Incore Bank: AI Agents and Banking Governance Are Compatible

When a Swiss bank tests AI agents, executives take notice – and for good reason. In January 2025, Incore Bank, Kyndryl, and Google Cloud presented a proof of concept showing how agentic AI can reduce KYC onboarding from months to days – without sacrificing the governance essential to the financial industry. For you as a CEO or board member, this means: highly regulated industries can deploy AI agents in production when traceability, human-in-the-loop, and audit trails are properly implemented.

Executive Summary

Incore Bank, Kyndryl, and Google Cloud published a proof of concept for AI agents in KYC onboarding in January 2025. The result: KYC processes that previously took weeks to months can be reduced to days – with full traceability, human-in-the-loop control, and audit trails. The case demonstrates that agentic AI is production-ready even in highly regulated industries when governance frameworks are part of the architecture from the start.

Partners

Incore Bank, Kyndryl, Google Cloud

Use Case

KYC Onboarding (Know Your Customer)

Time Savings

From months to days

Governance Elements

Human-in-the-loop, audit trails, traceability

Technology

Google Gemini, Kyndryl Agentic AI Framework

Why the Incore Bank Case Is Relevant for Swiss Decision-Makers

AI agents – autonomous systems that independently plan, execute, and adapt complex tasks – still seem like future technology to many executives. The proof of concept from Incore Bank, Kyndryl, and Google Cloud proves otherwise: AI agents are production-ready, even in industries with the strictest compliance requirements. For you as a CEO or board member, three insights are central:

• **Regulated processes are not an exclusion criterion.** KYC onboarding is subject to anti-money laundering, data protection, and banking supervision – the case shows that AI agents can meet these requirements when governance frameworks are part of the architecture from the start.
• **Human-in-the-loop remains central.** The proof of concept relies on controlled autonomy: critical decisions – such as the final approval of a customer account – remain with humans. AI agents handle the repetitive, time-intensive work (document verification, sanctions list checks, risk analysis), humans make the strategic approvals.
• **Traceability is an architecture decision.** Every agent action is logged, every decision path is auditable. This is not an afterthought feature but part of the platform – a critical difference from black-box automation.

What the Proof of Concept Demonstrated Concretely

Incore Bank built an AI agent system for KYC onboarding with Kyndryl and Google Cloud. The architecture combines Google Gemini (multimodal language model) with the Kyndryl Agentic AI Framework, which provides governance functions such as audit trails, human-in-the-loop control, and traceability.

The process runs in three stages:

1. **Document capture and verification:** The AI agent reads uploaded identity documents, commercial register extracts, and other KYC documents, extracts relevant data (name, date of birth, company headquarters, beneficial owners), and matches them against internal requirements.
2. **Risk analysis and sanctions list screening:** The agent automatically checks whether individuals or companies appear on sanctions lists (OFAC, EU, Swiss SECO), evaluates countries of origin and business models according to risk categories, and creates a structured risk profile.
3. **Approval with human-in-the-loop:** The system presents the complete dossier to a compliance officer – including all verification steps, sources, and justifications. The human decides whether the account is opened. Every decision is logged.

According to Incore Bank and Kyndryl, processing time is reduced from several weeks (sometimes months) to a few days – without the bank lowering governance requirements or forgoing external audits. The proof of concept was made public in January 2025; production deployment is planned, though no specific timeline was provided (source: finews.ch, 2025).

Governance Architecture: Why This Is Not a Black-Box Approach

The crucial point is that governance is not added retroactively but is part of the platform. The Kyndryl Agentic AI Framework – an enterprise infrastructure for AI agents – provides three core functions:

• **Agent-level audit trails:** Every agent action – every document read, every API call, every assessment – is logged. This is not an optional feature but mandatory architecture. For auditors and supervisory authorities, it is always traceable how a decision was made.
• **Human-in-the-loop gates:** The system defines decision points where a human must intervene – for example, in high-risk situations or unclear data. The agent prepares, the human decides.
• **Policy engine:** Compliance rules (e.g., sanctions list checks, country-of-origin assessment) are centrally defined and versioned. When legal requirements change, the policy is updated – all agents adopt it automatically.

Boomi – a provider of AI agent governance platforms – describes such systems as an "Agent Control Plane": a central control layer that monitors agents, enforces policies, and provides audit data (source: Boomi, 2025). The approach is similar to how Kubernetes orchestrates containers – except here, AI agents are orchestrated.

What This Means for Swiss Companies

The Incore Bank case is no longer an isolated example. GE Appliances has been operating around 800 AI agents in production since 2026 – for quality control, maintenance planning, and supply chain optimization (source: PYMNTS, 2026). The difference from traditional automation: the agents autonomously adapt to new situations, learn from deviations, and escalate only when real exceptions occur.

For you as a CEO or board member, four strategic implications emerge:

1. **Governance is an architecture decision, not an afterthought.** Anyone who wants to introduce AI agents must think about traceability, human-in-the-loop, and audit trails from the start – otherwise the project will fail due to compliance or audits.
2. **Highly regulated industries are no longer an exclusion criterion.** Banks, insurance companies, healthcare – all industries that were previously skeptical of AI agents can now start production-ready deployments if the governance architecture is right.
3. **Start small, scale fast.** Incore Bank began with a proof of concept – a clearly defined use case. The pattern is transferable: choose a process with high manual effort, clear rules, and measurable results (e.g., contract approval, invoice verification, compliance screening). Test there, learn, scale.
4. **Avoid vendor lock-in.** Check whether the platform uses open standards (e.g., OpenTelemetry for logging, standardized APIs for agent orchestration). Proprietary frameworks lock you into a vendor long-term.

Next Steps for Decision-Makers

If you want to assess whether AI agents make sense for your company, proceed in three steps:

1. **Identify a process:** Choose a process that currently costs a lot of manual work, has clear rules, and delivers measurable results. KYC onboarding, contract approval, invoice verification, and compliance screening are typical candidates.
2. **Clarify governance requirements:** Define which decisions an agent may make and where a human must intervene. Clarify which audit trails your industry requires (FINMA, Swiss Data Protection Act, EU AI Act).
3. **Proof of concept with governance focus:** Build a prototype that not only automates the process but also implements traceability, human-in-the-loop, and audit trails. Test whether auditors and business units accept the system.

The Incore Bank case shows: AI agents are no longer future technology. They are production-ready – if you think about governance from the start.