Banks could unlock productivity gains of up to 20 times by deploying a workforce of AI agents, according to McKinsey.
The consultancy’s estimate rests on a simple shift: rather than completing every task themselves, a single compliance professional could supervise between 15 and 20 specialised agents. In some compliance teams, humans and AI are already working side by side, said Duna.
These agents are not general-purpose chatbots. Each is a specialised worker with a defined remit, from investigating a screening alert to monitoring a business for changes long after onboarding. Humans oversee the work and intervene where judgement is needed.
Adoption, however, remains at an early stage. Research from the Capgemini Research Institute shows only 10% of financial institutions have deployed AI agents at scale. Banks identify fraud detection (64%) and customer onboarding (59%) as the leading candidates for large-scale rollout.
The value of these agents, though, depends heavily on the roles they are given. Duna’s onboarding and compliance platform illustrates the model. A policy engine gathers evidence from business registries, submitted documents and sanctions lists, then identifies what still needs verifying.
Purpose-built agents pick up the remaining work: an onboarding agent reads KYC documents and runs sanctions, PEP and watchlist screening; a due diligence agent maps ultimate beneficial owner networks and drafts risk narratives; a screening and monitoring agent ranks alerts and drafts SAR narratives; and a perpetual KYC agent tracks registry changes and re-scores customer risk.
In one worked example, a German manufacturer applying for a business account has its registry data pulled automatically, its Luxembourg holding structure mapped, and a false-positive sanctions match on a director cleared with documented evidence, all before an analyst opens the case. Duna says its agents helped one European eCommerce platform cut onboarding from eight days to under a minute.
A European eCommerce platform CDD project lead said, “We have done more than 20,000 screenings on Duna at speed, because we can see everything in one overview, without pulling sources from multiple systems.”
Governance remains critical. Agents can drift, approving cases they previously escalated, so every recommendation must be traceable to underlying evidence and stand up to audit. Duna sets out five principles: define the policy first, break outcomes into specialised tasks, give agents the same information a human would need, make every decision explainable, and keep humans in the loop for complex cases.
Mews CEO Matthijs Welle said, “Duna’s AI agents reduce manual reviews, lower false positives, and increase straight-through processing. Exactly what our compliance team needs.”
For banks weighing where agents fit, the message is clear, in that agents deliver value as part of a policy-driven system, not as disconnected automations.
Stay ahead of the regulatory curve with strategic intelligence and early insight trusted by industry leaders. Subscribe to the RegTech Analyst newsletter today.
Copyright © 2026 RegTech Analyst
Copyright © 2018 RegTech Analyst
