Beating buyer fraud with AI

Most of us are familiar with the idea of fraud. We’ve all received, at some point, one of those scam emails inviting us to share a large sum of money in return for help transferring money overseas. You may also have had your credit card cloned. And if you’re really unlucky, someone may even have attempted to steal your identity, for example by opening a bank account in your name.

These are the typical scenarios that jump to mind for most people when they think of fraud.

But fraud is not limited to citizens and consumers – it’s also a huge problem for business.

Let’s take a typical example from the construction industry.

A buyer places an order for some beams on credit from a steel supplier.

The supplier delivers the beams to the buyer and issues an invoice due in 60 days.

At the end of the payment deadline, the buyer fails to pay: when the supplier tries to contact them, they discover the company doesn’t exist.

In the meantime, the fraudulent buyer has sold the steel beams on to another (legitimate) company, and disappeared with the cash.

This type of scam is all too common today, particularly in industries featuring goods with high unit values, such as construction materials or IT equipment. In fact, Euler Hermes found in a study in one country that this type of fraud accounts for more than 85% of business fraud cases detected.

At Euler Hermes, we have first-hand experience of dealing with this issue.

Our core business of trade credit insurance involves guaranteeing the payment of buyer invoices. Our clients - the suppliers – can be the target of buyer fraud. It’s therefore our job to protect them, and us, by identifying suspicious buyers before goods are shipped and the buyer has a chance to disappear without payment.

Over the years, we’ve developed reliable ways to do this by identifying red flags. One example is volume of requests. Fraudsters tend to be greedy. They don’t just target one supplier – they send requests to lots of buyers at the same time.

At Euler Hermes, we see a red flag when multiple customers request cover for a new buyer in a short period of time. We’re in a better place to detect the fraud than our customers because we see what’s happening across a market segment at any given time.

The issue is that manual detection of buyer fraud is tricky and time-consuming: it involves an analyst collecting and analysing the buyer’s administrative and financial documents and cross-checking it with external information.

Enter the Group Data Analytics and Artificial Intelligence team! Two years ago, our team launched a machine learning tool that leverages all the data we collect on our customers’ buyers to predict fraud.

The tool retrieves buyer data and carries out automatic analysis on financial statements: do they contain suspiciously round numbers? Is the margin exceptionally high? It also looks at history, for example whether the buyer has been investigated in the past six years. Finally, it looks at behaviour: has this same buyer recently flooded the market with purchase orders?

By analysing data and building on its ever-increasing understanding of market activity, our machine learning algorithm will clear the buyer, or flag it.

Can a tool replace the experienced judgement of our people? Certainly not. But it can help them make decisions more efficiently. We’re proud of the role we play here in Euler Hermes’ Group Data Analytics and Artificial Intelligence team in supporting the business to detect buyer fraud – and we finding ever more sophisticated ways to do it!