Practice the AB-250 domain for configuring agents and AI capabilities. Questions cover Copilot assistance, knowledge sources, voice agents, IVR, speech, DTMF, and multilingual experiences.
10 questions covering this official AB-250 domain.
Q01 - Question
A bank must deliver a fixed phone menu for payment options. Every prompt and path is reviewed by the compliance team, callers select options by using DTMF keypad input, and the wording must not vary between calls. Which orchestration mode should you use for the voice agent?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Generative orchestration, because it dynamically adapts prompt wording.
- B. Generative orchestration, because it is required for collecting DTMF input.
- C. Classic orchestration, because it uses deterministic topics and trigger phrases.
- D. Classic orchestration, because it bypasses the one-hour limit on agent-handled calls.
C is correct.
Explanation: Classic orchestration uses deterministic topics and trigger phrases, which suits menu-based, compliance-sensitive, and DTMF-heavy flows where wording must remain fixed.
A is incorrect: Generative orchestration dynamically interprets intent and context, meaning responses can vary, which conflicts with the fixed wording required by compliance.
B is incorrect: Generative orchestration is meant for open-ended conversation and knowledge scenarios, not exclusively for DTMF input.
D is incorrect: The one-hour limit applies to agent-handled calls generally, and classic orchestration does not bypass this architectural limit.
Q02 - Question
You are configuring security and compliance settings for a voice agent. The agent collects credit card numbers and uses a workstream that records calls. You need to ensure that recording consent is obtained and that credit card numbers are not captured in the recording. What should you configure?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Configure consent before recording starts, and collect the card number with DTMF while recording is paused.
- B. Start recording immediately, and rely on transcription timeout settings to mask the card number.
- C. Ask callers to speak the card number, and use Event Grid to delete the recording after the call.
- D. Configure data residency policies to automatically encrypt the recording.
A is correct.
Explanation: To comply with varying jurisdiction requirements, configure consent before recording starts. Then, use DTMF and recording pause to ensure sensitive data like credit card numbers are not captured.
B is incorrect: Starting the recording immediately violates the requirement to obtain consent first, and timeout settings control input durations rather than masking sensitive audio.
C is incorrect: Event Grid is an architectural prerequisite; asking callers to speak the card number records the sensitive data, which violates the requirement to prevent capturing it in the first place.
D is incorrect: Data residency policies determine where recordings are stored, but they do not obtain caller consent or prevent the card number from being recorded.
Q03 - Question
A voice agent collects an order number and a reason for the call, then escalates to a representative in Dynamics 365 Contact Center. The representative must receive the collected values when the call arrives. Which two configurations support this requirement? Each correct answer presents part of the solution.
Domain: Configure agents and AI capabilities (10-15%) Type: Multiple choice
- A. Use a Transfer conversation node to pass user-defined variables.
- B. Configure X-Headers to pass an unlimited number of context fields.
- C. Pass the collected values as short key-value context in SIP UUI.
- D. Enable external grammars to capture and transfer variables automatically.
A and C are correct.
Explanation: Transfer conversation nodes pass user-defined variables to the Contact Center, and SIP UUI supports short key-value context during call transfer.
B is incorrect: X-Headers support up to five fields; they cannot be configured to pass an unlimited number of context fields.
D is incorrect: External grammars control speech input handling, not the transfer of variables to a representative.
Q04 - Question
You need to configure an automated solution to answer inbound calls, greet callers, resolve simple requests, and transfer calls to a representative with the collected context. What should you configure?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Configure a chat agent and prompt the caller to switch channels.
- B. Build a voice agent in Copilot Studio and connect it through a voice workstream.
- C. Record inbound calls and instruct representatives to replay the recording.
- D. Route the inbound call directly to a representative queue.
B is correct.
Explanation: The evidence states that voice agents are built in Copilot Studio and connected through voice workstreams to greet and route callers, verify details, resolve requests, and transfer calls with collected context.
A is incorrect: A chat agent does not natively answer inbound phone calls or satisfy voice workstream requirements.
C is incorrect: Recording a call for later replay does not resolve simple requests automatically or transfer active context to a representative.
D is incorrect: Routing directly to a queue skips the automated greeting, verification, and resolution steps that a voice agent provides.
Q05 - Question
You configure a voice agent that uses a single phone number. Callers sometimes start speaking in English and then switch to a different supported language during the call. The voice agent must dynamically switch languages mid-conversation without restarting the call. What should you configure?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Multiple phone numbers with channel-specific languages.
- B. Escalation routing by using the Conversation.CustomerLanguage context variable.
- C. Generative orchestration for the voice agent.
- D. Classic orchestration and add trigger phrases for each language.
C is correct.
Explanation: Generative orchestration can support dynamic mid-conversation language switching on a single phone number.
A is incorrect: Configuring multiple phone numbers with channel-specific languages forces the caller to use a specific number per language, preventing mid-call switching on a single number.
B is incorrect: The Conversation.CustomerLanguage variable is used to route escalations to a human representative, not to enable the voice agent to switch languages mid-conversation.
D is incorrect: Classic orchestration relies on deterministic topics and does not dynamically interpret and switch languages mid-conversation.
Q06 - Question
A contact center plans to turn on autonomous AI agents in Dynamics 365 Contact Center. The organization stores data in one region and the agent capabilities process data in another. Which two prerequisites must the administrator address before the AI agents can be enabled? (Choose two.)
Domain: Configure agents and AI capabilities (10-15%) Type: Multiple choice
- A. Provision pay-as-you-go or subscription capacity for the AI agents.
- B. Replace side-pane Copilot with immersive Copilot for every representative.
- C. Grant consent for cross-region data movement.
- D. Turn off Copilot conversation summaries until the agents are live.
A and C are correct.
Explanation: The evidence supports A and C because AI agents require pay-as-you-go or subscription capacity, cross-region consent, and agent-specific roles. Enable capabilities incrementally and monitor adoption and quality.
B is incorrect: Replacing side-pane Copilot with immersive Copilot controls how representatives view assistance during interactions, not a condition for enabling autonomous agents.
D is incorrect: Copilot conversation summaries provide representative assistance and do not need to be turned off before enabling agents.
Q07 - Question
A retailer configures the Customer Assist Agent for inbound voice. The business wants the voice experience to use real-time speech models rather than a fixed, deterministic call flow. Which voice option should the administrator select?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Classic voice
- B. GenAI voice
- C. Premium GenAI voice
- D. SMS proactive engagement
C is correct.
Explanation: The evidence supports C because Premium GenAI voice uses real-time speech models, which meets the requirement to replace a fixed, deterministic call flow.
A is incorrect: Classic voice is deterministic, meaning the conversation follows a predefined flow rather than using real-time speech models.
B is incorrect: GenAI voice supports flexible orchestration, but Premium GenAI voice specifically uses real-time speech models.
D is incorrect: SMS proactive engagement is an outbound messaging capability and does not change how inbound voice is processed.
Q08 - Question
A quality manager wants the Quality Assurance Agent to evaluate conversations across several queues and channels, and wants the same measures reused across multiple evaluations. What should the manager do first?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Build a reusable quality library of indicators and guardrails, then create evaluation plans that reference it.
- B. Create a separate evaluation plan for each representative and define measures inside each plan.
- C. Create route-to-queue rules that score each conversation as it is routed.
- D. Enable Copilot conversation summaries and use them as the quality score.
A is correct.
Explanation: The evidence supports A because administrators must build a reusable quality library of indicators and guardrails before configuring evaluation plans. The evaluation plans then define the queue, workstream, and channel scope.
B is incorrect: Defining measures inside each evaluation plan prevents the reuse of indicators and guardrails across queues and channels.
C is incorrect: Route-to-queue rules manage how conversations are distributed and are not used to score or evaluate quality.
D is incorrect: Copilot conversation summaries provide interaction context and drafting assistance; they are not a quality evaluation scoring mechanism.
Q09 - Question
An administrator uses the Service Operations Agent to configure a second orchestration playbook for voice. The new playbook uses triggers and conditions that overlap an existing active playbook. What should the administrator expect when trying to activate the new playbook?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Both playbooks activate and run one after the other for each call.
- B. The newest playbook automatically replaces the existing active playbook.
- C. The two playbooks are merged into a single playbook definition.
- D. Validation blocks the activation until the overlap and conflict are resolved.
D is correct.
Explanation: The evidence supports D because validation explicitly blocks incomplete, duplicate, conflicting, or overlapping active playbooks. The Service Operations Agent requires overlap resolution before applying actions.
A is incorrect: Overlapping active playbooks are blocked by validation, so they do not activate and run sequentially.
B is incorrect: Activation does not automatically replace existing active playbooks; validation reports the conflict instead.
C is incorrect: Playbooks are not merged; validation blocks the conflicting activation to ensure human confirmation before actions apply.
Q10 - Question
Representatives must view order details from an external system inside Copilot. The order data is sensitive and the system requires controlled access. Which approach should the administrator plan for?
Domain: Configure agents and AI capabilities (10-15%) Type: Single choice
- A. Add the external system as a trusted web domain knowledge source for Copilot.
- B. Use a custom connector plugin with agent authentication, and confirm Copilot Studio licensing and testing before rollout.
- C. Add custom instructions to the Dynamics 365 knowledge source so Copilot retrieves the order data.
- D. Configure Quality Assurance guardrails to retrieve the order data during the conversation.
B is correct.
Explanation: The evidence supports B because custom connector plugins extend Copilot with external data, use agent authentication for sensitive external data, and require testing and Copilot Studio licensing.
A is incorrect: A trusted web domain expands the knowledge Copilot searches, but it does not securely retrieve sensitive data requiring agent authentication.
C is incorrect: Custom instructions apply only to Dynamics 365 knowledge and do not connect to external order systems.
D is incorrect: Quality Assurance guardrails identify violations during evaluation and trigger live nudges; they do not retrieve external data for Copilot.