Microsoft Certified: Agentic AI Business Solutions Architect — Practice Set 1

Question 1

A solutions architect is scoping an AI project that will use customer data stored in Azure SQL and SharePoint to generate automated deal summaries for account executives. What is the FIRST thing the architect should assess about the data before designing the AI components?

Accepted Answer

The skills measured explicitly identify 'Review data for grounding, including accuracy, relevance, timeliness, cleanliness, and availability' as the required data assessment step before designing AI components. This determines whether the data is fit to ground AI responses reliably. Headcount (A), licensing cost (C), and storage format (D) are secondary considerations that do not directly determine grounding fitness.

Question 2

A retail company wants to adopt Microsoft AI services across its finance, HR, and customer service departments. The solutions architect has been asked to structure the adoption approach. Which framework should the architect use as the foundation for implementing AI adoption across the organization?

Accepted Answer

The skills measured content explicitly states 'Implement the AI adoption process from the Cloud Adoption Framework for Azure' as the required planning methodology for AI strategy. The Cloud Adoption Framework (CAF) provides structured guidance for enterprise-scale AI and cloud adoption at Microsoft. SCRUM (A) is an agile delivery methodology. ISO 27001 (C) is a security compliance standard. TOGAF (D) is a general enterprise architecture framework not specific to Microsoft AI adoption.

Question 3

Contoso is building a multi-agent solution to automate procurement approvals. The architecture team is debating which platform combination to use. The solution needs natural language policy guidance for employees (Microsoft 365 integration), custom workflow agents for procurement, and a machine-learning model for anomaly detection in invoices. Which platform combination should the architect recommend?

Accepted Answer

The skills measured explicitly calls out 'Design a multi-agent solution by using platforms such as Microsoft 365 Copilot, Copilot Studio, and Microsoft Foundry' — matching each platform to its strength: Copilot for Microsoft 365 integration, Copilot Studio for custom agent workflows, and Microsoft Foundry for custom AI models. Azure Logic Apps (A) is an integration platform without built-in LLM or agent capabilities. Using only Copilot Studio (C) would not address custom model training. SharePoint Syntex and Azure SQL (D) cannot fulfill the natural language or ML anomaly detection requirements respectively.

Question 4

A financial services firm wants to automate risk assessment for loan applications using AI. The solution will use both a large general language model for document understanding and a smaller, fine-tuned model for credit risk scoring. How should the architect handle routing requests between models?

Accepted Answer

The skills measured explicitly identifies 'Implement a model router to intelligently route requests to the most suitable model' as a cost and benefit optimization mechanism. Routing directs document understanding queries to the LLM and credit scoring queries to the specialized small model — maximizing accuracy and cost efficiency. Sending everything to the large model (A) is costly and may underperform on specialized scoring. Running both simultaneously (C) doubles cost with no benefit. Replacing the fine-tuned model with prompting (D) sacrifices the precision that fine-tuning provides for credit risk.

Question 5

A solutions architect is performing an ROI analysis for a proposed AI-powered customer service automation solution. Which TWO criteria should be included in the ROI evaluation according to the AB-100 skills measured? Choose 2.

Accepted Answer

The skills measured content specifies 'Select ROI criteria for AI-powered business solutions, including the total cost of ownership' (A) and 'Create an ROI analysis for the proposed AI solution for a business process' (C). TCO encompasses infrastructure and maintenance costs, while the business process ROI analysis quantifies the measurable improvement over the current state. Personal productivity of the architect (B), slide count (D), and language count (E) are not ROI criteria.

Question 6

A solutions architect is designing an AI solution for a mid-size logistics company that needs automated route optimization insights. The company cannot justify the cost of building a proprietary ML model. Which build-vs-buy decision framework should the architect apply?

Accepted Answer

The skills measured explicitly states 'Analyze whether to build, buy, or extend AI components for business solutions' — this three-way analysis weighs cost, time-to-market, and whether the capability is differentiating enough to warrant custom development. Blanket policies to always build (A) or always buy (C) ignore context. Avoiding data integration (D) would cripple most logistics AI solutions.

Question 7

A global manufacturing firm wants to use AI agents to streamline task automation across ERP, CRM, and HR systems. The architect is assessing how agents can be applied. What is the primary criterion for determining whether an agent is appropriate for a specific task?

Accepted Answer

The skills measured explicitly identifies 'Assess the use of agents in task automation, data analytics, and decision-making' as the requirement analysis step. Agents are appropriate across a wide range of scenarios — not limited to images (A), chatbots (C), or headcount thresholds (D). The fit is determined by whether agents can meaningfully automate, analyze, or support decisions in the target business context.

Question 8

A solutions architect is designing a Copilot Studio agent for a customer service scenario. The agent needs to gracefully handle queries that fall outside its defined topic scope — for example, questions about topics the agent was not trained to address. What must the architect design to handle this?

Accepted Answer

The skills measured explicitly includes 'Design topics for Copilot Studio, including fallback' — the fallback topic is the built-in Copilot Studio mechanism that gracefully handles queries outside the agent's defined topic coverage. A parallel AI model (A) adds unnecessary complexity. Power Automate to email (C) is a workaround, not a designed agent experience. A custom Azure Function for intent classification (D) is over-engineering when Copilot Studio's fallback topic handles this natively.

Question 9

A healthcare organization wants its Copilot Studio patient-support agent to interact with an external Electronic Health Records (EHR) system to retrieve appointment data. The EHR exposes a REST API following the Model Context Protocol (MCP). How should the architect design this integration?

Accepted Answer

The skills measured explicitly calls out 'Design agent extensibility with Model Context Protocol in Copilot Studio' — MCP is an open standard Microsoft supports natively in Copilot Studio for connecting agents to external services. Since the EHR already exposes MCP, the architect should use this direct integration path. Polling via Power Automate with SharePoint caching (A) adds latency and stale-data risk. Asking the vendor to rebuild to OData (C) is unreasonable. Converting MCP to SOAP with API Management (D) adds unnecessary complexity.

Question 10

Fabrikam wants to deploy an AI agent that can independently navigate and interact with its legacy internal web portal to extract data and perform form submissions on behalf of field technicians — without requiring API access or code changes to the portal. Which Copilot Studio capability should the architect propose?

Accepted Answer

The skills measured explicitly includes 'Design agents to automate tasks in apps and websites by using Computer Use in Copilot Studio' — this is the purpose-built Copilot Studio capability for interacting with UIs without API access. Power Automate Desktop (B) can do RPA but is a separate tool with its own infrastructure requirements. Azure Cognitive Services vision models (C) provide perception but not autonomous task execution. Azure Bot Service with speech (D) is a conversation interface, not UI automation.

Question 11

A solutions architect is designing an AI solution for a Dynamics 365 Customer Service implementation. The team wants to customize how Copilot assists agents with case resolution and extend it with additional knowledge from the company's internal product wiki. Which TWO design tasks are directly relevant? Choose 2.

Accepted Answer

Both 'Design customizations of Copilot in Dynamics 365 apps for customer experience and service' (A) and 'Design business terms for Copilot in Dynamics 365 apps for customer experience and service' (C) are explicitly listed in the skills measured under this domain. Business terms define company-specific vocabulary Copilot uses; customizations extend Copilot's behavior and knowledge sources. Conditional access (B), AKS clusters (D), and on-premises AD (E) are infrastructure or security concerns not part of the Copilot design tasks for this scenario.

Question 12

A solutions architect is choosing the natural language processing approach for a Copilot Studio agent that handles complex, free-form customer inquiries that don't fit predefined intents. The volume of inquiry types is too large to manage with hand-crafted intents. What orchestration approach should be selected?

Accepted Answer

The skills measured explicitly states 'Determine when to use standard natural language processing, Azure conversational language understanding, or generative AI orchestration in Copilot Studio.' When inquiry types are too numerous and free-form for predefined intents, generative AI orchestration is the appropriate choice — it handles open-ended queries without requiring exhaustive intent mapping. Standard NLP (A) and CLU (B) require predefined intents, which don't scale to this scenario. Regex matching (D) is far too rigid for natural language inputs.

Question 13

A solutions architect is designing an AI solution that extends Microsoft 365 Copilot to assist project managers in Microsoft Teams. The extension needs to pull live project status from an external project management API. Which design approach is most appropriate?

Accepted Answer

The skills measured includes both 'Design agents in Microsoft 365 Copilot' and 'Optimize solution design by using agents in Microsoft 365, including Teams and SharePoint.' Designing an M365 Copilot agent with a connector to the external API is the correct architecture for extending Copilot with live external data in Teams. A standalone Azure Bot (A) would not integrate into the Copilot experience. Power BI (C) provides visualization, not conversational AI assistance. A manually replicated SharePoint list (D) is not live and adds maintenance burden.

Question 14

An enterprise architect is designing AI solutions that span multiple Dynamics 365 apps — specifically Dynamics 365 Sales, Customer Service, and Finance. Each app requires AI components that interact with shared customer data. What should the architect prioritize in the overall design?

Accepted Answer

The skills measured explicitly states 'Design AI solutions that use multiple Dynamics 365 apps' — the architect should design for cross-app AI integration with shared data access, not isolated silos. Isolated components (A) would lead to inconsistent customer data across sales, service, and finance AI responses. Limiting to one app (C) fails the multi-app business requirement. Using a non-Microsoft platform (D) unnecessarily fragments the architecture.

Question 15

A solutions architect has deployed a Copilot Studio agent for field service technicians. The operations team reports that the agent sometimes provides outdated instructions. The architect wants to proactively identify when the agent's responses degrade before users report it. What should the architect recommend?

Accepted Answer

The skills measured includes both 'Recommend the process and tools required for monitoring agents' and 'Interpret telemetry data for performance and model tuning.' Proactive monitoring with telemetry analysis is the recommended approach to catch degradation before it impacts users at scale. Weekly manual testing (A) is reactive and infrequent. Replacing with a static knowledge base (C) removes AI capability. Increasing response length (D) does not address accuracy.

Question 16

A large insurance company has deployed an AI agent for claims intake. After three months, the product team has collected 500 user feedback submissions and a growing backlog of unresolved edge cases where the agent failed to route claims correctly. The architect needs to use this feedback to improve the agent. What process should the architect follow?

Accepted Answer

The skills measured explicitly identifies 'Analyze backlog and user feedback of AI and agent usage' and 'Apply AI-based tools to analyze and identify issues and perform tuning' as the deployment and improvement process. Analyzing failures in the backlog combined with AI-based tools for issue identification drives targeted tuning. Ignoring negative feedback (A) leads to recurring failures. Redeploying blindly (C) does not address root causes. Restricting usage (D) hides the problem rather than solving it.

Question 17

A solutions architect is designing the ALM (Application Lifecycle Management) process for a new Copilot Studio solution that includes custom agents, connectors, and actions. Which TWO ALM design elements are explicitly required for Copilot Studio components? Choose 2.

Accepted Answer

The skills measured lists both 'Design the ALM process for Copilot Studio agents, connectors, and actions' (A) and 'Design the ALM process for data used in AI models and agents' (B) as explicit AB-100 requirements. Data ALM ensures that training data and knowledge sources are versioned and governed alongside the agents. Agile sprint planning (C), server upgrades (D), and React pipeline configuration (E) are not AI-specific ALM design responsibilities for a Copilot Studio solution.

Question 18

A government agency is deploying an AI agent built with Microsoft Foundry that processes citizen benefit applications. The agency must comply with strict data residency laws requiring all data to remain within a specific geographic region. How should the architect address this?

Accepted Answer

The skills measured explicitly states 'Validate data residency and movement compliance' under the responsible AI, security, and governance domain. This requires the architect to actively design and validate that all data flows, model endpoints, and storage locations comply with the residency law — not assume automatic compliance. Documenting only in a charter (A) is insufficient. On-premises deployment (C) may not be necessary if the cloud region is configured correctly and may create operational burdens. Encryption alone (D) does not address geographic data residency requirements.

Question 19

A solutions architect is designing the security model for a Copilot Studio agent that accesses sensitive HR records for grounding. Only HR business partners should be able to trigger queries that retrieve employee salary data. How should the architect enforce this?

Accepted Answer

The skills measured explicitly calls out 'Design access controls on grounding data and model tuning' — the proper mechanism is enforcing data-layer access controls so the grounding data source itself enforces authorization, regardless of prompt content. System prompt instructions (A) can be bypassed with jailbreak prompts and do not enforce technical access control. Open SharePoint storage (C) removes all access control. Relying on user judgment (D) provides no technical safeguard.

Question 20

A solutions architect is reviewing an AI solution that uses Microsoft Foundry to fine-tune a language model on proprietary company data. A security audit has flagged potential vulnerabilities. Which TWO vulnerabilities should the architect address as part of the responsible AI and security design? Choose 2.

Accepted Answer

The skills measured explicitly identifies 'Analyze solution and AI vulnerabilities and mitigations, including prompt manipulation' (A) and 'Design audit trails for changes to models and data' (C) as required security design elements. Prompt manipulation is an AI-specific attack vector that must be mitigated. Audit trails ensure accountability for model changes. Font rendering (B), color scheme (D), and browser tabs (E) are UI/UX considerations, not AI security vulnerabilities.

Question 21

A solutions architect is building a testing strategy for a multi-agent AI solution that spans Dynamics 365 Sales, Customer Service, and Finance. What type of test scenarios must be included in the testing strategy?

Accepted Answer

The skills measured explicitly states 'Design end-to-end test scenarios of AI solutions that use multiple Dynamics 365 apps.' When the solution spans Sales, Customer Service, and Finance, end-to-end tests must validate cross-app data flows and agent handoffs. Unit tests alone (A) miss integration failures. Load tests alone (C) and security tests alone (D) are important but insufficient — end-to-end functional validation of the multi-app AI flows is required.

Question 22

A solutions architect is reviewing an AI solution before go-live. The legal team asks her to confirm that the solution complies with responsible AI principles, including fairness, reliability, privacy, and transparency. What should the architect do as part of deployment governance?

Accepted Answer

The skills measured explicitly states 'Review solution for adherence to responsible AI principles' as a deployment governance requirement. Responsible AI review is independent of functional testing (A) and is the architect's responsibility, not the infrastructure team's (C). While third-party ethics boards may be relevant in some regulatory contexts, the exam's focus is on Microsoft's responsible AI guidelines (D).

Question 23

A retail company has deployed a Microsoft Foundry-based AI agent for inventory forecasting. The data science team reports that the agent's predictions have degraded significantly after a change to the product catalog data pipeline. The architect needs to diagnose whether the degradation is due to model drift or a data quality issue. What should the architect do?

Accepted Answer

The skills measured includes 'Interpret telemetry data for performance and model tuning' — telemetry analysis is the diagnostic step to distinguish between data pipeline issues and model drift before deciding on remediation. Re-training from scratch (A) is premature and expensive without a diagnosis. Rolling back without investigation (C) only treats the symptom. Replacing the model (D) avoids the problem rather than solving it.

Question 24

A solutions architect is designing the ALM process for AI models that are fine-tuned using Microsoft Foundry. The model will be retrained quarterly as new business data arrives. What must the ALM process explicitly include for these custom AI models?

Accepted Answer

The skills measured explicitly includes 'Design the ALM process for custom AI models' — proper ALM for ML models encompasses versioning, automated validation criteria, staged deployment, and rollback capability. Manual email approvals (A) and SharePoint tracking lists (D) are manual workarounds that don't constitute a robust ALM process. Third-party vendor requirements (C) are a procurement decision, not an ALM design requirement.

Question 25

A solutions architect must design the governance model for agents deployed in Copilot Studio and Microsoft Foundry for a regulated financial services firm. Which TWO governance design tasks are explicitly required by the AB-100 skills measured? Choose 2.

Accepted Answer

The skills measured explicitly lists both 'Design governance for agents' (A) and 'Design security for agents' (B) as required deployment responsibilities. In a regulated environment, governance defines who can create, modify, and retire agents, while security design covers authentication, authorization, and network controls. Marketing strategy (C), office furniture (D), and programming language selection (E) are outside the scope of agent governance and security design.

AB-100 — Agentic AI Business Solutions Architect