Can AI be trusted to detect identity fraud in digital onboarding processes? This was the question raised during QuickSign’s latest webinar. A dense, interactive, and straight-talking exchange to demystify what artificial intelligence can—and cannot—do when facing new risks.
Between seamless customer experience and heightened security: the (im)possible balance?
Financial services are currently facing a dual requirement. On one side, users who demand a seamless, instantaneous experience available at any hour. On the other, a growing security imperative in the face of increasingly frequent identity fraud attempts. As Fabien Vittori points out, this tension puts institutions in a balancing act, because simplifying onboarding journeys sometimes means reducing control points.
This pressure is all the stronger given that identity documents were never designed to be verified remotely. Digital channels eliminate physical interactions, neobanks and fintechs set new standards, and regulatory constraints keep evolving. As a result, identity fraud remains a major point of friction in digital onboarding today.
Identity theft, forgery, mules: multiple frauds, sometimes undetectable
Contrary to popular belief, identity fraud is not limited to traditional identity theft. Admittedly, this method—which consists of using a third party’s documents without their consent—remains the most widespread. But other forms are emerging or becoming widespread: forged documents, synthetic identities, or the famous “mules”—people who open an account with their real identity before handing it over to a fraudster.
This confusion is all the more problematic because the boundary between cases is sometimes impossible to draw. For fraud detection teams, having a global vision capable of cross-referencing multiple weak signals becomes essential.
AI, an indispensable building block of secure digital onboarding
In this context, AI plays a decisive role. Not as a miracle solution, but as an accelerator of performance, precision, and scalability. At QuickSign, the numbers speak for themselves: in 2024, only 3% of fraud attempts managed to pass through all AI checks without triggering an alert.
Beyond raw performance, artificial intelligence allows high volumes to be processed without degrading control quality. It also adapts faster than fixed-rule systems, thanks to its continuous learning from fraudster behaviors. This makes it particularly relevant against shifting threats.
But the key element is complementarity with humans. Operators at the Fraud Expert Bureau (FEB) validate cases faster when they have been pre-approved by the machine—a sign of a genuine relationship of trust between the two.
Debunking misconceptions about AI and fraud
The webinar also served to debunk several misconceptions. No, AI is not useless: it triggers relevant alerts in the vast majority of cases. No, it is not a “black box”: every algorithm used in a QuickSign workflow is traceable, with the ability to explain decisions made at each step (OCR, biometrics, forgery detection…).
And no, AI is not limited to “gross” fraud. Thanks to advanced image analysis technologies, it detects signals invisible to the naked eye, such as colorimetry inconsistencies revealing a photo of a screen, or graphic signatures from fake document generators. This is precisely where AI surpasses human capabilities.
Humans remain indispensable — and the AI + human duo is the most effective
Why not automate everything? Because AI, however performant it may be, is not infallible. It can produce false positives. It can fail to decide in ambiguous situations. And it must be supervised so as not to create unnecessary friction for legitimate users.
At QuickSign, the model is not total automation, but smart automation. We automate everything we can without sacrificing workflow reliability. And we let humans step in where they bring real value.
What prospects for tomorrow?
Attacks will evolve. And defense systems will too. The next steps are already taking shape around so-called multimodal approaches: cross-referencing signals from documents, behaviors, text data, or open sources (OSINT). At QuickSign, this is notably what powers our Fudster Database project—a smart database built from real cases to strengthen our predictive models and anticipate new fraud strategies.
Conclusion: without AI, no effective fight against identity fraud
Artificial intelligence alone is not enough. But without AI, securing a digital onboarding process at scale has become virtually impossible.
It is by combining the power of algorithms with human judgment that financial services will be able to meet the number one challenge of digitalization: securing onboarding journeys… without sacrificing the user experience.
Written by Nicolas G.
