AI is moving faster than most budgets can keep up with. In this episode of Demystifying Microsoft, host Nathan Taylor sits down with Graham Rosenberg, Director of Intelligence and Automation at Sourcepass, to unpack one of the most urgent conversations in the Microsoft ecosystem today: AI FinOps for Microsoft Copilot.

What You’ll Learn:

Why Copilot, Copilot Studio agents, and Copilot Cowork require a new kind of financial discipline
The two control planes every IT and finance leader needs to understand
Hard-won lessons on setting spending policies, building environments, avoiding runaway agent loops, and identifying power users who may justify a full Microsoft 365 Copilot license
Guidance on Azure AI Foundry, Copilot Studio, and cost-effective model selection
How to build enforcement caps so no one wakes up to a surprise bill

About the Guest:

Graham Rosenberg is the Director of Intelligence and Automation at SourcePass, where he leads the RPA automation, business intelligence, and AI teams. With deep expertise across the Microsoft stack, Graham helps organizations build custom AI agents, workflow automations, and BI dashboards that drive real operational efficiency. He splits his focus between scaling Sourcepass's internal operations and applying that same playbook to help customers navigate the fast-moving world of agentic AI, Microsoft Copilot, and AI FinOps.

About the Host:

Nathan Taylor is the Senior Vice President and Global Microsoft Practice Leader at the Sourcepass Center of Excellence for Microsoft. With more than 20 years of experience supporting small and mid-sized organizations, Nathan brings a pragmatic perspective shaped by real-world deployments, Microsoft 365 security strategy, and close collaboration with Microsoft. He helps IT leaders cut through the noise of an ever-evolving Microsoft ecosystem so they can make informed, defensible decisions with confidence.

Episode Highlights:

03:26 Why Copilot Needs a FinOps Discipline

AI has shifted Microsoft licensing from predictable per-seat pricing to consumption-based billing. Every agent, prompt, and long-running task now runs a meter.

05:48 Where Teams Get Their AI Billing Model Wrong

Copilot credits flow through two separate control planes: one for Cowork, one for Copilot Studio agents. Confusing them is the #1 source of billing surprises.

09:02 The End of Free Cowork

As of July 1st, Cowork requires both a spending policy and an agent assignment to work. Miss either step, and your users are locked out.

15:00 How Copilot Studio Agents Factor Into Your AI Budget

Paid Microsoft 365 Copilot license holders get near-unlimited agent usage. Everyone else needs metered billing policies tied to dedicated Power Platform environments.

18:18 Steps to Avoid a Surprise Copilot Bill

Enable auto-shutoff at 100%, build per-agent Azure resource groups, and set threshold alerts at 50/70/90/100%. Enforcement caps beat notifications every time.

20:57 What IT and Finance Leaders Should Do This Quarter

Pick your control plane, set hard caps, and treat AI spend as a collaborative business decision. Start with capped pilots.

22:47 Agent 365, Observability, and the Future of Microsoft AI FinOps

Microsoft is rapidly closing FinOps gaps in the Copilot admin panel. Expect deeper observability and governance tooling over the next six to twelve months.

25:00 When to Move Beyond Copilot Studio Into Azure AI Foundry

Copilot Studio shines for chat-based agents. For agentic workflows without a human in the loop, Azure Functions and AI Foundry offer more flexibility at lower cost.

27:10 Cost-Effective Model Selection

Well-instructed agents can run on cheaper models like GPT-4o mini at $0.29 per million output tokens, a fraction of frontier model pricing.

28:52 Why Copilot Still Wins on Access to Frontier Models

Copilot delivers the latest OpenAI and Anthropic models inside Microsoft Enterprise Data Protection, giving you frontier capability with enterprise-grade security and governance.

Episode Resources:

Not sure where your Copilot credits are going, or whether your spending caps will actually hold? Connect with the Sourcepass MCOE team to build a Microsoft Copilot FinOps strategy with the right licenses, control planes, and enforcement caps, so AI spend stays predictable: https://sourcepassmcoe.com/demystifying-microsoft-contact

Graham Rosenberg on LinkedIn
Sourcepass Website
Nathan Taylor on LinkedIn

Quotes:

“Historically, small businesses have survived on cloud services with per-month licensing. Very repeatable, you know what your cost is going to be. AI and everything with it is very much like Azure consumption. The more you use it, the more you're going to pay.”
“You can look at just Copilot credits as one currency but two control planes.”
“I like to tell people when you're setting up these limits, think of it very similar to when you go gambling in Vegas. Only bring the money you're willing to lose.”
“It's really a muscle we all have to work at and get good at. There's really no right or wrong answer to how to use it. You have to figure out how to use it.”
“There is a trap here where the shortcut answer is just give them a paid license and a lot of these problems go away. And if you can afford that, by all means, it simplifies everything. But it's not always the most cost effective way of going about it.”