Last year, financial institutions in the German, Austrian, and Swiss (DACH) region prioritized the organizational design and institutionalization of their efforts in agentic artificial intelligence (AI).

This year, their focus shifted to engaging in building foundations, negotiating works council approvals for employee-related changes, and selecting high-return-on-investment (ROI) use cases, according to a recent study conducted by the University of St Gallen and French consulting firm Wavestone.

The study, conducted between October 2025 and May 2026, involved 30 semi-structured interviews with banks, insurers, and selected technology providers in the DACH region. The objective was to gain insights into the patterns, challenges, and developments shaping agentic AI in financial services.

The study explored various aspects, including organizations’ AI organization and operating models, current AI maturity, and their outlook on AI for the next two years. It revealed that despite enthusiasm for agentic AI, deployment remains limited.

In the DACH financial services sector, classical machine learning (ML) is mature and widely adopted among the sampled organizations, yet agentic AI is largely absent from production. Approximately two-thirds of the surveyed organizations had generative AI (genAI) in place but had no AI agents deployed in production. Only a few organizations had successfully implemented their first AI agents, and overall, industry participants anticipate working with agentic systems in earnest by 2027.

However, many firms expressed a shift from reactive market and technical developments toward proactive scenario planning around agentic AI. They are now focused on making AI work systemic, with orchestration as the source of value and multi-component systems replacing standalone tools. This underscores a transition from tactical adoption of isolated AI tools to building coordinated, integrated AI ecosystems.

Three main scenarios

The interviews yielded three potential future scenarios for agentic AI in the financial services industry.

In the first scenario, human supervisors oversee AI agents. These AI assistants perform tasks similar to robotic process automation, such as summarizing, drafting, routing, and answering queries from a knowledge base. This delivers productivity gains without significant structural changes, and is the current state of affairs for firms with AI agents in place.

In the second scenario, multiple agents collaborate end-to-end within a single value stream, such as claims intake, recourse, or underwriting support. This process is executed first in the back-office and mid-office, followed by customer-facing interactions. It is the predominant approach for most of the sample.

Finally, in the third scenario, agents and multi-agent systems are freely composed and reconfigured for each specific use case. These systems extend beyond the firm’s boundaries in two directions. The first direction involves collaboration across organizations, where the firm’s agents work with those of its partners and technology providers. The second direction involves interaction with agents on the open internet, where agents discover and coordinate with agents that the firm does not control.

AI efforts at DACH financial institutions

The study also examined the organizational structure of AI efforts. It found that most financial institutions adopt a hub-and-spoke structure where a central “hub” manages shared concerns such as strategy, governance, and infrastructure, while “spokes” embedded in individual business units handle actual AI delivery, since that’s where domain expertise and real use cases live.

This model takes four recurring shapes, primarily depending on the location of governance, budget, and works council involvement:

• In a hub-heavy model, the central hub holds governance, data, and infrastructure, while spokes execute the AI processes. This structure provides control and scalability but may reduce local ownership.
• In a spoke-heavy model, business units assume a significant role in the delivery of AI and may even participate in codetermination. This approach increases ownership but risks fragmentation.
• In the lab model involves a dedicated unit that manages its own budget and steering committee. Depending on the firm, works council involvement may be associated with the lab or remain with the spokes.
• Finally, the external setup involves placing innovation outside the regulated core, for instance, in a digital subsidiary, venture arm, or IT subsidiary.

The study revealed that a “60/40 central-to-decentral” model is the most prevalent. This approach involves around 30 to 40 units feeding into a single prioritized backlog ranked by business value, and is employed by 25% of respondents. The spoke-heavy model is also prominent, with 25% of respondents following an “AI community/champions” approach with a decentralized focus.

However, 33% of respondents are still forming their AI organization, underscoring persistent gaps.

Findings from the University of St Gallen and Wavestone study align with those from a separate research by KPMG, which found that agentic AI implementation remains limited among financial institutions but that plans to accelerate its adoption are in place. In Q2 2025, only 11% of institutions had put agents into production, although 99% planned to do so.

However, 65% of the companies polled were piloting AI agents, a number that doubled in three months, up from 37% in Q1 2025. This indicates a significant acceleration in AI deployment.

This year, financial institutions are prioritizing agentic AI. A 2025 survey by PwC of financial services executives found that for 30% of organizations, AI is a top investment priority in 2026.

Key applications of agentic AI include financial crime and regulatory processes, enabling banks to automate know-your-customer (KYC) and onboarding as agents compile, summarize, and validate source-of-wealth information that previously required extensive manual work.

In customer engagement and personalization, AI agents can resolve about 30% of routine requests without human intervention, bringing automation while improving customer satisfaction and reducing operating costs.

In lending and credit decisioning, AI agents can transform the traditionally manual and paperwork-intensive process into a dynamic, end-to-end digital journey. They automate document gathering, verification, and risk assessment, leading to faster, more accurate lending decisions and streamlined coordination among all parties involved.

Early adopters of agentic AI are already witnessing significant improvements. A separate May 2025 survey by PwC polled 300 senior executives and found that 66% of those adopting AI agents reported increased productivity, 57% reported cost savings, 55% faster decision-making, and 54% improved customer experience.

Featured image: Edited by Fintech News Switzerland, based on image by farknot via Magnific