GermanyFinance and insurance2026 – ongoingUmweltBank
On-premise knowledge chatbot and overnight call transcription
UmweltBank put an on-premise AI chatbot on its internal policies live after 12 months of testing
Problem
Staff at the Nuremberg bank needed quick answers on standard procedures. The bank did not want customer or internal data to leave its own systems, as it can with public chatbots.
Approach
The bank built a retrieval chatbot on its internal process and policy documents. It used the Leadvise Reply agent framework and Mistral language models, and runs them on Nvidia A100 and H100 chips at a noris data center in Nuremberg. Hundreds of outdated or duplicate documents had to be revised first, and staff in each department now curate the data and check answers. A second stage reads customer call recordings from the Atruvia core banking system through RPA, then an AI agent transcribes and summarizes them overnight. Fraud detection and next-best-offer use cases are planned.
Results
No published figures
Takeaway
The bank said document quality, not the model connection, was the main hurdle. It gave no usage figures.
Categories
- Industry
- Finance and insurance
- Business function
- Internal work and knowledge managementCustomer service
- Innovation type
- Process innovation
- AI technology
- Generative AIAI agents
- Adoption
- Partner co-development
- Evidence
- Company's own claim
Sources
Source summaries
The trade magazine described UmweltBank's in-house AI setup. A knowledge chatbot on internal policy documents went live after 12 months of testing, and customer call recordings are transcribed and summarized overnight. The article names the framework, model, hosting and hardware, and gives no usage figures.
Full details are in the original sources.