Building agentic tools on Microsoft Copilot -RUBIS
Business Case
Computer Revolution Africa Group (CRAG) has designed and delivered an AI-powered HR Policy Agent that is:
- Built on Microsoft Copilot Studio,
- Grounded on the organization’s HR policy documents
- Stores data in a designated SharePoint document library.
The agent will provide employees and the HR team with instant, mostly accurate, self-service answers to routine policy questions through a familiar conversational chat experience in Microsoft Teams or the web..
Key Stakeholders
The HR Policy Agent will serve the following stakeholder groups, each with tailored needs:
Employee/End Users
Prevented the unauthorized leakage of confidential and sensitive data
HR Team
Reduce repetitive enquiries, ensure consistent guidance, and direct complex cases to the right contact.
HR Content Owner
Maintain the approved policy library in SharePoint as the single source of truth and validate agent responses.
IT / Administrators
Govern access, identity, and security in line with existing Microsoft 365 and SharePoint controls.
Solution Overview
The proposed solution is a conversational HR Policy Agent built in Microsoft Copilot Studio that uses a designated SharePoint document library as its single grounding source
Using Retrieval Augmented Generation (RAG), the agent retrieves relevant passages from approved policy documents and composes accurate, natural-language answers, each accompanied by a
citation/link back to the source policy.
Measurable Business Outcomes
The agent allows for:
Securely retrieving answers from approved policy documents
Returning instant, cited responses,
Improving HR productivity,
Enhancing the overall employee experience.
Ensuring consistency
Conclusion
The solution operates entirely within the Microsoft 365 trust boundary. It honours SharePoint permissions, relies on Entra ID for authentication, and keeps data in-tenant. For the POC, the library is scoped to general-staff policies, and responses are strictly grounded to the approved document set to prevent ungrounded or out-of-scope answers.
