AB-100 — Frequently Asked Questions

Dataverse stores the data, but the AI system needs an explicit governance signal (endorsement) to trust and query the semantic model.

Both are forms of endorsement that make a model usable for AI grounding. Certified implies higher rigor, but both satisfy the 'trusted source' requirement.

The Admin Center manages environments and security but does not provide the financial modeling templates required to calculate ROI or ROAI.

ROAI specifically measures the return on investment derived from AI-enabled features within a solution, whereas ROI is a broader term.

Success by Design is an implementation methodology providing checklists and steps for deployment. The Well-Architected Framework provides architectural principles and best practices for design and quality.

The Microsoft Power Platform Well-Architected Framework contains the specific guidance for Experience Optimization and Conversational User Experience design.

Success by Design focuses on implementing Dynamics 365 and Power Platform business applications, whereas CAF covers the broader infrastructure and cloud strategy.

The CoE provides starter kits and governance tools to manage, monitor, and secure the Power Platform environment after it has been established.

An Autonomous Agent manages complex, multi-step workflows independently without constant human input. A Task Agent performs specific, bounded business functions or analyses, often triggered by events or schedules.

Prompt-and-response agents are conversational and require user initiation for each step. Order fulfillment requires automated, sequential execution across systems, which only an Autonomous Agent can handle reliably.

Copilot Studio is the platform for building and configuring AI agents. Copilot in Power Apps is a feature within Power Apps that allows users to interact with those agents via natural language.

Use Microsoft Foundry when you need deep customization, integration with Semantic Kernel, complex orchestration, or multi-agent patterns that go beyond the standard low-code capabilities of Copilot Studio.

Cost Management tracks actual spend of running resources. ROAI requires estimating potential future costs before the solution is deployed.

While often used for migration comparisons, it is also the standard tool for modeling the total cost of a new proposed solution including infrastructure and services.

AI Summaries provide quick, personalized overviews of key metrics across various pages. Collections Coordinator Summary is a specialized tool specifically for managing customer debt, payment history, and follow-ups.

While the Supplier Communication agent helps with vendor interactions, AI Summaries with Copilot are more direct for simply identifying and surfacing overdue invoice details to staff for initial review.

Log Analytics is a data repository. Application Insights is the instrumentation service that captures the detailed performance metrics and dependencies you need.

Power BI visualizes data but does not collect telemetry. You would typically send Application Insights data to Power BI for advanced reporting after collection.

Log Analytics stores raw logs but lacks the pre-built, comprehensive agent performance dashboards and metrics found in the Analytics tab.

Microsoft Purview is used for data governance, compliance, and risk management, not for real-time agent performance or telemetry monitoring.

Agent management focuses on activating and configuring already selected agents. Immersive Home is the dedicated entry point for discovering and browsing available templates.

No. Copilot Studio is for building custom agents. Prebuilt agents for D365 are discovered through the Immersive Home interface within the application.

Surveys measure user satisfaction, not technical accuracy. They don't help you debug the agent's logic or knowledge base.

While B shows failed sessions, it doesn't analyze the text content. E directly flags poor quality answers based on grounding and relevance.

Blob Storage has native, granular version control features that explicitly track every change to an object, which directly satisfies the requirement to 'track and version data changes'.

It binds the fine-tuning job to a specific, governed dataset source, ensuring that every retraining run uses the exact same approved data rather than pulling from undefined sources.

Azure Policy enforces organizational standards and scope at deployment time (preventive). Microsoft Defender for Cloud continuously monitors and assesses security posture and compliance after deployment (detective/corrective).

No. Microsoft Purview is for data governance and classification. Azure Policy is the native tool for controlling resource creation parameters like location, type, and tags.

Limiting retrieved content via prompts can inadvertently exclude critical context required for accurate answers, thus degrading quality.

No, proper chunking improves accuracy by ensuring the model receives semantically focused, relevant snippets rather than noisy, large documents.