ESPC26
Conference Sessions
Fixing Broken Project Governance with an AI Agent (Lessons from a Real Customer Implementation)
SPEAKERS
Chiara Carbone
microtom ag
ABOUT THE SESSION
Most organizations have project management standards and governance in place, but project managers don’t always stick to them in practice.
In one of our customer projects, this became a real issue. The customer was looking for a way to better support project managers in following defined processes without adding more manual effort or control overhead.
The Challenge:
Project managers work under pressure, processes are interpreted differently, and required steps or documentation are sometimes skipped. This leads to inconsistent project setups and limited transparency.
I will share how we approached this by building an AI agent using Microsoft Copilot Studio to support project governance in practice.
The agent guides project managers through required steps, checks inputs against defined rules, and brings together data from Microsoft 365 and external systems. Instead of just suggesting actions, it helps structure how projects are set up and executed.
A key aspect was keeping the human in control. Users can override decisions, but everything remains transparent and traceable. The agent also explains its recommendations, so users understand not only what to do, but why.
I will focus on what worked in practice, what didn’t, and the key technical and design decisions we made while implementing this in a real customer environment.
Assumed Knowledge:
You should have a general understanding of what AI agents and Microsoft 365. Basic understanding of project management concepts is recommended. And familiarity with Power Platform or Copilot Studio is helpful.
Practical Takeaways:
– A practical pattern for combining Microsoft 365 and third-party data to validate project inputs, including data mapping, integration, and decision support – How to design a Human-in-the-loop setup by defining which decisions require human approval, and implementing clear approval flows, explainable decisions, and proper logging for full traceability – Common pitfalls and how to avoid them when building and using such an agent
Out of Session Scope:
This session will not cover a full step-by-step implementation of Copilot Studio, detailed architecture or code-level deep dives. It will also not focus on broader project management methodologies or organizational change and adoption strategies.
MEET THE SPEAKERS
Chiara Carbone
microtom ag