AI Combined with Human Expertise: A Powerful Weapon Against Fraud

Tax, social security, credit card fraud… Fraud takes many forms, and with the diversification of payment methods, schemes are becoming increasingly widespread and sophisticated, impacting individuals, businesses, and governments with enormous financial costs.

In 2020 alone, payment fraud reached €1.28 billion* (including €644 million during the first quarter of 2021 alone), while estimates of tax fraud range between €80 and €100 billion**. Fighting fraud has therefore become a major challenge.

To tackle this growing threat, governments, businesses, and financial institutions are strengthening their fraud detection systems and increasingly turning to Artificial Intelligence. The French Ministry of the Economy and Finance has reported improved tax fraud detection performance through the use of datamining algorithms. AI is undoubtedly a key ally in the fight against fraud — provided it is used correctly.

Fraud prevention is a perfect use case for understanding the true potential of AI. At the heart of the system remains the business expert.

First step: thoroughly understanding the context for optimal data modeling

It is important to remember that, despite its significant cost, fraud remains a relatively rare event. Modeling rare events requires an extremely rigorous approach to data preparation and the ability to identify even the weakest signals.

As with any data modeling initiative, the first step is to create a relevant dataset while also gaining a deep understanding of the fraud context: What is the target of the fraud? How does the compromised system operate? Where are the vulnerabilities? Who can commit the fraud? How can fraudsters monetize their gains?

A detailed understanding of the company’s operating environment, supported by strong business expertise, makes it possible to identify targeted fraud typologies and design algorithms capable of meeting operational fraud detection requirements.

For example, in the context of credit card fraud detection, it is essential to understand that fraud reports are not immediate and that fraud patterns continuously evolve. These elements must be considered during the design and modeling phases. Otherwise, discrepancies may appear between the proof of concept (POC) and production phases, leading to results that fail to meet expectations.

Once the full context and constraints are clearly understood, how can organizations make the best use of AI’s capabilities?

AI provides multiple benefits for fraud prevention, especially when combined with human expertise. It opens new possibilities for analysis and performance improvement by enabling experts to:

• Optimize existing fraud detection systems
  Before introducing machine learning algorithms, many organizations already rely on fraud detection systems based on rules created by experts according to field observations and feedback. AI can challenge these rules and suggest optimizations (for example, adjusting thresholds) to improve relevance and reduce false positives, which often lead to financial losses such as wasted investigation time or legitimate financial transactions being incorrectly blocked.
• Identify new fraud patterns
  AI can detect emerging fraud schemes. However, it is critical that the outputs generated by algorithms remain interpretable for experts, enabling them to understand the newly identified fraud patterns. “Black-box” models (such as XGBoost), while effective for detection, are often less suitable for explaining and understanding fraud mechanisms. The deeper the understanding of the fraud pattern — enriched by human expertise and experience — the more effective and sustainable the response will be.
• Increase responsiveness and vigilance
  Fraudsters adapt quickly. As soon as they realize that a fraud scheme has been mitigated through new controls, they rapidly invent new ones. Organizations must therefore remain vigilant and agile. AI helps detect unusual behaviors or anomalies in real time (for example, the sudden appearance of a new e-commerce website or a sharp increase in activity on a platform) and flag them for investigation. However, anomalies do not necessarily indicate fraud — they may simply result from a product launch or promotional campaign — which is why expert validation remains essential.

In many ways, AI is a powerful lever for improving fraud detection. But like any form of intelligence, it must be “trained” on the fraud context it is expected to address in order to be truly effective and unlock its full potential. Combining the power of AI with business expertise is therefore essential to building a fraud prevention system that is comprehensive, agile, and highly effective.