Extending the Azure AI for Operations Framework with Copilot Studio Agents for FinOps and Update Management
Solution ideas
The original framework introduced several Logic App and Function App patterns for SQL BPA, Update Manager, Cost Management, Anomaly Detection, and Smart Doc creation.
In this article we add two Copilot Studio Agents, packaged in the GitHub repository Microsoft Azure AI for Operation Framework, designed to be deployed in a dedicated subscription (e.g., OpenAI-CoreIntegration):
- Copilot FinOps Agent – interactive cost & usage analysis
- Copilot Update Manager Agent – interactive patch status & one-time updates
Architecture
Copilot FinOps Agent
A Copilot Studio agent that lets stakeholders chat in natural language to retrieve, compare, and summarise cost data—without leaving Teams.
Dataflow
| # | Stage | Description |
|---|---|---|
| Initial Trigger | User message (Teams / Copilot Studio web) invoke topic | The conversation kicks off the topic “Analyze Azure Costs”. |
| 1 | Pre-Processing | Power Automate flow captures tenant ID, subscription filters, date range. |
| 2 | Cost Query | Azure Cost Management APIs pull actual and previous spend, returning JSON rows (service name, cost €). |
| 3 | OpenAI Analysis | Data is analyzed by OpenAI\Copilot Agent following the flow structure. |
| 4 | Response Formatting | Copilot Studio flow format the output as a table. |
| 5 | Chat Reply | Copilot agent posts the insight list. Users can ask any kind of question related the FinOps topic. |
Components
- Microsoft Copilot Studio (Developer licence) – low-code agent designer
- Power Automate Premium – orchestrates REST calls, prompt assembly, file handling
- Azure Cost Management + Billing – source of spend data (Rest API)
- Azure OpenAI Service – GPT-4o and o3-mini reasoning & text generation
- Microsoft Teams – chat surface for Q&A, cards, and adaptive actions
Potential use cases
- Finance teams asking “Why did VM spend jump last week?”
- Engineers requesting a monthly cost overview before sprint planning
- Leadership dashboards that can be drilled into via natural-language chat
Copilot Update Manager Agent
A Copilot Studio agent that surfaces patch compliance and can trigger ad-hoc One-Time Updates for selected VMs directly from the chat.
Dataflow
| # | Stage | Description |
|---|---|---|
| Initial Trigger | User message (Teams / Copilot Studio web) invoke topic. | The conversation kicks off the topic “Analyze Azure Costs”. |
| 1 | Pre-Processing | Flow validates RBAC and captures target scope (subscription / RG / VM). |
| 2 | Patch Status Query | Azure Update Manager & Resource Graph query patchassessmentresources for KBs, severities, pending counts. |
| 3 | OpenAI Report | GPT-4o - o3-mini generates: • VM-level summary (English) • General Overview |
| 4 | Adaptive Card | Power Automate builds an Adaptive Card listing non-compliant VMs with “One-time Update”- "No action" buttons. |
| 5a | User Action – Review | User inspects details or asks follow-up questions. |
| 5b | User Action – Patch Now | Clicking One-time Update calls Update Manager REST API to start a One-Time Update job. |
| 6 | Confirmation | Agent posts job ID, live status, and final success / error summary. |
Components
- Microsoft Copilot Studio – conversational front-end
- Power Automate Premium – API orchestration & status polling
- Azure Update Manager – compliance data & patch execution
- Azure OpenAI Service – explanation & remediation text
- Microsoft Teams – Adaptive Cards with action buttons
Potential use cases
- Service owners getting a daily compliance digest with the ability to remediate on demand
- Security officers validating zero-day patch rollout status via chat
- Help-desk agents triaging “Is VM X missing critical updates?” without opening the Azure portal
Prerequisites
| Resource | Quantity | Notes |
|---|---|---|
| Copilot Studio Developer licence | 1 | Assign in Microsoft 365 Admin Center |
| Power Automate Premium licence | 1 user | Needed for HTTP, Azure AD, OpenAI connectors |
| Microsoft Teams | 1 user | Chat interface |
| Azure subscription | 1 | Dedicated OpenAI-CoreIntegration recommended |
| GitHub repo | latest | Microsoft Azure AI for Operation Framework Copilot Agent |
Copilot Studio User Experience
Deployment steps (high level)
- Assign licences – Copilot Studio Developer + Power Automate Premium
- Create Copilot Studio Agent
New Agent → Skip to configure → fill basics → Create → Settings → disable GenAI orchestration - Import topics
- Copilot topic Update Manager (link to configuration file)
- Copilot topic FinOps (link to configuration file)
- Publish & share the agent to Teams.
- Verify permission scopes for Cost Management and Update Manager APIs.
- Start chatting!
Feel free to clone the GitHub repo, adapt the topics to your tag taxonomy or FinOps dashboard structure, and let us know in the comments how Copilot Agents are transforming your operational workflows and... Stay Tuned for the next updates!
Contributors
Principal authors
Tommaso Sacco | Cloud Solutions Architect
Simone Verza | Cloud Solution Architect
Special thanks
Carmelo Ferrara | Director CSA
Antonio Sgrò | Sr CSA Manager
Marco Crippa | Sr CSA Manager
Microsoft