Know-your-customer onboarding has been the slowest, least automatable part of opening a bank account for institutional clients, often taking months of document chasing and manual risk review. Swiss transaction bank Incore Bank, IT services firm Kyndryl, and Google Cloud say a proof of concept just showed that gap closing fast, with agentic AI compressing onboarding from months to days.
The system runs on Kyndryl’s Agentic AI Framework, layered with a semantic and Policy-as-Code capability, on top of Google’s Gemini models. Multiple AI agents work across structured and unstructured data, pulling from internal systems and trusted external sources to extract and validate customer information, gather evidence, identify risk factors, generate explainable risk scores, and produce auditable decision records. The proof of concept hit up to 99% accuracy automatically extracting data from onboarding paperwork, though that figure measures document reading, not the underlying credit or risk judgment, which still sits with a human compliance officer.
Why it matters: “in highly regulated industries like banking, success depends on strong governance, explainability, auditability, secure data access and human oversight,” said Jacqueline Wild, managing director and vice president of Kyndryl Alps. That framing matters more than the speed claim. Regulators have shown they treat bank AML and KYC output as load-bearing evidence, not paperwork, so an automation pitch that cannot produce an audit trail a regulator trusts is not actually deployable, whatever its accuracy score. This is also part of a broader shift toward identity and risk vendors building AI agents directly into compliance workflows rather than selling standalone scoring tools.
The original insight: Incore Bank chief executive Mark Dambacher described the goal as helping banks “assess customers more efficiently while maintaining rigorous oversight and controls,” which is a more conservative pitch than most agentic AI launches make. The banks most likely to adopt this kind of system first are not the ones chasing the fastest onboarding time, but the ones that can prove to a regulator that faster did not mean less controlled.
Source: Kyndryl
