AB-100 Practice Questions

The scenario requires a secure, private, and auditable integration for a healthcare provider handling Protected Health Information (PHI). An authenticated Omnichannel adapter provides a secure, pre-built connection between the agent and the contact center platform (Dynamics 365). This meets the "no public exposure" and "auditable handoff" requirements by leveraging the platform's native authentication and security model. It avoids custom code, aligns with the exhibit showing a contact center integration, and is designed for low-latency, high-volume interactions, satisfying all constraints for the initial production release.

A. A direct Teams bot endpoint may not integrate seamlessly with a multi-channel contact center's routing and reporting, and could require custom code for PHI governance.

B. Exposing a public webchat directly violates the "avoid public exposure" constraint, which is critical for handling sensitive PHI data.

C. An email channel is asynchronous and does not meet the "tight response-time target" required for a real-time triage scenario.

1. Microsoft Learn, "Integrate a Power Virtual Agents bot". Section: "Prerequisites". This documentation outlines connecting a bot to Dynamics 365 Omnichannel for Customer Service, which uses a secure adapter.

2. Microsoft Learn, "Configure authentication settings in Omnichannel for Customer Service". This document details how to set up authenticated conversations, ensuring only verified users can interact, which is crucial for PHI.

3. Microsoft Learn, "Overview of security and compliance in Dynamics 365". This resource discusses the platform's capabilities for meeting regulatory requirements like HIPAA for PHI.

This combination provides a secure, compliant, and integrated solution using the Microsoft stack.

(B) Copilot Studio agent in Customer Service allows for extending the native Copilot within the existing agent workspace, avoiding a new chat surface.

(D) Dataverse knowledge search with security roles leverages the built-in, secure data platform for grounding, respecting PII regulations and security boundaries without an external data store.

(F) Power Automate approvals for high-risk actions is the standard, low-code method to integrate approval workflows directly with Copilot Studio actions, ensuring that sensitive operations like changing a case status are properly controlled and auditable.

A. Out-of-box Copilot only: This lacks the required customizability for specific actions and approval workflows, failing to meet the high-risk action requirement.

C. Custom Azure-hosted bot with separate audit data lake: This introduces a new chat surface and an external data store, violating explicit constraints of the scenario.

E. External vector database for case transcripts: This violates the constraint of not introducing a new external data store for grounding.

1. Microsoft Learn. (2024). Overview of extending Copilot in Dynamics 365 Customer Service. This document states, "You can extend Copilot to use... actions through Power Automate flows... by creating custom topics in Microsoft Copilot Studio." This directly supports using Copilot Studio (B) and Power Automate (F).

2. Microsoft Learn. (2024). Use actions in Microsoft Copilot Studio. It details how to "call a Power Automate flow from a topic" to perform actions, including those requiring approvals.

3. Microsoft Learn. (2024). Security concepts in Microsoft Dataverse. This documentation explains how security roles and field-level security protect sensitive data (like PII) within Dataverse, supporting its use for secure grounding (D).

The best design is Entra group-based Role-Based Access Control (RBAC) per environment. This approach directly meets the requirements by allowing the creation of a specific security group in Microsoft Entra ID for the supervisors. This group can then be granted permissions to use the agent's submission capabilities only within the production environment. This enforces the principle of least privilege, supports environment separation (dev/test/prod), and contains the governance to the specific solution without imposing tenant-wide restrictions, as requested.

B. Environment Admin role for agent identity: This violates the principle of least privilege by granting excessive permissions to the agent's service principal.

C. Delegated user token for tool calls: While useful for auditing, this doesn't enforce the rule that only supervisors can submit; it relies on the tool's permissions, which is better managed by RBAC.

D. Tenant-wide connector allowlist with conditional access policies enforced: This is a tenant-level control that would create the "tenant-wide friction" the requirements explicitly forbid.

1. Microsoft Power Platform documentation, "Environment strategy for ALM." This guide emphasizes using distinct environments for development, testing, and production to isolate changes and manage access.

2. Microsoft Power Platform documentation, "Security concepts in Microsoft Power Platform," Section: "Microsoft Entra security groups." It states, "Using security groups is a best practice for securing user access to Power Platform resources."

3. Microsoft Power Platform documentation, "Share a bot with security groups." This page details the specific mechanism for using Entra ID (formerly Azure AD) security groups to control access to bots (agents) on an environment-by-environment basis.

A declarative agent (built with a low-code tool like Copilot Studio) combined with a Microsoft Graph connector is the optimal approach. The Graph connector securely ingests the bank's policy data into the Microsoft Graph, keeping it within the compliant tenant boundary and region. The declarative agent provides a centrally governed, auditable, and low-code way to build the conversational experience and trigger actions. Approvals for the "HoldPayment" action can be implemented using Power Automate, which integrates seamlessly with Copilot Studio, satisfying the separation of duties and governance requirements with a faster time-to-deliver.

A. Custom engine agent with backend orchestration: This is a high-effort, code-intensive approach that increases time-to-deliver and management overhead compared to a declarative solution.

B. Copilot Studio agent with public web search enabled: Enabling public search is explicitly forbidden by the "No public exposure" requirement and poses a significant data exfiltration risk for a regulated bank.

C. API plugin with delegated admin consent: This describes a component, not a full solution. It doesn't address how policy questions are answered and relies on a potentially risky consent model.

1. Microsoft Copilot Studio Documentation, "Overview of enterprise chatbots". This outlines the benefits of using a low-code, declarative platform for building and governing enterprise-grade conversational AI.

2. Microsoft Search Documentation, "Microsoft Graph connectors overview". Section: "Key benefits". This document explains how connectors bring external content into the Microsoft 365 ecosystem securely, respecting source permissions and tenant boundaries.

3. Microsoft Power Automate Documentation, "Create approval flows with Power Automate". This shows how to build and manage approval workflows that can be triggered from various services, including Copilot Studio, to meet governance needs.

A. An app-packaged declarative agent definition (e.g., a manifest file) enables consistent, repeatable deployments and central governance of the agent's capabilities across dev/test/prod environments, which is a core requirement.

C. Using connector-based grounding with Access Control List (ACL) trimming allows the agent to access supplier data in-place, respecting data residency and the "no new data movement" constraint, while enforcing user-specific security permissions.

D. An action allowlist with approval gates provides a robust safety and governance mechanism. It restricts the agent to only pre-approved actions and enforces separation of duties for high-risk operations, meeting a key safety requirement.

B. Per-user prompt override in Teams chat threads: This approach undermines central governance, making the agent's behavior inconsistent and difficult to audit, which is contrary to the requirements.

E. Separate agent per subsidiary tenant: This would lead to fragmented management and violate the "centrally governed" requirement, increasing operational complexity and cost.

F. Direct embedding of SOPs into prompts: This is not a scalable or maintainable solution for grounding. It is subject to token limits and is far less effective than using a dedicated retrieval mechanism like a connector.

1. Microsoft Copilot Studio Documentation, "Plugins for Microsoft Copilot". This explains how plugins (actions) can be defined and governed, including the use of allowlists and approval flows for safety. (See section: "Build and use plugins").

2. Microsoft Learn, "Implement a data connection for Microsoft 365 Copilot". This resource details how Microsoft Graph connectors index data in-place, respecting source permissions (ACLs) and avoiding data duplication. (See section: "Security and compliance").

3. Microsoft Power Automate Documentation, "Create and run approval flows". This document describes how to build approval gates for actions, which can be triggered by agents to enforce separation of duties. (See section: "Get started with approvals").

The solution requires a code-first, AI-driven user interface, a supervision surface, and strict governance controls. A generative page (B) provides the modern, code-first React-based UI for case triage within the model-driven app. An agent feed with approval-required actions (D) is the designated built-in surface for supervising autonomous tasks and enforcing human-in-the-loop for high-risk actions, ensuring auditability. An environment-level Data Loss Prevention (DLP) policy (F) allows a central admin to restrict connector usage, enforcing private-only data access and maintaining compliance across all environments.

A. Public website grounding for policy updates: This violates the explicit "private-only access" and in-region data residency requirements for a regulated entity.

C. Power Apps code app for custom React portal: A generative page is a more integrated and specific solution for a page within a model-driven app than a standalone code app.

E. Tenant setting for Copilot in Power Apps: This is a general enablement switch, not the specific granular control needed for connector restrictions, which is managed by DLP policies.

1. Microsoft Learn. (2024). Add an AI-generated page to a model-driven app (preview). In Power Apps documentation. Section: "How it works". This document describes how generative pages create a React web resource for a code-first UI.

2. Microsoft Learn. (2024). Microsoft Copilot Studio: Human-agent escalation and monitoring. This documentation outlines the patterns for agent supervision and human-in-the-loop approvals, which are surfaced in an agent feed.

3. Microsoft Learn. (2024). Data loss prevention (DLP) policies. In Power Platform documentation. Section: "Environment-level DLP policies". This explains how administrators can scope policies to specific environments to control connector availability.

Microsoft’s prompt-engineering guidance for Copilot Studio and Microsoft 365 recommends: 1) a consistent role–task–context–output pattern to reduce variance, 2) explicitly governed tool invocation with approval gates to satisfy ALM and regulatory obligations, and 3) source-controlled, versioned prompt assets per dev/test/prod so every environment carries identical, approvable content and in-region storage. These three directly address consistency, governance/approvals, and reliability (≤3 s latency) without encouraging unsupported “best-guess” responses.

B. Citations are useful but not mandated for internal agents and do not solve governance or latency.

C. Self-critique chains add latency and may leak chain-of-thought, violating Microsoft guidance.

F. Higher temperature increases randomness, hurting consistency and reliability.

1. Microsoft Learn – “Engineering effective prompts for Microsoft Copilot” § Pattern (role-task-context-output).

2. Microsoft Power Platform ALM documentation – “Solutions and environments” § Version control of prompt assets.

3. Microsoft Responsible AI Standard, v2 § Input and tool governance – approval requirements.

4. Azure OpenAI latency guidance – <https://learn.microsoft.com/azure/ai-services/openai/concepts>.

The correct sequence prioritizes safety, governance, and business rules before technical implementation and testing. First, defining decision boundaries and approval gates (3) establishes the fundamental safety and business logic for high-risk actions. Next, binding the agent to governed data sources (4) enforces data security and access control. With the guardrails in place, end-to-end scenario evaluation and regression checks (1) can be safely performed. Finally, configuring monitoring (2) is a post-deployment or late-stage pre-deployment step to observe the agent's performance, cost, and reliability in a production-like environment. This order ensures a secure, compliant, and well-tested rollout.

A. This option incorrectly places testing (1) before defining the critical safety and decision boundaries (3), which is a high-risk approach.

C. This option incorrectly prioritizes monitoring (2) before testing (1) and defining business rules (3), which is illogical in a standard development lifecycle.

D. This option incorrectly starts with monitoring (2), which is typically one of the final steps before or immediately after production deployment, not the first.

1. Microsoft. (2023). Responsible AI Standard, v2. "Principle: Human agency and oversight," Section 2.4, p. 13. This document emphasizes establishing human oversight mechanisms (like approval gates) as a foundational step.

2. Microsoft Azure Well-Architected Framework. (2024). Security pillar. "Design principles: Protect data." This resource details the importance of establishing data access controls and binding services to governed sources early in the design phase.

3. Microsoft. (2023). Guidance for production machine learning systems. "MLOps stage: Model deployment." This documentation outlines that comprehensive testing and validation precede monitoring in a production rollout.

The requirements for environment alignment, no auto-upgrades, SLA adherence, stable costs, and fast rollback capabilities point directly to a controlled, version-specific deployment strategy. A "Pinned deployment" explicitly links a specific model version to an environment (e.g., prod uses v1.2), preventing unexpected changes from auto-upgrades and ensuring cost stability. A "rollback runbook" is a predefined procedure to quickly revert to a previous stable version, directly satisfying the requirement to roll back within a single change window. This combination provides the necessary stability, control, and safety for a mission-critical system.

A. A single shared deployment is a poor practice that eliminates environment isolation, making it impossible to test changes without risking the production environment.

C. Fine-tuning without governance directly contradicts the requirement for central enforcement and alignment across dev/test/prod, leading to instability and non-compliance.

D. Auto-upgrading to the latest version is explicitly forbidden by the constraint "can’t allow auto-upgrades" due to the risk of SLA-breaching outages.

1. Microsoft. (2024). MLOps: Model management, deployment, and monitoring with Azure Machine Learning. Microsoft Learn. This document details MLOps best practices, including controlled rollout strategies. The concept of deploying specific, registered model versions aligns with "pinned deployment" to ensure predictability and enable rollbacks. See the "Deploy and manage models with MLOps" section.

2. Microsoft. (2024). Deploy models for inference and prediction in Microsoft Fabric. Microsoft Learn. The documentation on model deployment in Fabric emphasizes versioning ("ML model versions"). Pinning is the act of selecting a specific version for a deployment endpoint, providing the stability required by the scenario.

3. Kreuzberger, D., et al. (2023). Machine Learning Operations (MLOps): Overview, Definition, and Architecture. IEEE Access, vol. 11, pp. 31646-31661. DOI: https://doi.org/10.1109/ACCESS.2023.3262108. This academic review of MLOps highlights the importance of reproducibility and versioning of all assets, including models, as a core principle for reliable production systems. See Section III-B, "Reproducibility."

The constraints—tight deadline, strict compliance, no new connectors, and the need for quick value demonstration—make a "big bang" enterprise rollout infeasible. A phased pilot with KPI gates (D) is the best strategy. This approach de-risks the project by starting with a limited scope, allowing the team to prove value quickly and secure stakeholder buy-in. It respects compliance by tackling systems one by one, adheres to the "least privilege" principle, and allows for predictable cost management. Each phase's success, measured by KPIs, gates the investment for the next phase, ensuring alignment with business value.

A. An enterprise-first rollout is too high-risk, slow, and likely to violate the strict compliance and least-privilege access constraints from the start.

B. Choosing the lowest-cost model might not meet the complex cross-system requirements and could fail to deliver the value the COO expects.

C. Building a custom model in-house is time-consuming and expensive, directly contradicting the six-week timeline and predictable cost requirements.

1. Microsoft Cloud Adoption Framework for Azure. (2024). Strategy methodology. "Define motivations." This framework advocates for identifying business outcomes and starting with projects that can demonstrate value early to build momentum.

2. Microsoft Power Platform Adoption. (2023). Power Platform adoption maturity model. "Stage 2: Envision." This model emphasizes starting with pilot projects and defining success criteria (KPIs) to guide a phased and successful enterprise adoption.

3. Pressman, R. S., & Maxim, B. R. (2020). Software Engineering: A Practitioner's Approach (9th ed.). McGraw-Hill. Chapter 4, "An Agile View of Process," discusses the value of incremental delivery for managing complexity and stakeholder expectations.