By: Ronan Kerouedan, SVP of Global Value Solutions Consulting at Coupa
Financial fraud is a constant threat to businesses. According to PwC’s 2020 survey of economic crime and fraud, 47% of global organisations have been victims of fraud, of whom 13% said they had suffered fraud-related losses of over $50 million.
While there are many different types of fraud, enterprise fraud — such as duplicate invoicing, fraudulent expenses, and asset misappropriation — is particularly problematic for businesses. Asset misappropriation, where revenue is either accidentally or intentionally lost or stolen by employees, is present in 86% of all fraud cases, according to a report from the Association of Certified Fraud Examiners released last year, which also found that organisations are losing 5% of their revenue to fraud each year. These revenue leakages severely impact an organisation’s ability to grow and execute its goals.
Despite enterprise fraud consistently wiping off millions in revenues each year, attempts to combat it have fallen woefully short so far. Spotting fraud is incredibly challenging, which is not helped by the fact that much of this work is still done manually. People make mistakes and understandably cannot spot every fraudulent invoice or expense claim, especially if there are hundreds or thousands to process. However, artificial intelligence (AI) can help businesses stop fraud in its tracks. Companies must embrace AI to stop the rot – it is one area where AI can make a substantial, immediate and tangible impact on any organisation.
The trouble with tackling fraud
Detecting financial fraud is challenging, costly, and time-consuming, and there are major problems with the traditional techniques used for fraud monitoring.
Consider duplicate invoices, which occur when multiple invoices with slightly different attributes (supplier name, invoice numbers, dates, and sometimes amounts) are submitted for the same goods or services. When undetected, multiple payments are processed contributing to significant amounts of spend leakage. In just the Coupa community alone, we have detected an average of $1.7 million per year in duplicate invoice spend per customer.
Most of the technology used to detect duplicate invoices are rules-based, making them hard to maintain. They work by looking for exact matches of the attributes listed above. Unfortunately, this will miss many duplicates, because in the real world these duplicate invoices will have slightly different names and attributes. This often happens because of multiple errors during the process of entering invoice data from invoice images into the system.
Some of these technology solutions do already use AI systems, but they can often be very basic and lack the Natural Language Processing (NLP) engines needed to flag duplicates. Companies using non-specialised AI for fraud detection will continue to miss these fraudulent invoices and leak revenue.
Finance teams also have trouble accurately monitoring and identifying non-compliant spend. This is often a consequence of businesses building up gradually from more humble beginnings, where processes are manual, such as on spreadsheets or paper. But as a business grows, these processes do not scale and can quickly become time consuming and error-prone.
In addition, existing auditing methods generally require companies to audit a random sample of data. This prevents them from getting full visibility into their data because they’re only looking at a small segment, which means they could miss critical errors or fraud. Also, auditing usually detects fraud after-the-fact, meaning customers must then spend time and effort reclaiming payments, and this missing money can impact cash flow.
Bonfire of the silos
A direct result of businesses scaling up and building upon the manual processes they started out with, is that key functions like Procurement, Finance, Accounts Payable, and Treasury become entrenched in silos.
Enterprise fraud thrives in the shadows created by a company’s silos, as the gaps between systems which can’t talk to one another form dark alleyways for criminals to lurk.
Businesses must cut across these silos to detect fraud. Because this is resource-intensive, companies often take a manual approach to focus on one area, such as expense management. As a result, they may miss fraud occurring elsewhere, for instance in invoicing or procurement.
Similarly, companies may look to anti-fraud products designed for a specific silo but they will still miss the bigger picture, as well as being too costly and inefficient to scale across all areas of spend.
A first step to tackling fraud must be to dismantle silos and bring all spend-related data to a single, unified platform. This allows disparate systems to begin working much more intelligently with each other, and enables businesses to start shining a bright light into those dark alleys.
AI in action
With this data on one platform, it is much easier to apply AI technology, which has the speed and power to automatically analyse and audit 100% of transactions, rather than just a small sample.
In addition, advances in NLP and machine learning mean that AI can be trained to understand semantics. So instead of identifying duplicate invoices by looking for exact attribute matches, AI can look for patterns and variations to spot duplicates. Humans are simply not as good at determining patterns as AI, meaning that the technology can better identify fraud. AI can also spot fraud in real-time, thus preventing or intercepting suspicious payments before they are made.
At Coupa, we have developed an AI-powered fraud detection system called Spend Guard, which has been trained to spot and identify fraud using our mass of community data. At one large consumer goods customer, we found a duplicate invoice of $1.5 million within a few weeks of going live with Spend Guard. This invoice had already been approved through their previous processes.
In another example, we helped the industrial services provider Brock Group reduce fraudulent spend. Brock employs more than 13,000 employees across the US and Canada. With so many workers submitting expenses, it was challenging for the company to review all expense reports for suspicious and erroneous transactions.
But within the first hour of activating our AI, Brock identified tens of thousands of dollars of suspicious transactions, including duplicate transactions and even discovered an employee who had submitted thousands of dollars of seemingly legitimate small-dollar expenses that were fraudulent.
AI has the potential to all but eliminate fraud from enterprises – not in 10 years’ time but today. The technology works and can stop the vast majority of all business fraud almost immediately, helping companies to save money. With the economy still recovering from the struggles of the pandemic, preventing fraud must become a top business priority as organisations look to build resiliency
