Procurement fraud and error detection – A global £6.2tn opportunity
Procurement fraud has been a forgotten topic and many organisations have accepted the financial implications as part of their cost of doing business. However, most organisations are unaware of the scale of the problem. Estimates for the true cost of procurement fraud vary anywhere between 3 percent to 10 percent of annual revenues. Based on these estimates, the global cost of procurement fraud adds up to a staggering £1.8 trillion to £6.2 trillion annually.
In 2013 grant Thornton published a report that identified estimated the impact of fraud in the construction industry alone at £665.53 billion, projecting it to grow to £1.16 trillion by 2025.
While the cost of procurement fraud dwarfs the GBP of most countries, organisations have done little to nothing to address the problem. Those organisations that have acted, have made improvements to their procurement practices with enhanced controls, vetting procedures, and KYC requirements. Unfortunately, these improvements have fallen short in preventing offenders from exploiting organisations. For example, in recent years the UK government has put an increased emphasis on more responsible spending of public funds of which the detection and prevention of procurement fraud is a critical area of focus. Nevertheless, governments manage to detect less than 1 percent of fraud committed.
Most fraud solutions on the market are focused on transactional fraud in commerce and banking using a sophisticated rule base and machine learning technologies. The reason why these systems have not been applied to procurement is that procurement fraud much more subjective, nuanced and not always of a transactional nature. For example, collusion on bids has a financial motivation and impact but the actions themselves are not transactional in nature. In addition, procurement fraud is also highly varied and often tailored to internal processes and policies which make organisations vulnerable.
Using an AI-based detection approach
We have been working with several clients to tackle the problem of procurement fraud by building our Booly fraud detection engine. Through a combination of business rules, parameters and modelling we have recently helped a large government organisation to put in place an advanced fraud detection capability based on Booly.
The solution uses a variety of approaches and models to construct relationships between transactions, suppliers, and processes. The aim of the system is to make contextual and dynamic judgements on whether a transaction is an anomaly. It is not enough to look at this from a point in time as procurement relationships evolve and deal values and spend patterns often change.
We therefore took a different approach to traditional fraud identification which we call “process originated inductive and deductive learning”. Rather than sifting through data and trying to identify abnormally high or frequent payments, we develop a data representation of an organisation’s relationships with it’s suppliers. In a nutshell this means we learn what a normal relationship with any supplier looks like at any point in time and develop risk indicators when there are events that are “unusual”.
In the example above we deployed several models to provided different “lenses” to look at fraudulent behaviours at the transaction level. We processed annual spend of £500 million and identified 15 percent or £80 million of transactions as fraudulent.
While these were impressive findings, we were more surprised by the subtlety of the identified cases. When people think about fraud, conversations quickly end up debating funds being transferred to secret offshore accounts to fund lavish lifestyles for individual offenders. While these cases certainly exist, the bulk of fraudulent activities we uncovered were of a systemic nature, exploiting vulnerabilities in weak procurement processes.
Post award cost inflation, undisclosed partnerships and conflicts of interest were some of the most significant contributors to the £80 million identified.
In conclusion, organisations must urgently upgrade their capabilities to detect fraud within their supply chains. While trust is a noble virtue, governance is becoming increasingly more important to operate efficiently. Deploying machine learning applications such as our Booly engine will not only recover the initial investment within a matter of months but also boost the bottom line and add additional working capital. In addition to the immediate financial benefits, identifying organisations engaging in fraudulent activities can also provide insights into the resilience and health of a supply chain.