A custom connector is required to integrate with the third-party AI service because the integration needs a defined authentication scheme and custom request handling. The connector wraps the service’s API as named operations, with host information, paths, methods, headers, parameters, request bodies, and response schemas. Power Automate or Power Apps can then call those operations through governed connections instead of reconstructing raw requests in every solution. Power Fx is an expression language and does not by itself define or secure an external API contract. Microsoft 365 Copilot chat is a user productivity experience, not a Dataverse integration mechanism. An environment variable can store deployment-specific configuration such as a base URL or identifier, but it neither authenticates nor sends the API request. The connector should be packaged in a solution, use the service’s supported API key or OAuth design, and comply with data loss prevention policies. This directly aligns with the Study Guide cloud-flow objective “evaluate and recommend connectors” and Microsoft’s custom connector architecture.

Study Guide reference/topic: “Create cloud flows — Evaluate and recommend connectors.” AB-410 Study Guide | Microsoft Learn technical reference

Continue until status is Complete — Do until; Process the fixed retrieved list — Apply to each

Do until is the correct loop when processing must continue until a runtime condition becomes true. The flow can recheck the status and repeat its actions until Status equals Complete, subject to configured count and timeout limits. Because the condition is reevaluated between iterations, it can respond to a status change that occurs while the flow is running. Apply to each is correct for the fixed array of records retrieved at the beginning. It enumerates each item in that array once; records added to the source later are not automatically inserted into the already obtained collection. Scope groups actions for error handling or organization but does not implement either iteration pattern. Concurrency settings affect parallelism and ordering within a loop but do not replace the loop itself. The Do until design should include a delay or appropriate polling interval and explicit termination safeguards to avoid excessive requests. Apply to each concurrency should be chosen according to ordering and connector limits. The Study Guide’s cloud-flow objectives include “implement flow control with conditions and loops.” The verified mapping matches a condition-driven loop and a collection-driven loop to their intended behaviors.

Study Guide reference/topic: “Create cloud flows — Implement flow control with loops.” AB-410 Study Guide | Microsoft Learn technical reference

The app should call the existing AI Builder model from its app logic. The model is added to the canvas app and invoked from Power Fx associated with a user event such as a button selection, passing the current image as input. The returned class, confidence, or other prediction values can immediately update controls or be stored with the business record. Retraining for every image is neither required nor operationally appropriate; training is a controlled model-lifecycle activity performed with a curated dataset. Reusing a stored prediction would classify the previous input rather than the newly submitted image. Exporting a solution transports the model and app components between environments but does not execute inference. Runtime code should check that an image exists, handle prediction failures, and apply a confidence threshold or human-review path where the business decision requires it. The AB-410 Study Guide includes the exact objective “consume an AI model in apps.” Option D describes that native runtime consumption and is the only action that generates a prediction during the user’s current interaction.

Study Guide reference/topic: “Create prompts and models in AI Hub — Consume an AI model in apps.” AB-410 Study Guide | Microsoft Learn technical reference

The Microsoft 365 administrator should change the user’s sign-in name in the Microsoft 365 or Microsoft Entra administration plane. Dataverse user records are synchronized from the tenant identity, so the authoritative user principal name must be updated at its source. The existing identity object remains the same, preserving its directory object identifier and the corresponding Dataverse system-user relationship. Consequently, record ownership, activity history, audit associations, and security-role assignments are not replaced simply because the sign-in name changes. Administrators should allow synchronization to complete and then verify the updated username and mailbox information in the Power Platform environment. Creating a new user would risk separating the historical record from the new identity and would require reassignment work. The scenario asks for a name change without losing application history, which is precisely why the existing directory account must be modified instead of replacing the Dataverse user. This aligns with the Study Guide’s security and identity collaboration expectations and its “configure security” objective for Dataverse-backed applications. The proposed Microsoft 365 administrator action therefore meets the goal.

Study Guide reference/topic: “Build data models — Configure security.” AB-410 Study Guide | Microsoft Learn technical reference

Dataverse table columns are the appropriate knowledge source for a prompt that must use Retrieval-Augmented Generation with current business information. Grounding the prompt with selected Dataverse data allows the model to incorporate governed product specifications at runtime instead of relying only on its pretrained knowledge. The maker can identify the relevant table and columns and then pass or retrieve the business context required by the prompt. A PNG image is an input artifact, not the structured knowledge source needed for specification retrieval. Plain text can provide fixed instructions or context but does not create a maintained enterprise data source. A PDF can be useful in other grounding experiences, but under the available prompt knowledge options in this scenario, the Dataverse table provides the direct, structured, updateable source. Access remains subject to environment and table security, so only approved data should be exposed to the prompt. The AB-410 Study Guide explicitly measures the ability to “add knowledge to a prompt.” Selecting Dataverse table columns implements that objective while minimizing duplication of product data.

Study Guide reference/topic: “Create prompts and models in AI Hub — Add knowledge to a prompt.” AB-410 Study Guide | Microsoft Learn technical reference

The app should pass the current user-entered values as prompt parameters and invoke the prompt directly from the button’s Power Fx logic. Parameters ensure that the generated summary is grounded in the text currently displayed in the app rather than stale or hard-coded data. Direct invocation from the button provides the immediate interactive behavior required by the scenario. A cloud flow can also consume a prompt, but it adds an orchestration layer that is u nnecessary when the requirement is limited to a synchronous canvas-app interaction. Publishing the prompt is part of making a completed component available, but it does not execute the prompt or provide the current input. Updating general app settings is unrelated. The button formula should validate blank content, handle errors, and place the returned summary in a variable or control only after a successful response. The prompt’s inputs and output should be typed and tested for expected maximum lengths. The Study Guide explicitly includes “add inputs to a prompt” and “consume a prompt in apps.” Options B and C meet those two runtime responsibilities.

Study Guide reference/topic: “Create prompts and models in AI Hub — Consume a prompt in apps.” AB-410 Study Guide | Microsoft Learn technical reference

A global choice provides one centrally maintained option set that multiple Dataverse columns can reference. This directly satisfies the requirement that Priority must be added to several tables and that a newly added option must become available automatically wherever the shared choice is used. Microsoft’s Dataverse documentation states that columns can refer to a “common set of global options.” Maintaining the values centrally prevents separate Priority columns from drifting out of synchronization.

A local choice is defined specifically for one column. Adding a value to one local choice would not update corresponding columns on other tables. Status reason is tied to a row’s active or inactive state and is unsuitable for general priority classification. A formula column calculates a value; it does not provide a centrally managed, user-selectable list.

Because users must select one value, the column should be a single-select Choice referencing the global Priority choice, not a multi-select Choices column.

Assigning the appropriate security roles is the complete action because model-driven app access and Dataverse data access are both role driven. The app must be shared with the role, and the role must grant the accounting team the table privileges required to read and create or update the financial records. Configuring app visibility privileges alone may make the app tile appear, but it does not establish the underlying table rights. Setting an app as default affects navigation or launch preference and does not grant access. Granting table-level permissions is only one part of the required configuration and, in Dataverse, those privileges are normally delivered through a security role; option C represents the combined mechanism expected by the question. The administrator should use least privilege, choose the correct access depth, and include privileges for dependent tables and processes used by the app. Existing visibility in the launcher is not proof of effective authorization. The AB-410 Study Guide includes both “configure access to model-driven apps” and “configure security.” Appropriate role assignment aligns the user, the shared app, and the table operations in one enforceable access model.

Study Guide reference/topic: “Create model-driven apps — Configure access to model-driven apps.” AB-410 Study Guide | Microsoft Learn technical reference

The solution should use a prebuilt AI capability where it satisfies the complaint-processing requirement and create the generative prompt from blank for the organization-specific inspection report. Using a prebuilt model minimizes custom model development and training effort, which directly reflects the stated constraint. A blank custom prompt gives the maker control over the inspection template, instructions, inputs, tone, and output format required by the business. Starting from a prompt template can accelerate common scenarios, but the case requires alignment to an existing custom inspection format; a blank prompt avoids carrying irrelevant template assumptions and permits deliberate grounding. Building a custom trained model would add data preparation, training, validation, and lifecycle overhead that the requirement seeks to minimize. The prompt should define required inputs explicitly and be tested with representative data to prevent the inconsistent output problem identified in the case. The Study Guide covers “build prompts from templates or from blank,” “add inputs to a prompt,” and consumption of AI capabilities in apps and flows. Options A and C provide the lowest-custom-development design that still supports the specific generative output.

Study Guide reference/topic: “Create prompts and models in AI Hub — Build prompts and consume AI capabilities.” AB-410 Study Guide | Microsoft Learn technical reference

A custom prompt is the native low-code mechanism for generating an inspection report from current app data while following the company’s existing template instructions. The maker can define inputs for the inspection fields, specify required sections, tone, order, and formatting, and invoke the prompt from the new app. This keeps the report-generation experience within Power Platform and minimizes custom AI development. A paginated report is well suited to deterministic, pixel-precise reporting, but it does not perform the requested generative composition and generally presents a separate reporting artifact. A Word template performs deterministic field substitution and is useful when no generative rewriting is required; it does not supply the AI-driven custom narrative described in the case. The prompt should be grounded only with approved inspection data, tested for missing inputs, and reviewed for output consistency before production use. The Study Guide covers “build prompts,” “add inputs to a prompt,” and “consume a prompt in apps.” Option C is therefore the component that directly implements native generative inspection-report creation.

Study Guide reference/topic: “Create prompts and models in AI Hub — Consume a prompt in apps.” AB-410 Study Guide | Microsoft Learn technical reference