Microsoft - AB-100: Agentic AI Business Solutions Architect

Sample Questions

Explanation:

The scenario highlights two key challenges:

• The company has no AI governance framework.
• The staff has limited AI experience.

According to Microsoft's AI adoption guidance in the Cloud Adoption Framework (CAF) for Azure, organizations should begin by:

1. Defining an AI strategy aligned to business goals.
2. Establishing AI governance and responsible AI practices.
3. Building organizational readiness, skills, and operating models.
4. Then scaling AI solutions and deployments.

Deploying AI solutions before governance and strategy are in place increases risks related to security, compliance, data management, and responsible AI.

References:

AI strategy - Guidance to set your organization's AI strategy

Guidance to set up your organization's AI governance process

Explanation:

The requirements are:

• The company already uses Dynamics 365 Sales.
• They need AI-powered recommendations and sales insights.
• They must integrate with a proprietary risk scoring system hosted in Azure.
• They want to balance time-to-value with custom business requirements.

Microsoft's sales-focused Copilot is designed to work with Dynamics 365 Sales and can be extended with custom integrations and business-specific data sources. Microsoft documentation specifically describes extending the Sales agent/Copilot for Sales with partner or custom applications and APIs to bring additional insights and recommendations into the sales workflow.

This approach provides:

• Faster implementation than building an AI solution from scratch.
• Reuse of existing Dynamics 365 Sales investments.
• Integration of proprietary risk scores through custom connectors.
• Context-aware recommendations for relationship managers.

References:

Extend Sales agent with partner applications

Manage extensions for Sales in Microsoft 365 Copilot

Explanation:

We should create separate topics for policy enquiries, claims submissions, and general questions, with a fallback topic that collects customer details and creates a support ticket.

This follows Microsoft Copilot Studio best practices:

• Each major business function should have its own topic to keep conversation flows manageable and easier to maintain.
• A fallback topic should handle situations where the agent cannot determine the user's intent.
• The fallback topic can collect information such as customer name, policy number, contact details, and issue description, then invoke backend processes (for example, creating a support ticket). Microsoft documentation specifically describes using the fallback topic to capture unrecognized intent and call backend systems or workflows.

References:

Configure the system fallback topic

Use the Fallback topic

Explanation:

We should export the agent as part of a Power Platform solution, including the connectors and knowledge source configurations, and import it into the production environment through the managed solution pipeline.

For Copilot Studio ALM, agents and their dependent components should be managed through Power Platform solutions. The solution can package the agent together with its related components, allowing controlled movement between development, test, and production environments.

The appropriate flow is:

Development → Test → Production

Using a managed solution for production provides controlled deployment and helps maintain a consistent, repeatable ALM process.

The solution should account for dependencies such as:

• Custom topics
• Custom connectors
• Knowledge-source configuration
• Other Power Platform components required by the agent

For environment-specific resources such as a SharePoint site or connection, environment variables and connection references can be used rather than hard-coding development resources.

References:

Establish an application lifecycle management strategy

Create and manage solutions in Copilot Studio

Explanation:

We should review the agent's conversation transcripts and topic analytics to identify which topics are triggering incorrect responses, then analyse the knowledge source for gaps in warranty policy content.

Copilot Studio provides monitoring and analytics that can be used to diagnose agent behavior. For incorrect warranty answers, the support team should:

1. Review conversation transcripts to see the actual questions and responses.
2. Use topic/agent analytics to identify patterns and determine which topics or scenarios are generating poor responses.
3. Examine the knowledge source to determine whether warranty policies are missing, outdated, ambiguous, or poorly structured.
4. Correct the underlying topic or knowledge content and then test the agent again.

This is a data-driven troubleshooting approach rather than immediately rebuilding or disabling the agent.

References:

Monitor overview

Monitor conversational agents

Understand downloaded session data from Copilot Studio