The challenge
The HR organization of a multinational energy group faced a high volume of repetitive employee questions on policies, processes and entitlements. Employees had no fast self-service route to accurate answers, and HR staff carried a continuous load. The group had already attempted the task using the platform's default configuration and had not reached usable answer quality.
At a glance
PlatformMicrosoft Copilot Studio, M365 / SharePoint
MethodCustom RAG: chunking, vector retrieval, prompt design
OutcomeClient's internal platform standard
What we did
SWICON designed a custom retrieval-augmented generation (RAG) architecture on the group's existing Microsoft stack (Copilot Studio, Microsoft 365 / SharePoint knowledge base). The differentiator was SWICON's own retrieval design on top of the platform: precise document chunking, vector retrieval, knowledge-base structuring, prompt and system-instruction design, and an answer-quality evaluation framework.
The result
A working HR chatbot pilot was delivered in two months. Based on SWICON's architecture and recommendations, the group's IT was able to scale and extend the solution independently, and the design became the foundation of the group's internal chatbots. The production chatbots built on that foundation achieve a correct answer rate above 96%.
