What we took away from RegTech Day

We spent last Thursday at RegTech Day in Paris. This was a dense event: the people there were the ones actually running compliance and payment operations. And a dense agenda. Our CEO, Maxence, kicked it off with a panel on “AI Act and data sovereignty: how should financial institutions adapt?” on the Operations Stage.

And the conversations that followed were as rich as the program. What emerged are three threads which ran through the day. And all three resonated with what we keep hearing from our European customers.

Regulation does not arrive, it accumulates

Every new text lands on top of the last one. The cost is rarely the rule itself. The cost is re-implementing it across backoffice tooling that was never designed to absorb change.

One of the clearest illustrations came from Sébastien Dalhiner, general secretary at Alma, on Consumer Credit Directive 2 (CCD2). The directive will require checking each customer’s solvency before credit is granted. He gave us a view of how Alma will adapt.

It is a new step in a live operational flow, on every file, with a trace to keep. And we believe that, as new regulation continues to land, tools like Forest will bring the agility needed to adapt.

Standardization of conformity, but not at the expense of UX

Thomas Courtois, president of Nickel, made the point of the day on Anti-Money Laundering Regulation (AMLR). While the regulation will help standardize processes across Europe, this standardization must not come at the cost of customer experience.

That is a classic and ongoing concern. Every control you add is friction you pay for in conversion. The market is not asking for fewer controls but for checks that fire on signal instead of on everyone.

Adaptive KYC

Not every customer journey deserves the same level of scrutiny, and the data should decide which situation does.

This is the deterministic / probabilistic split we keep coming back to now that fintechs have entered the AI agent era. Some steps have to be provable: identical every time, auditable line by line. Others are judgment calls, where a model reads context better than a static rule ever will. The mistake would be to run every process or step in one of these modes only.

The common thread

All three are operations problems as much as compliance ones.

What we heard in conferences or in discussion on the floor points to the same topic. Several large groups have stood up dedicated AI entities this year, with governance still being written, which means projects wait. Smaller teams went the other way and built their own agents for enhanced review, because nothing available gave them the control they needed. Both routes end in the same place: AI agents are only as useful as the permissions, the logs and the workflows around them.

That is the part we care about. The model is no longer the hard part. The hard part is letting humans and agents act on regulated data through the same controlled surface, with the same permissions and the same audit trail, whether the actor is a compliance officer or a script.

That is what we mean by agentic ops, and why we are building the fintech backoffice for the agentic era.

Thanks also to everyone who came by the stand: Xpollens, Marble, Iroko, Tessi, SBS, Cleardil, ComplyAdvantage, Alphaguard, Meelo, Regvantage.