Kyndryl and Google Cloud Pilot Agentic AI to Automate Bank KYC Onboarding
A proof of concept with Incore Bank demonstrates 99% data extraction accuracy in high-stakes regulatory workflows.
Kyndryl, Google Cloud, and Zurich-based Incore Bank have completed a proof of concept (PoC) using agentic AI to streamline the Know Your Customer (KYC) onboarding process. The trial proves that AI agents can accelerate digital onboarding while maintaining the strict auditability required for financial compliance.
The solution, built using Google's Gemini models and Kyndryl's Agentic AI Framework, focused on automating the extraction of data from customer onboarding documents and performing initial risk assessments. According to project data, the PoC achieved up to 99% accuracy in extracting information, significantly reducing the manual effort typically required to verify client identities.
The Compliance Burden
KYC is a mandatory regulatory requirement for financial institutions designed to prevent fraud and money laundering. Traditionally, this process is labor-intensive, relying on human analysts to manually review documents and assign risk scores. While standard large language models (LLMs) can summarize text, agentic AI is designed to execute multi-step workflows—such as verifying a document against a database and subsequently flagging a risk—making it more suitable for complex regulatory tasks.
Shifting the AI Paradigm
This pilot signals a critical shift in how the financial sector deploys artificial intelligence. While many institutions have limited AI use to internal productivity tools like chatbots, this project moves the technology into critical, customer-facing regulatory workflows. By demonstrating that agentic AI can handle high-stakes compliance with near-perfect accuracy, the project sets a precedent for other banks to automate onboarding to reduce operational costs and improve the client experience.
Governance and Oversight
Despite the level of automation, the project emphasized that AI does not replace human judgment in regulatory decisions. The framework was specifically designed to ensure that all risk decisions remain explainable, auditable, and subject to human governance. This "human-in-the-loop" approach ensures that the bank can justify its compliance decisions to regulators, mitigating the "black box" risk often associated with generative AI.
What's Next
The success of the Incore Bank PoC provides a blueprint for the broader integration of agentic AI in RegTech. Industry observers will now be watching to see if this model scales across different jurisdictions with varying regulatory requirements and whether other global financial institutions adopt similar frameworks to replace legacy manual onboarding systems.