AI can clear most of a KYC check on its own. It's the last 20% that still needs a person. As part of our CTO vs Status Quo and Fintech Recoded series of interviews, we’ve spoken to Dennis Overbeeke who had a lot to say about this subject.
Fintechs run some of the heaviest identity and fraud checks in software, and AI agents are already reshaping the fraud detection use cases that matter most this year.
TL;DR
Full KYC automation still isn't realistic,
Fintechs keep their own decision engine,
Humans stay in charge of edge cases,
Accountability can't be handed to a vendor.
Why do KYC and AML resist full automation?
More data now feeds every KYC and AML decision, and AML automation keeps expanding to keep pace. Regulators keep adding checks on top of that.
For Dennis Overbeeke, CTO of New10, AI fraud detection is the obvious next step for connecting it all. It brings its own risks too. Those risks include hallucinations, weak auditability, and transaction chains that stretch across dozens of connected entities.
A 2026 industry poll named AI the single most impactful factor shaping the fraud landscape this year. It ranked ahead of criminal sophistication and regulatory change.
Can KYC be outsourced to a third party entirely?
Third-party vendors already offer solid automated fraud detection and KYC tooling. Full outsourcing still doesn't work in practice, according to Overbeeke.
"[...] it's you who's going to be held accountable. You can't outsource accountability. That's why fintechs or banks prefer to keep that in-house." — Dennis Overbeeke, CTO at New10
Every fintech using an automation vendor still wants its own business rules sitting on top. Risk appetite is company-specific, not vendor-specific.
TSH ran into a similar constraint helping a Dutch bank move onto cloud infrastructure.
The new setup had to satisfy strict EU banking rules from day one, not bolt compliance after launch. Every vendor integration sat inside a compliance layer the bank controlled itself.
What is the 80-20 rule, and why does the other 20% matter?
Overbeeke puts a number on where automation tops out.
"In practice, the need for human intervention here boils down to the 80-20 rule. 80% of KYC could be automated, but there are always edge cases that need to be double-checked or re-interpreted by a human." — Dennis Overbeeke, CTO at New10
An unusual ownership structure. A transaction chain that doesn't fit a template. That's where the remaining 20% lives.
The same split was held at Blanco, Overbeeke's previous company. It built a digital onboarding tool, but every customer profiled risk differently. The tooling had to stay customizable, not one-size-fits-all.
Where is AI cutting the most friction in fraud detection today?
AI agents show up first in real-time transaction monitoring. They flag anomalies as they happen, not days later in a batch review.
The bigger win for AI fraud detection often isn’t about catching more fraud. Cutting the false positives that used to eat up analyst time can be an even more important job for AI.
That matches what's showing up in the previously mentioned industry research. Data completeness now ranks as the top concern for compliance teams, ahead of tooling limits or the cost of compliance itself.
It tracks with Overbeeke's point. Connecting fragmented data is the harder problem underneath the AI layer.
What would make cross-border KYC easier?
Right now, switching banks means redoing KYC and AML from scratch, even within the EU.
In response to this, Overbeek doesn’t recommend more automation. He believes that standardization is the way to. That means industry-accepted digital identities and uniform KYC data that travels with the customer.
"Right now, there is no way for a customer to say, 'Bank A already verified me, could you use their data to verify me?'" — Dennis Overbeeke, CTO at New10
Until that exists, AI agents can only optimize each institution's own process. They can't remove duplication that's built into the system itself.
How much control should a fintech hand to AI agents?
AI agents absorb volume. Humans keep the judgment calls. The business keeps its own rules on top of whatever vendor tooling it buys.
The goal here is not full automation, but a defensible, audit-ready decision.
Read the original conversations with Dennis Overbeeke for more details:
The interview with Dennis Overbeeke , CTO at New10.
Authors
Adrian Senecki
Copywriter and budding fiction writer, interested in (but not limited to) the business side of software development. Likes acquiring new skills and foretelling the future.
Andrzej Wysoczański
Frontend developer with 10 years of experience. With The Software House for almost 7 years, going from a regular dev to the Head of Frontend. He loves keeping tabs on the latest frontend technologies, especially React-related. Regular of the Taby & Spacje podcast (tsh.io/taby-vs-spacje) for Polish speaking programmers.