How Generative AI Turned 1 Million Support Documents Into Instant Answers

One Million Documents. Zero Efficient Ways to Search Them.

Customer service reps at a global risk management insurance firm were losing the race against their own documentation. More than one million support files. Multiple systems. Every troubleshooting request meant combing through repositories while customers waited on the line.

Manual search wasn't a process problem. It was a customer experience problem.

Built to Find the Right Answer, Not Just an Answer

We deployed a GenAI chatbot powered by Azure OpenAI and Azure AI Search, starting with 25 device types and scaling to more than 10,000 devices indexed and searchable in production.

What makes it work:

• Natural language queries pull from both Salesforce repositories and the client's proprietary system, delivering device-specific answers in seconds
• Internal documentation is prioritized first, with automatic fallback to external sources like OEM sites and YouTube when needed
• A back-end pipeline built on Azure Data Factory and custom Python scripts keeps documentation current and searchable without manual intervention
• An orchestration layer filters responses by user need, with prompt engineering that enables memory-aware conversations
• CI/CD pipelines, Cosmos DB logging, and strict governance standards keep it compliant and auditable

Faster Answers. A Foundation for What's Next.

CSRs resolve device issues in minutes, not hours. Queue times drop. Every customer gets a consistent, accurate response — not an answer that depends on how well someone navigates internal systems.

The foundation is already in place to extend the chatbot directly to end users, deflecting routine queries before they reach a rep. What started as an internal efficiency play is becoming a self-service platform.

That's the difference between AI being built to understand the work, not just automate around it.