Microsoft AB-410 - Building Intelligent Applications

A calculated column is the correct option because it can evaluate a mathematical expression over numerical columns in the same Dataverse row without custom code. Dataverse stores the calculation definition and derives the exposed read-only value from the underlying columns rather than requiring the application to maintain an ordinary stored result. This satisfies the requirement for a dynamic, non-manually persisted value. A business rule would typically set a column value during a create or update event, resulting in stored data that can become stale if the rule does not run for every path. JavaScript operates only in supported client experiences and would not provide a consistent table-level value to all consumers. A dataflow transforms and loads data on a schedule or run; it materializes data rather than evaluating the field whenever accessed. The expression should account for blanks, numeric precision, and supported functions. The Study Guide explicitly lists “configure calculated, rollup, and formula columns.” Under the choices provided, D is the native declarative behavior that matches the calculation, access, storage, and low-code requirements.

Study Guide reference/topic: “Implement business and process logic — Configure calculated columns.” 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 existing AI Builder classification model should be invoked from the canvas app through Power Fx. The maker adds the model as an app data source or AI capability and calls its prediction operation from the relevant user action, passing the current image. The returned classification and confidence can then be stored in variables, displayed to the user, or written to Dataverse. A static formula cannot reproduce a trained image model’s inference. Refining the model with additional images is a training-lifecycle task and may improve quality later, but it does not consume the existing model during the current interaction. Exporting the containing solution transports components between environments; it does not execute a prediction. The app formula should validate that an image is present, handle service errors, and define what happens when confidence is below the accepted threshold. The Study Guide contains the exact objective “consume an AI model in apps.” Option A is the native runtime integration that classifies each user-submitted image during the active canvas-app session.

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

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

Separate development environments and separate solutions provide the required isolation between teams, while managed deployments provide production control. Each team can own a functionally scoped solution in its own development environment, preventing concurrent makers from overwriting the same unmanaged component layer. The solutions can be source-controlled, validated, and deployed independently through an approved pipeline. Importing them as managed solutions in production gives the organization a controlled, upgradeable package and supports removal through solution uninstall when dependencies permit. Keeping separate solutions in one shared development environment does not fully isolate changes to common components. Direct changes in production bypass validation and make the deployed state difficult to reproduce. Unmanaged solutions in production merge customizations into the active layer and are not the recommended release artifact. Dependencies and shared components must still be planned deliberately so that one solution does not silently require another team’s unmanaged work. This approach reflects the Study Guide requirement to “apply a Microsoft Power Platform solution and ALM strategy” and the standard separation of development, test, and production responsibilities.

Study Guide reference/topic: “Design Power Platform solutions — Apply a solution and ALM strategy.” 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 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.

Calling the prompt from a Power Automate cloud flow provides a reusable service-level entry point that can be invoked by different Power Platform experiences. The flow can accept parameters from each caller, pass those values into the prompt, and then route the generated result to an app, Dataverse, email, or another process. This addresses reuse more effectively than binding the prompt directly to one canvas app. An input variable is necessary for dynamic data, but by itself it does not establish cross-service orchestration; the flow exposes and manages those inputs for multiple callers. Returning JSON can standardize output handling but does not make the prompt reusable by itself. Direct canvas-app invocation is valid for an app-only requirement, not for the stated need to span Power Platform services. The flow should define explicit input types and validate required values so the case’s inconsistent-input problem is not repeated. The AB-410 Study Guide lists both “add inputs to a prompt” and “consume a prompt in cloud flows.” Option B provides the reusable execution layer while allowing dynamic values to be supplied at runtime.

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

The maker should define a prompt that references the support-case fields and create the destination column with the Prompt data type. The prompt contains the instructions for producing the case summary and declares the record inputs needed to ground that output. The prompt column then makes the generated summary available in the Dataverse/model-driven app context for the current row. A formula column performs deterministic Power Fx computation and cannot generate a natural-language summary from multiple fields. A business rule can set or validate values, but it does not invoke a generative model. The prompt should limit inputs to the approved columns, specify the required structure, and handle empty or contradictory case data so the generated content remains useful. Security on the source fields must still be respected, because an AI-generated summary must not become a route around restricted data. The AB-410 Study Guide lists “configure prompt columns” under Build data models and “add inputs to a prompt” under AI Hub. Options A and D supply the instruction definition and the Dataverse column behavior required for record-context generation.

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