Jens Verboven: "The fraud indicator is the Holy Grail for an industry"

"Artificial intelligence and machine learning have significantly improved the first fraud model since 2020. Self-learning algorithms have made the fraud indicator more accurate and robust. Thanks to customer feedback and broader integration of external data, current models better detect fraudulent activities. Without fine-tuning the basic model, we already achieve a prediction rate of 75% today. This standard model forms the starting point for further refinement."

Increasingly Accurate Score

To tailor the model to the client's needs and further improve the prediction rate, GraydonCreditsafe delves into the specific fraud cases the client has experienced. Ideally, these cases go back up to five years.

"We can see the exact situation at that time. Then we apply our basic model to it, which not only shows if the model works, but also provides insight into the prediction rate and how much improvement is possible," continues Jens Verboven.

"Next, we reactivate the technology to see if any other elements emerge. This allows us to continue developing the basic model. If you can add a few exceptions to the internal data, you build a very strong and efficient model. And that's not all. As soon as customers use the score intensively, internal data is continually fed back to improve the score."

Fraud Indicator Between 0 and 10

The fraud indicator ranges between 0 and 10. To some extent, the score is open to interpretation. The customer chooses which threshold to apply and should ideally divide it into three categories: high, medium, and low risk of fraud.

"The customer has a choice between a lot or a little manual work, depending on the threshold. They decide what 'high' and 'medium' mean. Reject the file immediately or examine it manually more thoroughly? You obviously want the segment requiring further research to be as small as possible. Ultimately, the customer decides whether to engage with a company or opt for an alternative approach, such as asking for a guarantee. In any case, all parties benefit from the highest possible prediction rate."

Customers receive the fraud indicator via an API or an online application where they can enter a business registration number individually. In addition to the result, the user also receives an overview of suspicious elements, making it easier to dig deeper during manual handling.

The Added Value of a Fraud Indicator

The prediction model automatically indicates which companies have a higher chance of fraud. It is much faster than manual checks. Additionally, the algorithms consider more factors.

"Thanks to the model, employees have more time to thoroughly investigate questionable cases. If they used to have to manually review ten cases, they can now exclude seven parties thanks to the model. This gives them much more leeway to closely examine the remaining three cases. In practice, they can prevent more fraud cases."

A crucial note is that the indicator does not accuse any company of fraud.

"No, the indicator simply indicates whether the risk of fraud is real and which elements make a party suspicious. A user can work on two axes: on one side, the prediction rate, and on the other, the efficiency of manual checks. The more accurate the score, the more efficient the control process can be. Therefore, it's more beneficial to combine internal and external data so that the prediction rate—and efficiency—can continually improve."

The Holy Grail

Companies need partnerships more than ever to improve data sharing. Industry organisations and federations play a crucial role as neutral parties.

"At the sector level, a fraud indicator is actually the Holy Grail," Jens Verboven concludes. "Fraud is a persistent societal problem, and I'm convinced that legitimate companies want to share information that will benefit everyone. Fraudulent companies need to be removed. Full stop. Ideally, we collaborate with a federation that acts as a neutral party and provides information to feed the model. This is necessary to refine the model and improve detection."

"Does a case still slip through the net? We want to be part of the feedback loop. Customer feedback on fraud cases is crucial for further refining the score. Each new case is carefully analysed and used to improve the model, making future cases easier to detect. Who could object to that? It's ultimately in everyone's interest."

To Conclude

In 2024, the fraud indicator will continue to be an important tool for companies to identify potential fraudsters. The ongoing development of AI models, cross-sector collaboration, and a strong feedback loop contribute to the accuracy of this solution. For companies, it's crucial to continue investing in advanced prevention models to ensure a secure business environment.

Do you want to test a specific case against the model? Or would you like to know what the fraud indicator can mean for your company or sector? Contact us immediately for a live demo of the platform.