Case Study: Zurich Insurance Builds ZuriChat – A Secure Internal Generative AI Assistant

As organizations build sophisticated AI Digital Workforces, internal chatbots are evolving from simple Q&A tools into critical entry points and orchestrators for intelligent, multi-agent systems.

They serve as the natural, conversational gateway that connects employees to powerful backend AI agents, automating processes while maintaining a familiar user experience.

Traditional internal chatbots were limited to scripted responses, FAQs, and basic ticket creation. Modern chatbots — powered by large language models and agentic capabilities — have become much more capable. They understand context, reason through requests, and act as intelligent routers within a larger digital workforce.

Instead of just answering questions, today’s internal chatbots can:

• Interpret complex, natural-language requests
• Break down tasks and delegate them to specialized AI agents
• Execute multi-step processes across systems
• Escalate to humans with rich context when needed

Zurichat

An exemplar case study of this is from the Zurich Insurance Group, a major global insurer serving over 55 million customers across more than 200 countries, developed ZuriChat as its enterprise-grade internal AI chatbot.

Launched globally in 2023, ZuriChat functions as a secure, company-controlled alternative to public tools like ChatGPT. It empowers employees with capabilities such as document summarization, translation, internal content search, data analysis, coding assistance, and technical support.

The project aligns with Zurich’s AI360 strategy to become an “AI-native insurer.” By running ZuriChat in a private, protected environment, the company mitigates data privacy and security risks inherent in public LLMs, while accelerating productivity and enabling responsible AI adoption across a highly regulated industry.

Background and Business Drivers

Insurance is data-intensive and heavily regulated, with strict requirements around customer data protection, explainability, and risk management. As generative AI gained traction in 2022–2023, Zurich recognized its potential for productivity gains but faced challenges with public tools:

• Risk of sensitive corporate or customer data leaking to external providers.
• Need for consistent, governed access to AI across a global workforce.
• Desire to move beyond experimentation to scalable, everyday productivity tools.

Zurich had already been investing in traditional AI (e.g., NLP for claims processing with partners like expert.ai) and exploring generative AI applications such as data extraction from claims documents. Experiments with ChatGPT in early 2023 highlighted both the promise and the governance gaps.

The solution was to build an internal chatbot that mirrors the user experience of consumer GenAI tools while operating entirely within Zurich’s secure infrastructure. This addressed the “AI paradox” — where many firms invest heavily in AI but see limited measurable impact — by focusing on safe, high-adoption productivity tools.

Development and Architecture

ZuriChat was developed in-house by Zurich’s AI teams, with the same group responsible for other proprietary tools like the Catastrophe Intelligent Agent (CATIA). Key design principles included:

• Security and Data Sovereignty: The chatbot runs in Zurich’s own controlled environment (private cloud or on-premises-equivalent setup), ensuring no sensitive data leaves the company’s perimeter. This was critical for a regulated insurer handling vast amounts of personal and financial information.
• Integration with Enterprise Systems: It supports tasks tied to internal knowledge bases, documents, and tools, likely incorporating retrieval-augmented generation (RAG) patterns for accurate, context-aware responses grounded in Zurich’s proprietary data.
• Broad Capabilities: Employees use it for summarization/translation of documents, searching internal content, basic data analysis, coding help, and technical support — freeing time for higher-value work.
• Governance Layer: Access is managed centrally, with monitoring, responsible AI guidelines, and human oversight to ensure compliance.

ZuriChat operates as part of a broader GenAI Lounge — a secure global gateway and central platform that provides access to multiple large language models from various cloud providers (including partnerships with Microsoft Azure OpenAI and AWS). The Lounge handles metering, dashboards, workload distribution, safety measures, and responsible use, making it the single entry point for GenAI experimentation and deployment across business units.

The architecture emphasizes modularity, scalability, and reuse, aligning with Zurich’s use of platforms like Amazon SageMaker for MLOps and productionizing AI assets globally.