At Rise of AI 2026 in Berlin, Julia Koch, managing director at Finanz Informatik, the central IT partner of the German Savings Bank Group (Sparkasse), explains how her organization is scaling sovereign AI from assistance tools to agentic systems across more than 350 independent legal entities and about 200,000 employees. Finanz Informatik runs its own data centers, built before ChatGPT existed, and operates an on-premise AI platform using open-source models where possible to avoid vendor lock-in, developed with engineering support from Nvidia. Its flagship tool, the SKI-Pilot (SAI-Pilot), is used by 200,000 employees, generated 2.8 million prompts in a recent ten-day period, and its meeting-preparation agent cut advisor prep time from 30-45 minutes to seconds, with over 175,000 meeting preparations completed in April. Koch describes deep integration with core banking systems and Outlook as key adoption drivers, and previews a widget-based, role-specific dashboard planned for deployment in June to support further agentic capabilities in a highly regulated environment.
Key takeaways
- Finanz Informatik built its own data centers over 10-20 years, letting it run AI in-house years before ChatGPT existed, which Julia Koch frames as the basis of its sovereign AI approach.
- The SKI-Pilot agentic assistant is used by 200,000 employees and generated 2.8 million prompts in a recent ten-day period.
- The meeting-preparation agent cut advisor prep time from 30-45 minutes to seconds and supported over 175,000 meeting preparations in April.
- Finanz Informatik avoids vendor lock-in by switching between open-source and other language models depending on the use case, developed with Nvidia's engineering support.
- A new widget-based, role-specific employee dashboard is planned for deployment in June as the foundation for broader agentic capabilities.
Detailed notes
Finanz Informatik's role and scale
Julia Koch introduces Finanz Informatik as the central IT partner behind the German Savings Bank Group, serving more than 15 million common-foundation customers, over 35 million online banking customers and more than 20 million users of the Sparkasse app, who together generate more than 22 million digital banking interactions per month, up from about 7 million per month when she joined in January 2022. She frames trust as the most valuable asset of banks and stresses the responsibility that comes with deploying AI across more than 350 independent legal entities, each subject to ECB and local regulatory requirements, comparing the scale to running '350 times Deutsche Bank.'
Why sovereignty rests on owning the data center
Koch says Finanz Informatik has operated its own data centers for 10 to 20 years and is one of the largest data center operators in Germany, which let it work with AI years before ChatGPT existed. This in-house infrastructure lets the platform connect directly to bank data, processes and institutional knowledge, including partners like Deka and LBS. The platform runs on open-source models where possible, switches between small and large language models depending on the use case, and avoids vendor lock-in. Koch credits Nvidia's engineering support for helping the team solve enterprise-scale problems specific to operating across roughly 400 legal entities, something neither side had faced before.
From assistance to agentic AI: the SKI-Pilot
Finanz Informatik's agentic tool, the SKI-Pilot, integrates directly into employees' Outlook environment, which Koch calls the adoption 'icebreaker.' It started with text generation and summarization and has since added agents that prepare emails, advisory meetings, requirement processes, financial statements and controls. In the ten days before the talk it logged 2.8 million prompts. Its meeting-preparation agent, given a customer number, pulls together customer data, transaction analysis and advisory suggestions in seconds, replacing a manual process that previously took advisors 30-45 minutes; in April alone it supported more than 175,000 meeting preparations. Koch attributes the tool's effectiveness to deep integration with the Sparkasse core banking systems, via a routing layer she refers to with the MCP icon shown on her architecture slide, connecting agents directly to bank processes.
Outlook, organizational change and closing principles
Koch previews a next-generation, widget-based dashboard tailored to each employee's role and regulatory needs, due for deployment in June, which she says is the foundation for deeper agentic capabilities across sales and service processes in a highly regulated environment. She frames the technology shift as equally an organizational one: business, design and technology experts must work in shared 'collaboration rooms' to find real use cases, many of which are discarded when they do not prove helpful. She notes Germany's retiring baby-boomer workforce makes AI-driven efficiency essential since replacement hires will not be available in sufficient numbers. She closes by summarizing sovereign AI at Finanz Informatik as strategic principles built on their own AI platform: customer-centric, compliant, effective and explainable to supervisors.
Quotes
Our own data centers are our prerequisite to talk about everything in a secured and flexible environment.
Julia Koch
Sovereign AI means for us really the strategic principles on our AI platform. That is our common and broad foundation for everything.
Julia Koch
Questions answered in this talk
How is Finanz Informatik scaling AI across the German Sparkasse group?
According to Julia Koch at Rise of AI 2026, Finanz Informatik runs a sovereign AI platform on its own data centers, serving more than 350 legal entities and 200,000 employees, and has moved from text-generation assistance to agentic tools embedded in core banking and Outlook workflows.
How much has the SKI-Pilot AI assistant reduced meeting preparation time for bank advisors?
According to Julia Koch at Rise of AI 2026, the SKI-Pilot's meeting-preparation agent reduced advisor preparation time from 30-45 minutes to a few seconds, and supported over 175,000 meeting preparations in April alone.
Why does Finanz Informatik run AI on its own data centers instead of the public cloud?
According to Julia Koch at Rise of AI 2026, owning the data centers lets Finanz Informatik connect AI directly to bank data and processes securely, avoid vendor lock-in by switching between models, and meet strict ECB and local regulatory requirements across hundreds of legal entities.