Fraud and real-time: a new lens on payment risk

Instant payments, increasingly sophisticated fraud schemes, and exploding data volumes are reshaping the fraud landscape at an unprecedented pace. In this context, financial institutions must fundamentally rethink their detection strategies. Arnaud Schwartz, Co-Founder of Marble, shares his perspective on the ongoing transformation, spanning real-time processing, friction, and artificial intelligence.

Arnaud, could you briefly introduce yourself and Marble?

Four years ago, together with my co-founder Pascal Delange, I launched Marble. Prior to that, we both worked at the neobank Shine, which we joined at an early stage of its journey. This gave us first-hand exposure to compliance challenges, particularly during the process of obtaining a payment institution license.

That experience revealed the limitations of the tools available on the market at the time: there was a clear gap between the actual needs of compliance teams and the solutions being offered.

This ultimately led us to create Marble, a transaction monitoring platform designed specifically for financial institutions. The platform covers the entire value chain, including scoring, screening, PEP and sanctions checks, and, most importantly, real-time transaction validation. It also provides comprehensive investigation and reporting capabilities.

Our strength lies in delivering an end-to-end platform enhanced by AI agents. Marble enables Compliance Officers to focus on high-value tasks – reviewing and assessing transactions – rather than spending time on configuration or continuous optimization. Because the platform was built specifically for financial institutions, clients can choose to deploy it on their own infrastructure to ensure data sovereignty or rely on our SaaS offering, which meets the industry’s strict regulatory and security requirements.

Real-time processing has become essential. In which cases does it truly make a difference?

In fraud prevention, the answer is quite straightforward: if an event is not detected in real time, it is already too late. At that point, it is no longer prevention, it is simply observation. The situation is somewhat different in Anti-Money Laundering (AML), where certain retrospective analysis approaches remain relevant. However, real-time capabilities help bridge the gap between fraud prevention and AML, generating significant operational efficiencies, particularly in how data is leveraged.

Real-time processing is also transforming analytical methodologies. Instead of relying on aggregated reviews of transactions after the fact, institutions can assess each transaction individually and immediately.

The primary challenge lies in latency. Decisions must be made within seconds while maintaining a very low margin for error. Achieving this requires highly performant, well-structured systems. Without them, real-time monitoring remains largely theoretical.

Can this immediacy create new forms of friction?

Yes, real-time processing can introduce new frictions, even as it removes others. When decisions must be made within seconds, institutions may need to temporarily block a transaction or request additional verification before authorizing a payment.

The key challenge is managing false positives. The more sensitive a detection system is, the greater the risk of interrupting legitimate transactions. On the contrary, controls that are too permissive leave more room for fraud. The objective is therefore to find the right balance between security and a seamless customer experience.

That being said, we must be realistic: preventing fraud requires action before funds leave an account. Without the ability to intervene in real time, institutions do not prevent fraud, they simply confirm it after it has already occurred.

How is the boundary between fraud prevention and anti-money laundering evolving?

There are strong synergies between fraud prevention and AML. At Marble, we believe every piece of data can provide value. Our platform can aggregate information from multiple sources – internal systems, accounting data, telephony monitoring tools, and more – consolidating everything into a single repository to improve the effectiveness of investigations and analysis.

From an operational perspective, the two disciplines remain separate, in line with the regulator’s requirements. However, the overlap between them continues to grow. Certain behaviors or transaction patterns can simultaneously indicate fraud and money laundering risks, particularly when identifying mule accounts or organized criminal networks.

As real-time detection capabilities become more widespread, the distinction between these two domains is becoming increasingly blurred, even though their regulatory objectives remain different.

Why does fraud continue to grow despite significant investments?

Because the primary fraud vector today is human.

We are witnessing the rise of social engineering fraud, where end users become the entry point for attacks against financial institutions – often to their own detriment. There is a fundamental information asymmetry between what the customer knows and what the bank knows. The purpose of fraud detection is to reduce that asymmetry and identify suspicious behaviors and anomalies.

In this context, the answer is not simply to invest in more specialized tools. Instead, institutions need to focus on anti-fraud design, rethink how banking interfaces are built and how users interact with them.

The goal is to help users recognize when they are about to perform a potentially risky transaction. This can be achieved through explicit warnings, risk-based customer journeys, or interfaces that are less automated and more contextual.

The introduction of Strong Customer Authentication (SCA) represented an important first step. However, it is no longer sufficient. When screens always look the same, users stop paying attention and tend to approve actions automatically.

Institutions must therefore introduce variation and contextualization, making customers active participants in their own security. The industry’s work around Verification of Payee (VOP), introduced last October, is another important step in that direction.

False positives remain a major challenge. How can institutions address them today?

This remains a critical issue. For many years, false positives generated significant operational costs by consuming resources, slowing down processes, and negatively impacting customer experience.

Today, artificial intelligence is fundamentally changing the equation. AI can analyze complex situations in seconds, 24/7, while filtering vast numbers of alerts.

We are witnessing a genuine paradigm shift. Whereas human teams previously had to investigate every case manually, AI can now perform a highly effective first-level assessment. This dramatically reduces the volume of false positives and allows human experts to focus on genuinely complex investigations.

Today, LLMs leveraging transaction histories without requiring institution-specific training are already delivering promising results. With the right level of tuning and data structuring, these models can achieve remarkably high levels of performance.

The second part of this interview will be published in the next few days. Stay tuned.