AB-410 Practice Questions and Answers: 2026 Intelligent Applications Builder Prep

Last updated: August 2026. The practice questions below are original items written by Examinotion, modelled on the official AB-410 study guide and skills outline verified on 10 August 2026.

TL;DR These are 17 original AB-410 practice questions with worked answers, grouped by Microsoft's three official domains. They cover Dataverse, model-driven and canvas apps, cloud flows, AI Hub prompts and Copilot Studio agents. Use them to find gaps against the skills outline, not to predict real exam content or scores.

Microsoft released Exam AB-410: Building Intelligent Applications on 30 July 2026, and it leads to the Microsoft Certified: Intelligent Applications Builder Associate credential [1][2]. It is a Power Platform associate exam that adds AI capability to the classic functional consultant skill set, and it now occupies the slot vacated by PL-200, which retires on 31 August 2026 [3][4]. If you are preparing for it, structured practice against the published skills outline is the most efficient way to spend your revision time.

These questions are original practice items written by Examinotion to model the published AB-410 skills outline. They are not real Microsoft exam questions, and no site can legitimately give you those. Microsoft's exams are protected by a non-disclosure agreement, so any resource claiming to reproduce live questions is both breaking that agreement and, in our experience, routinely wrong. What good practice questions do is rehearse the way Microsoft frames a decision, so that on exam day the phrasing is familiar and the underlying concept is second nature. For the wider context on the exam itself, see our full AB-410 exam guide.

How to use these questions

Work through one domain at a time and answer before you read the explanation. The rationale under each question is the point of the exercise: it teaches the concept and, where it helps, explains why the wrong options are wrong. If a question exposes a gap, do not just memorise the answer. Open a free developer environment and build the thing, because AB-410 rewards hands-on fluency over recall, especially in the automation and AI Hub material.

Treat your score here as a diagnostic, not a prediction. There is no pass mark on a 17-question set, and no honest practice provider can promise you a result on the real exam. Weight your follow-up study by domain, and give the heaviest domain, business logic and automation, the most attention.

AB-410 exam format at a glance

Microsoft publishes some of the exam mechanics and deliberately withholds others. The confirmed items and the flagged unknowns are below.

Question formats you may meet. Microsoft does not pre-announce which formats a specific exam uses, but it publishes the general catalogue for associate role-based exams. Expect multiple choice with a single correct answer, and multiple response where you select a stated number of answers such as choose two. You may also meet case studies, where a detailed scenario is followed by several questions you can refer back to while inside that case, drag and drop, build list where you arrange items into the correct order or grouping, and hot area where you select regions of an image. The practice questions below use single and multiple response, because those transfer best to self-study, but do rehearse the interactive formats in Microsoft's public exam sandbox before the day.

Create a foundation for intelligent applications (25-30%)

This domain covers solution design and data modelling in Dataverse, the groundwork before any AI capability is added [1]. Here are five practice questions.

Question 1. You are designing a Dataverse solution that will move from a development environment to test and then to production. A downstream administrator must be able to add columns to your tables in production without your components being overwritten at the next deployment. Which approach should you recommend?

A. Ship an unmanaged solution to production so the administrator can edit it freely. B. Ship a managed solution to production and have the administrator make additive customisations in a separate unmanaged solution layered on top. C. Export the tables as a data package and re-import them in production. D. Give the administrator system administrator rights and ship no solution at all.

Correct answer: B

Why: Managed solutions are the correct artefact for downstream environments because they are not directly editable, which protects your components and keeps application lifecycle management clean. Additive customisations belong in a separate unmanaged layer. Shipping an unmanaged solution to production, option A, is an anti-pattern that makes the environment hard to maintain. A data package, option C, moves data, not schema and components. Elevating rights, option D, ignores the ALM problem entirely.

Question 2. A business wants a Dataverse column that automatically generates a short natural-language description of each record from other field values, using generative AI. Which column type should you configure?

A. A calculated column. B. A rollup column. C. A prompt column. D. A formula column.

Correct answer: C

Why: A prompt column is the AI-native Dataverse feature that populates a column from a generative prompt, which is exactly what this requirement describes. Calculated and formula columns evaluate deterministic expressions over other fields, and a rollup column aggregates values from related rows. None of those three invokes generative AI, so they cannot produce a natural-language summary. Prompt columns are one of the items PL-200-era candidates most often overlook.

Question 3. You are modelling a scenario where one Account can be linked to many Contacts, and each Contact belongs to exactly one Account. Which Dataverse relationship should you configure?

A. A many-to-many relationship between Account and Contact. B. A one-to-many relationship from Account to Contact. C. A one-to-many relationship from Contact to Account. D. No relationship; store the Account name as a text column on Contact.

Correct answer: B

Why: The requirement is a one-to-many relationship where Account is the "one" side and Contact is the "many" side, so the relationship is defined from Account to Contact with a lookup created on Contact. Option C reverses the cardinality. A many-to-many relationship, option A, would wrongly allow a Contact to belong to several Accounts. Storing a plain text column, option D, breaks referential integrity and loses the lookup, filtering and cascade behaviour that a real relationship provides.

Question 4. A team of makers needs an isolated space to build and test a new solution without any risk to production data, and you want Dataverse available so they can create tables. Which environment type should you recommend?

A. The default environment. B. A production environment shared with the live app. C. A sandbox environment with Dataverse provisioned. D. A developer environment tied to a single maker's licence, shared by the whole team.

Correct answer: C

Why: A sandbox environment with Dataverse is the standard isolated space for building and testing, and it can be reset or copied without touching production. The default environment, option A, is shared tenant-wide and is a poor place for project work. A production environment, option B, is the very thing you are trying to protect. A developer environment, option D, is scoped to an individual and is not intended as a shared team build space, so it does not fit a team requirement.

Question 5. A solution requires that sales representatives can see only the Opportunity records they own, while sales managers can see every Opportunity in their business unit. Which Dataverse mechanism should you configure to enforce this?

A. Form-level field security only. B. Security roles with record-level privileges scoped to User and Business Unit. C. A Power Automate flow that hides records the user does not own. D. A canvas app filter that only shows the current user's records.

Correct answer: B

Why: Record-level access in Dataverse is governed by security roles, where each privilege can be scoped to levels such as User, Business Unit or Organisation. Giving representatives User-scoped read and managers Business Unit-scoped read enforces the requirement at the data layer for every client. Field security, option A, controls individual columns, not whole-record visibility. A flow or an app filter, options C and D, only hide data in one surface and are trivially bypassed through other clients, so they are not a security control.

Create intelligent applications (25-30%)

This domain is the app-building core, split between model-driven and canvas apps, and it includes generative pages and creating a Copilot Studio agent from a canvas app [1]. Here are five practice questions.

Question 6. A department needs an application over an existing Dataverse model that gives consistent forms, views, dashboards and business process flows, and that automatically adapts its layout across devices with minimal design effort. Which app type should you build?

A. A canvas app, because it offers pixel-level layout control. B. A model-driven app, because it generates a data-first, responsive interface over Dataverse. C. A Power Pages site, because it is public-facing. D. A SharePoint list, because it needs no development.

Correct answer: B

Why: A model-driven app is data-first: it renders forms, views, charts, dashboards and business process flows over the Dataverse model and produces a responsive layout with little manual design work, which matches the requirement. A canvas app, option A, gives fine layout control but requires you to build the interface by hand, which is the opposite of minimal effort. Power Pages, option C, is for external websites. A SharePoint list, option D, is not a Power Platform app and lacks the required capabilities.

Question 7. You are building a canvas app and want to define a single reusable calculation that other formulas can call by name, improving readability and maintainability without a separate component library. Which Power Fx feature should you use?

A. A global variable set with Set(). B. A collection created with ClearCollect(). C. A named formula in App.Formulas. D. A context variable set with UpdateContext().

Correct answer: C

Why: A named formula, defined in the App object's Formulas property, gives a value a name and recalculates automatically when its inputs change, which is ideal for a reusable, readable calculation. Variables and collections, options A, B and D, store state that you must explicitly set and update, so they add moving parts rather than a declarative reusable definition. Named formulas and user-defined functions are recent Power Fx additions and appear explicitly in the skills outline, which signals they are testable.

Question 8. In a model-driven app, a business analyst wants to build a new page by describing the layout and data they need in plain English rather than configuring it component by component. Which capability supports this?

A. Creating a generative page by using natural language. B. Importing a Power BI report. C. Editing the sitemap in the app designer. D. Writing JavaScript in a web resource.

Correct answer: A

Why: Generative pages let a maker describe the page in natural language and have Copilot compose it, which is exactly the described workflow and is a named skill in the outline. Importing a Power BI report, option B, embeds analytics but does not build the page from a description. Editing the sitemap, option C, changes navigation, not page content. Writing JavaScript, option D, is a pro-code customisation that contradicts the plain-English, low-code requirement.

Question 9. A canvas app calls a connector that occasionally returns an error, and you must ensure the app shows the user a friendly message and continues running rather than crashing. Which approach is most appropriate?

A. Disable the connector so errors cannot occur. B. Use IfError() or the Trace() and error-handling functions to catch the failure and display a message. C. Wrap every control in a timer. D. Rely on the platform to suppress all errors automatically.

Correct answer: B

Why: Structured error handling with functions such as IfError lets you catch a failed operation, show a controlled message and keep the app responsive, which is the requirement and a named skill. Disabling the connector, option A, removes the feature rather than handling its failure. Timers, option C, do not catch errors. The platform does not silently suppress all errors, option D, so relying on that would leave the user facing raw failures.

Question 10. You have built a canvas app that answers common HR questions from a Dataverse knowledge table, and you now want to expose the same capability as a conversational agent employees can chat with in Teams. What should you do?

A. Publish the canvas app to Teams unchanged and tell users to type questions into a text input. B. Create a Copilot Studio agent from the canvas app and deploy it to Teams. C. Rebuild the logic as a cloud flow with a manual trigger. D. Convert the canvas app into a model-driven app.

Correct answer: B

Why: Creating a Copilot Studio agent from a canvas app is a named AB-410 skill and is the intended bridge from an app to a conversational agent that can be deployed to channels such as Teams. Publishing the app unchanged, option A, gives a form, not a conversational agent. A cloud flow, option C, automates a process but is not a chat experience. Converting to a model-driven app, option D, changes the app type without delivering the requested conversational interface.

Build business application logic and automation (40-45%)

This is the heaviest domain and where the AI content concentrates, spanning cloud flows, AI Hub prompts and models, and business and process logic [1]. Here are seven practice questions.

Question 11. A cloud flow must run whenever a new row is created in a Dataverse table, and it must run automatically without any user action. Which trigger should you use?

A. A manual trigger, "When Power Apps calls a flow". B. A recurrence trigger set to every five minutes. C. The Dataverse trigger "When a row is added, modified or deleted", configured for Added. D. An HTTP request trigger.

Correct answer: C

Why: The Dataverse "When a row is added, modified or deleted" trigger, configured to fire on the Added action, responds to the row-creation event directly and runs automatically, which matches the requirement precisely. A manual trigger, option A, requires a user to invoke it. A recurrence trigger, option B, polls on a schedule and would introduce delay and unnecessary runs rather than reacting to the event. An HTTP trigger, option D, waits for an external call, not a Dataverse event.

Question 12. A flow must send an expense claim to a manager, wait for approval or rejection, and then branch on the outcome. Which combination of building blocks should you use?

A. A "Start and wait for an approval" action followed by a Condition that checks the outcome. B. A Send an email action and a manual follow-up. C. A Switch on the flow's trigger type. D. A Do until loop with no exit condition.

Correct answer: A

Why: The "Start and wait for an approval" action creates the approval, pauses the flow until the approver responds, and returns an outcome you can branch on with a Condition, which is the standard managed-approval pattern. Sending an email, option B, does not capture a structured decision. A Switch on the trigger type, option C, is irrelevant to an approval outcome. A Do until loop with no exit, option D, would never terminate and is simply incorrect.

Question 13. In AI Hub you have built a prompt that classifies incoming support messages by urgency. You now need that classification to run inside a cloud flow that processes each new message. What should you do?

A. Rebuild the prompt logic manually with Condition actions in the flow. B. Consume the prompt in the cloud flow by adding the prompt action and passing the message text as input. C. Copy the prompt text into a Compose action. D. Export the prompt as a Power BI dataset.

Correct answer: B

Why: AI Hub prompts are designed to be consumed in cloud flows through a dedicated action, where you pass the required inputs, such as the message text, and receive the model's output. That is a named AB-410 skill. Rebuilding the logic with conditions, option A, discards the generative capability. A Compose action, option C, only stores static text and does not call the model. Exporting to Power BI, option D, has nothing to do with running a prompt in a flow.

Question 14. You are building an AI Hub prompt that must answer questions using your organisation's internal policy documents rather than general knowledge. Which prompt configuration meets this need?

A. Increase the model's temperature setting. B. Add the policy documents as knowledge grounding the prompt. C. Add more example inputs to the prompt. D. Switch the prompt to a larger model.

Correct answer: B

Why: Adding knowledge to a prompt grounds its responses in your supplied documents, so it answers from that content rather than from general training data, which is the requirement and a named skill. Changing temperature, option A, adjusts response randomness, not the source of facts. Adding example inputs, option C, shapes format or behaviour but does not supply the policy content. A larger model, option D, is still not grounded in your documents, so it can still answer from general knowledge.

Question 15. A requirement states that a Contact record's Full Name column must always equal the First Name and Last Name columns joined with a space, recalculated whenever either changes, without any code. Which Dataverse feature should you use?

A. A business process flow. B. A calculated or formula column that concatenates the two columns. C. A cloud flow triggered on modification. D. A business rule that shows an error if the names do not match.

Correct answer: B

Why: A calculated column, or a formula column, evaluates an expression over other columns and updates automatically when its inputs change, which is precisely the low-code, no-code way to derive Full Name. A business process flow, option A, guides a user through stages and does not compute a value. A cloud flow, option C, works but is heavier and less immediate than a column expression for a simple concatenation. A business rule showing an error, option D, validates rather than computes the value.

Question 16. You need a simple, declarative rule inside a model-driven form that hides the "Reason for discount" field unless the "Discount applied" field is set to Yes, and it must work in the form without writing JavaScript. Which feature should you configure?

A. A business rule with a condition and show/hide actions. B. A Power Fx formula in a canvas app. C. A rollup column. D. A cloud flow with a Dataverse trigger.

Correct answer: A

Why: Business rules apply declarative condition-and-action logic, including showing and hiding fields, directly on model-driven forms without code, which matches the requirement exactly. A canvas app formula, option B, does not apply to a model-driven form. A rollup column, option C, aggregates related records and has nothing to do with field visibility. A cloud flow, option D, runs server-side and cannot dynamically show or hide a field in the live form as the user edits it.

Question 17. A cloud flow must process each item in an array returned by a previous action, and for each item call a Dataverse action, but you are hitting throttling and want to reduce parallel runs. Which two steps address this? (Choose two.)

A. Use an Apply to each loop over the array. B. Reduce the Apply to each concurrency control setting to a lower degree of parallelism. C. Replace the loop with a single Compose action. D. Add a Terminate action at the top of the flow.

Correct answer: A and B

Why: Iterating an array requires an Apply to each loop, and its concurrency control setting lets you lower the degree of parallelism, which directly reduces the simultaneous calls that cause throttling. A Compose action, option C, cannot iterate and call an action per item. A Terminate action, option D, would stop the flow immediately, which does not process the items at all. This is a multiple-response item, so both correct steps must be selected.

Common AB-410 mistakes to avoid

• Treating AB-410 as PL-200 with a new number. Prompt columns, row summaries, generative pages, AI Hub prompts and models, and Copilot Studio agent creation are genuinely new measured skills. A functional consultant who studies nothing new walks into the heaviest domain underprepared.
• Confusing AI Hub with Microsoft Foundry. AI Hub is Power Platform's own prompts and models workspace and it is what AB-410 tests. Microsoft Foundry, the renamed Azure AI Foundry, belongs to Azure developer exams such as AI-103, and studying it for AB-410 wastes your time [1].
• Reading instead of building. Half of the AI Hub skills are about consuming prompts and models inside apps and flows [1]. That is muscle memory, not recall, and a free developer environment removes every excuse not to practise it.
• Guessing at the question count or price. Microsoft publishes neither for AB-410, so anyone quoting exact figures is inferring them. Anchor on the certification page and check Pearson VUE for your own regional price [2][5][6].
• Skipping accessibility and performance. "Design apps for accessibility, performance, responsiveness, and usability" is one bullet covering four bodies of guidance inside a 25 to 30 percent domain, and it is the most commonly skipped Power Apps topic [1].
• Choosing the wrong logic tool. Business rule, business process flow, calculated column or cloud flow: the exam asks you to pick, so practise matching each requirement to the lightest tool that satisfies it.

Frequently Asked Questions

How many questions are on the AB-410 exam?

Microsoft does not publish a question count for AB-410 or for any individual exam. Its general guidance says most certification exams contain between 40 and 60 questions, with variation by exam. Treat that as a benchmark for the associate tier, not a confirmed AB-410 figure, and discount any source quoting an exact number.

Is AB-410 hard?

AB-410 is an associate exam needing 700 out of 1000 in 120 minutes. Experienced Power Platform makers will find the app-building and data-modelling domains familiar, but the automation domain carries 40 to 45 percent of the marks and includes AI Hub prompts and models, which are new material for most PL-200-era candidates and where difficulty concentrates.

What is the AB-410 passing score?

The passing score is 700 out of 1000, confirmed in Microsoft's official study guide. This is a scaled score, not a percentage, so it does not mean answering 70 percent of questions correctly. Microsoft scales scores to account for varying question difficulty across exam forms, which makes raw practice percentages an unreliable readiness gauge.

Is PL-200 being replaced by AB-410?

In practical terms, yes. PL-200 retires on 31 August 2026 and can no longer be earned or renewed after that date, while AB-410 released on 30 July 2026 and now occupies the same Power Platform slot. Microsoft never uses the word "replaces", so treat the succession as a reasonable inference rather than a verbatim Microsoft statement.

How long is the AB-410 exam?

Microsoft states 120 minutes to complete the AB-410 assessment, confirmed on the certification page. Your total seat time at the test centre or online is longer than that, because it also includes time for instructions and the non-disclosure agreement before the timed portion begins. Plan your day around the seat time, not just the 120 minutes.

Do AB-410 practice questions reflect the real exam?

Good practice questions rehearse how Microsoft frames a decision and the concepts in the published skills outline, which is genuinely useful preparation. They do not reproduce live exam questions, and any resource claiming to is breaking Microsoft's non-disclosure agreement and is usually wrong. The Examinotion questions here are original items aligned to the outline, built to teach, not to leak.

Where to go next

Examinotion's dedicated AB-410 practice product is on the way, and until it lands the exam is not yet in our catalogue, so there is no AB-410 course page to link to yet. In the meantime, browse everything we currently cover at Examinotion's exam hub, and if you are still mapping out your certification path, our Microsoft AI certification roadmap shows how the associate exams fit together. Because AB-410 leans heavily on Copilot Studio agents, our guide to Microsoft Copilot agents for business is useful background, and the related AB-620 exam guide covers the agent-builder associate credential many AB-410 candidates consider alongside it. If you are weighing up sitting AB-410 while it is freshly out of beta, read our Microsoft beta exam guide, keep an eye on ways to save with our certification vouchers and discounts guide, and for a worked example of exam-focused revision, see how to pass the AB-730 exam.

Sources

1. Exam AB-410: Building Intelligent Applications study guide - Microsoft Learn, accessed 2026-08-10
2. Microsoft Certified: Intelligent Applications Builder Associate - Microsoft Learn, accessed 2026-08-10
3. Exam AB-100: Agentic AI Business Solutions Architect - Microsoft Learn, accessed 2026-08-10
4. Microsoft Certified: Power Platform Functional Consultant Associate - Microsoft Learn, accessed 2026-08-10
5. Exam duration and exam experience - Microsoft Learn, accessed 2026-08-10
6. Certification exam frequently asked questions - Microsoft Learn, accessed 2026-08-10