Choosing the right Copilot and Azure AI service
Microsoft offers many AI products with similar names. The exam rewards leaders who match a business need to the right one quickly.
The Copilot family
| Product | Use it for |
|---|---|
| Microsoft 365 Copilot | AI in Word, Excel, PowerPoint, Outlook and Teams, grounded in work data. Per-user add-on license. |
| Microsoft 365 Copilot Chat | Free, web-grounded chat for Entra ID users. Agents can be used pay-as-you-go. |
| Copilot Studio | Low-code building of custom agents and workflows that connect to business systems. |
| Microsoft Security Copilot | Security operations. |
| Copilot in Azure | Help for cloud administration. |
Agents such as Researcher and Analyst come with Microsoft 365 Copilot. Researcher searches across work data such as SharePoint, so it needs the per-user Microsoft 365 Copilot license.
What data Copilot uses
Microsoft 365 Copilot uses Microsoft Graph by default: emails, files in OneDrive and SharePoint, meetings and chats the user can already open. It does not bypass permissions. Data outside Microsoft 365, such as an on-premises file share or a custom CRM, is not used unless you add a Copilot connector or build an agent that reaches it.
Extending Copilot
- Declarative agent (lite experience in Copilot Studio): the cheapest, lowest-effort way to tailor Copilot to a process.
- Custom agent (full experience in Copilot Studio): for complex, custom workflows and automation.
- Microsoft Foundry (formerly Azure AI Foundry): the platform for developers building custom AI apps and agents with models, including Azure OpenAI.
When the scenario says “minimize development effort and cost”, pick the low-code option, not Foundry.
Azure AI services in the exam
| Need | Service |
|---|---|
| Index and search large document sets, knowledge mining | Azure AI Search |
| Extract fields from invoices and forms | Azure Document Intelligence (in Foundry Tools) |
| Analyze images | Azure Vision |
| Predict from historical data | Azure Machine Learning |
| Build and manage custom AI apps | Microsoft Foundry |
Many of these were called “Azure AI services” or “Azure Cognitive Services” before; the exam may use either.
Pricing models for Azure OpenAI
Standard (on-demand) charges per use and suits pilots. Provisioned (PTUs) reserves capacity for predictable, high-volume workloads. Batch suits large non-urgent jobs.
Exam traps
- Copilot Studio builds agents; Microsoft 365 Copilot is the ready-made assistant.
- An on-premises file share is not a default Copilot data source.
- Document Intelligence extracts data; Azure AI Search finds it.
- Machine Learning is for prediction, not for summarizing transcripts.
- Provisioned throughput is not the starter option for a proof of concept.
Practice questions
Choose an answer, then review the explanation to check your understanding.
1. An HR team wants an agent that answers employee questions and connects to internal HR data sources. What should you recommend?
Show answer
Answer: B.Copilot Studio is the low-code tool for building agents that automate conversations and connect to business data.
2. By default, which data can Microsoft 365 Copilot use to answer a user?
Show answer
Answer: B.Copilot works through Microsoft Graph and respects the user's existing permissions.
3. Accounts payable needs to extract invoice numbers, vendor names and totals from invoices automatically. Which service fits best?
Show answer
Answer: B.Document Intelligence extracts structured fields from forms and invoices.
4. A pilot agent must start small and scale cost with demand. Which Azure OpenAI pricing model fits?
Show answer
Answer: B.Standard is pay-as-you-go, ideal for proofs of concept. Provisioned throughput suits steady, high-volume production.