Fraud and scams are rising across Europe, and the view of Frédéric Lebeau, Founder of Datavillage.ai, is that the type of fraud growing fastest matters as much as the volume. He highlighted the rise of scams such as investment scams, impersonation scams or romance scams: attacks that tend to land hard when successful.

His point was simple: fraudsters are spending more time “setting up” each victim because the payoff is higher. He referenced an average loss of around €3,000 per scam (all types combined) and €73,000 per investment scam (yes 73K !). Large amounts make these scams particularly painful for customers .

Why scams are winning: the weakest link has changed

Frédéric’s core explanation for why scams are getting worse is that banks and regulators have genuinely improved protection of the banking system itself. He cited measures like transaction monitoring, stronger onboarding checks, and strong customer authentication as examples of progress that have made classic “break-in” fraud harder.

But fraudsters have adapted. Instead of trying to bypass bank controls directly, they are targeting what he called “the weakest link”: the customer. In practice, that means social engineering and convincing a legitimate customer to do something that benefits the criminal. The bank sees a “real” customer logging in and making a “real” payment, but that action is being driven by manipulation.

This reframing is important because it changes what “good fraud prevention” looks like. If the attack is mostly psychological and happens outside the bank’s channels, then a bank can’t rely only on single rules, static thresholds or even segment-based deviation analysis .

Instant payments raise the stakes: prevention has to happen earlier

Instant payments featured heavily in the discussion, not because they create scams, but because they make scams more efficient. Frédéric’s point wasn’t that criminals have invented new tricks, it’s that they exploit every new capability the industry rolls out.

When money moves in seconds, the window for intervention collapses.

He explained the practical consequence: if funds can move across accounts quickly, the best defence is to catch a suspicious payment before it is executed, not after. Once the payment is gone, recovery becomes dramatically harder, and the bank is left dealing with downstream steps in the lifecycle: investigation, customer support and reporting, instead of stopping harm in the first place.

He also hinted at what’s next in the fraudster’s toolkit: as new models like “agent commerce” emerge, the industry should assume criminals will find ways to exploit them too.

The hardest part is context: no single “magic signal”

When the conversation turned to “early warning signs”, Frédéric pushed back on the idea that there is one reliable indicator. For scams, he argued, detection is less about a single red flag and more about a scenario : a set of signals that only makes sense when you see the wider context.

He also made a key operational point: many of the most useful signals happen outside the bank’s visibility. Scams often begin through phone calls, SMS, social networks, or messaging apps. Banks may only see the final step, the payment, but not the trail of manipulation that led to it.

His argument is that if fraudsters have shifted to attacking users, then prevention systems need to shift too: focus on understanding what the user is doing and why, rather than only looking for abnormalities in isolated transactions.

Regulation is moving, but the big lever is liability

On the regulatory outlook, Frédéric’s view was that change is coming, but not fast enough to match the speed of criminals. He pointed to PSD3 and the Payment Services Regulation (PSR) work as the most interesting developments, particularly because of their focus on customer protection and liability.

His logic here was that liability is one of the few levers capable of moving an entire ecosystem in the same direction. If different parties share responsibility for losses, they also share incentives to improve prevention and collaboration.

He used the UK as an example, referencing the mandatory reimbursement approach that came into effect in October 2024 and the “50/50” concept where responsibility is shared between sending and receiving institutions. He noted that reimbursement rates improved (he cited 88% reimbursed after a year versus 65% before – you can find more interesting insights in their summary here), while fraud levels still increased, suggesting that reimbursement helps victims, but on its own does not stop criminals.

He also mentioned that liability models differ across markets, which underlines why cross-border consistency is hard, but also why Europe may eventually push towards clearer shared obligations.

What banks should prioritise while timelines are still shifting

Asked where banks should focus while regulation is still being finalised, Frédéric’s advice was not to wait for perfect clarity. He expects regulation and criminal behaviour to keep evolving, and he argued banks need a clear view of the full fraud journey, even if they can only improve one step at a time.

He also made a comparison that will resonate internally: banks have invested heavily in AML over many years, but in his view, fraud has lagged. The opportunity now is twofold:

• better prevention for customers, especially against scams

• better operational efficiency, because the cost of handling cases is huge

His recommendation was essentially to build a strategic target across the end-to-end lifecycle, then improve step by step, but without losing sight of how the whole machine should work together.

The real bottleneck isn’t a missing tool — it’s too many tools

One of the most practical insights from our discussion came from his description of daily life inside fraud operations. He walked through the lifecycle: prevention, detection, triage, investigation, recovery, and reporting and then said the main issue often isn’t that teams lack tools, but that they have too many, and they don’t connect smoothly.

He gave an example from time spent observing a UK bank’s operations, where an analyst might need to use eight different tools in a day, moving between them, copying and pasting data, and manually building a picture of what happened. That’s not just slow, it also reduces the time analysts can spend on the actual thinking required to make good decisions.

Combine that with high false positives, and teams get flooded. The result is predictable: attention goes to volume handling, not to the most important cases or to learning lessons that would strengthen prevention upstream.

“Shift left”: move learning earlier in the lifecycle

Frédéric borrowed a concept from software engineering that maps neatly onto fraud: “shift left”, meaning move work that happens late in the process to earlier stages, where it is cheaper and more effective.

In fraud terms, that means taking what investigators learn during triage and investigation and feeding it back into prevention. But he was clear that this requires automation, because teams won’t have the capacity to do it consistently while they are overwhelmed.

This is where his view of AI becomes more grounded than the usual hype.

Where AI agents help today, without pretending they replace humans

Frédéric was pragmatic about AI agents. He sees them as useful across the lifecycle, but he pointed to three areas where the return is especially fast and tangible:

• alert triage

• case investigation

• compliance reporting

The benefit, in his framing, is not “AI makes the decision”, but “AI does the heavy lifting”: gathering information, summarising evidence, building a coherent narrative, and keeping an audit trail, so a human analyst can decide faster and with better visibility.

He also challenged the fear that AI inevitably replaces people. His argument was that fraud analysts should not be afraid of working hand in hand with these tools, because they can take away the least rewarding parts of the job and give analysts time back for higher-value work, including creating the feedback loop that improves prevention.

Crucially, he emphasised that banks still want a human in control. In his words, there’s still a “final button to click” that needs to be clicked by a person. That design choice is as much about trust and accountability as it is about accuracy.

A collaboration model built around sharing signals, not raw data

A repeated theme throughout the discussion and within the industry was the need for broader context and external signals, paired with the reality that sharing data is difficult. Frédéric explained that Datavillage’s approach centres on enriching transactions with both internal and external data, and leveraging AI agents to connect disparate activities into coherent scenarios that mirror how fraud victims are approached and exploited.

The takeaway: protect customers by modernising the operating model

To summarise the key messages, it’s that scams have forced a shift in mindset.

• The battleground is moving outside the bank, so banks need context, not just transaction rules.

• Faster payments shorten the response window, so prevention must happen before execution.

• Regulation is pushing harder on customer outcomes and liability, but it won’t solve the problem alone.

• Operations teams are overloaded, not because they lack tools, but because the workflow is fragmented and noisy.

• AI agents can help immediately in triage, investigation and reporting, not by replacing humans, but by giving them time back and making decisions better supported.

His closing advice to fraud teams was practical: embrace technology, use it to remove the drudge work, and reinvest that time into preventing harm earlier in the chain.

That’s how banks reduce both customer losses and the operational burden that follows every successful scam.

(You can find Frédéric and the Datavillage team at our flagship conference in Brussels on May 28 where we will be digging deeper into top-of-mind topics for fraud-fighters on our dedicated Fraud and Compliance stage - secure your seat for #TBSCONF26BXL today and join the conversations.)

Summaries are great for a quick skim-read, but if you want the complete, in-depth insights, you can find the full interview below or follow along on your favourite podcast channel here (and don't forget to subscribe!).