SAP builds a Mistral-powered multilingual chatbot that cuts SBB's software-migration support queries by 80%

Overview

To support end users during Swiss Federal Railways' (SBB CFF FFS) multi-year software migration affecting 30,000 employees, SAP built a multilingual chatbot using Mistral Large hosted entirely on SAP-operated infrastructure, making the deployment 100% European and locally hosted. The chatbot uses retrieval-augmented generation to draw on FAQs, training materials and unstructured content, with glossary-based query rewriting for domain-specific abbreviations and a human-in-the-loop process that captures expert answers into the knowledge base. It integrates into existing workflows via SAP's Build Work Zone portal and Microsoft Teams. The implementation achieved an estimated 80% reduction in repetitive support queries and scaled from 1,000 pilot users to SBB's full 30,000 employees, supporting German, French and Italian natively.

The challenge

Modern enterprises face universal challenges that span every sector: repetitive support queries that overwhelm expert teams, knowledge silos that prevent efficient information sharing, and manual processes that consume valuable human resources. These pain points become particularly acute during major transformations and migrations—whether it's SBB's five-year software migration affecting 30,000 employees. The complexity multiplies in European markets when there are multilingual situations. Organizations operating in German, French, and Italian face additional barriers when AI solutions can't effectively handle regional languages and dialects.

The solution

To help address end-users support needs during SBB's multi-year software migration, the SAP team built a multilingual chatbot leveraging Mistral Large hosted entirely on the SAP operated infrastructure. The chatbot employs sophisticated retrieval-augmented generation (RAG) to draw from multiple knowledge sources including FAQs, training materials, and unstructured content like recordings and screenshots, using advanced prompt engineering and glossary-based query rewriting to handle domain-specific abbreviations. A critical innovation is the human-in-the-loop functionality: when the AI cannot resolve a query, expert responses are automatically captured and added to the knowledge base. The solution integrates directly into employees' existing workflows through SAP's Build Work Zone portal and Microsoft Teams.

Reported business value

SBB achieved an estimated 80% reduction in repetitive support queries, freeing expert "super users" to focus on complex problem-solving. Scalable coverage: From 1,000 pilot users expanding to 30,000 employees, demonstrating enterprise-wide scalability.