An approach to procurement integrity that is driven by data analytics means that you can identify issues faster, reducing losses significantly. Data analytics is significantly more successful at identifying potential fraud than other methods of detection. PWC, for example, found that 25 percent of economic crimes can be discovered by using data analytics, which is a much higher rate of detection than any other method.
The reason for this is that a good analytical approach will use several methods to detect potential problems. Useful techniques include:
- Anomaly detection, to identify either historical or real-time behaviour that does not fit the expected pattern. Techniques may detect changes over time, note variations from the behaviour of a “peer group,” or even compare behaviour against predictions and models of either “good” or “bad” behaviour.
- Text mining can be used to investigate unstructured sources of data such as emails and social media, and again identify behaviour that does not fit the expected pattern.
- Associative linking identifies relationships between groups or individuals, for example, by phone numbers, addresses or bank accounts, or by transactions between them, such as providing references.
Of course, not all unexpected relationships, anomalies or unexpected behaviour will signal fraud. The idea behind all these techniques, however, is to identify issues that are worth further investigation.
Each of these analytics techniques is, on its own, useful, and can identify a particular type of fraud. Put them together, however, and a much better picture emerges. But even putting techniques together or using advanced analytics powered by artificial intelligence is not enough. Analytics, by itself, cannot identify fraud and take action to address it. It can only highlight potential patterns and insights.
What is needed is an extra ingredient: People
People are required to take the insights from analytics and turn them into tangible outcomes. Just like an indicator shows you where there might be problem and suggests areas for further investigation, so an analytical approach to procurement fraud – or any other form of fraud – only highlights probabilities. Someone with financial expertise is then needed to investigate the issue further and identify whether there really is a problem. The insights from analytics must be interpreted and contextualised.
People are also needed to consider whether there are any systemic problems that may need to be addressed. Do the business rules, for example, lend themselves to particular fraud scenarios? Are there organisational changes that are needed to counter these problems? Analytics by itself cannot do this – but it can point specialists in the right direction and provide evidence to support change.
There is also a final issue. Ethical considerations also require people to be involved. As artificial intelligence-driven analytics systems become more widespread, there is increased potential for bias. AI systems learn from the data that they encounter. If that data is biased, then the learning, and therefore future outcomes, will also be skewed in some way. Experts are needed to look out for and counter this type of institutional bias.
The combination of analytics and experts, however, is increasingly powerful enough to start to detect and counter procurement fraud effectively while remaining fair to all those involved.