A dashboard can tell you that revenue dropped 30%. But what happens when the next question is: why?
In this session, we will go beyond traditional BI and build an AI agent that investigates business problems using governed analytical tools rather than relying on the LLM to invent queries, calculations or explanations.
Using a real-world-inspired DecisionOps scenario, we will start with an anomaly surfaced in Power BI and follow the investigation through an AI agent built with Python, Microsoft Agent Framework and Microsoft Foundry. The agent will select analytical tools, inspect the underlying data, combine evidence from multiple business signals and explain what most likely caused the change.
We will also look behind the demo: how the analytical tools are designed, how they are tested, how agent tool calls can be traced and evaluated, and where human approval belongs in the process.
You will leave with a practical architecture for moving from dashboard → investigation → evidence → decision, and an understanding of how to build AI-powered analytics that are not only impressive in a demo, but also testable and governable.

Target audience: Power BI developers, BI engineers, analytics professionals and data architects interested in applying AI agents to analytical use cases.
Level: Intermediate. Basic familiarity with Power BI, analytics or data modelling is useful. No previous experience with Microsoft Foundry or AI agents is required.
Format: 60-minute technical session including a live demo. The demo uses fully synthetic business data and contains no customer or confidential information.
This is a new session developed around the DecisionOps AI reference project.