Answers locked behind reports and DAX
Staff who needed a number had to find the right report, filter it correctly or ask an analyst. The group wanted people to simply ask a question and get an accurate answer from the same trusted data, without opening up raw data or building a separate chatbot.
Put an AI agent on top of a semantic model built for it
We built a Power BI semantic model with clear measures and business descriptions, then connected an AI agent through Microsoft Copilot and Fabric. The agent reads the question, writes the DAX, runs it against the model and returns a table or visual, all inside the group's own Microsoft tenant.
A model written for AI
Measures, hierarchies and plain-language descriptions tell the agent what each number means, so answers use the business's own definitions.
Tested prompt templates
Structured prompts translate everyday questions into DAX consistently, so the same question gets the same answer.
Secure by default
The agent connects through Entra ID and stays inside the group's tenant, so it only sees what users are allowed to see.
Checked for accuracy
Controlled test cases with real business questions validated every answer against the model before the demonstration.
Tuned for reliability
Prompt parameters, filters and DAX translation were refined until answers matched the business's expectations.
Ready to roll out
The handover covers how to extend the agent to Microsoft Teams or a web interface for the wider organisation.
Where the data lives and how it flows
Hosts the workspace and the data the semantic model is built on
What changed
Staff can now ask a question in plain English and get an answer from the same trusted model the reports use, instead of hunting through dashboards or waiting for an analyst.
Because the agent sits on the semantic model and Entra ID, answers follow the business's definitions and its security, and the same pattern extends to Teams when the group is ready.