We’ve been here before – so easy to build, so hard on the CIO’s budget. The path to cost-effective AI in the Microsoft stack is strewn with pitfalls from autonomous agents burning Copilot Credits, makers and developers choosing top-end AI models, Computer Use Agent step pricing, unexpected Cowork consumption, and that’s before we get to simple M365 Copilot license wastage. The solutions might be impressive, the bill less so.

Agent workloads are making AI FinOps and cost governance one of the most important conversations in AI. This session cuts through the complexity we’ll walk through the major gotchas, governance levers, and cost avoidance strategies. We’ll also look at hard numbers from real-world deployments.

You’ll leave with a better view of common trouble areas, tips for building affordable agents, methods for forecasting and optimising agent expenses, recommended governance practices, and some ideas of how to tie costs to C-level value when they just cannot be reduced!

Assumed Knowledge:

Some level of understanding of Microsoft’s AI ecosystem – M365 Copilot, Copilot Studio, Foundry etc.

Practical Takeaways:

A framework for M365 Copilot/Copilot Studio/agents/Foundry cost optimisation, and real-world experience in calculating agent costs.

Out of Session Scope:

M365 licensing strategy (e.g. E7). Azure cost management more broadly. Agent building techniques and governance