What’s actually slowing this PC down?

Pick the symptom - the matching free tool is one click away.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Use Copilot to help build an app or cloud flow; use AI Builder models and prompts to make that app or flow perform AI tasks at runtime. The right setup depends on the task, where a person needs the result, licensing and credit availability, and whether the feature is available in your region and ready for production.

How do I use AI in Power Apps?

There are two distinct ways AI can appear in a Power Apps project: help for the maker creating the solution, and an AI capability that runs inside the finished solution.

Use Copilot to help create an app

Power Apps Copilot can take a natural-language description of a business need and generate an app and data model to get started. It is a starting point for configuration and review, not a guarantee that the generated tables, relationships, or app behavior fit your process without changes. Microsoft describes this capability in its Copilot in Power Apps overview. Some Copilot features are preview, with possible functional restrictions; check the feature’s release label, region, and prerequisites before relying on it in production.

Add AI to the app’s workflow

AI Builder models and prompts can provide capabilities such as extracting information, classifying text, recognizing text in images or PDFs, analyzing sentiment, translating content, and generating text. A prompt can also be called from Power Apps using Power Fx; see Microsoft’s guide to using a prompt in Power Apps. Choose this route when a person needs the AI result during an app interaction or data-entry task.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Common scenarios to assess include extracting fields from an invoice, sorting incoming requests into categories, or summarizing a submission for a reviewer. Treat these as possible workflows, not as a promise of accuracy: the result depends on the input, configuration, and task, and the cited Microsoft documentation does not establish a universal accuracy benchmark.

Can Power Automate use AI to process documents?

Yes. AI Builder capabilities can be used as steps in Power Automate cloud flows, including document processing and text recognition. A flow might receive a document, send it to an AI step to extract fields, then route the result for review or onward processing. Microsoft’s AI Builder in Power Automate overview describes supported flow capabilities. Whether a particular document task is supported depends on the selected model or feature.

Power Automate Copilot serves a different purpose: it can use natural-language instructions to suggest a cloud-flow design. It helps the maker construct the workflow; it is not itself the runtime document-processing model. See Microsoft’s Copilot in Power Automate overview.

Should the AI run in Power Apps or Power Automate?

Choose When it fits Typical placement
Power Apps A person needs to see, enter, or act on an AI result within an app interaction. For example, a prompt summarizes text as part of a user’s submission process.
Power Automate AI is one step in a cloud workflow that responds to an event or moves work between steps. For example, a flow processes a document and routes the extracted information.

The same business process may use both: an app can collect or display information while a flow handles background processing. Decide based on who needs the result and when, then confirm the required connectors, licenses, credits, region availability, and release status for each feature.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Should you use a prebuilt model, custom model, or prompt?

  • Prebuilt capability: Consider it for a common task, such as a supported document or text scenario. Check the exact feature’s supported inputs and prerequisites.
  • Custom model: Consider it when the task needs a tailored model, such as custom document processing, category classification, entity extraction, object detection, or prediction. You will need suitable example data and a way to evaluate outputs; the documentation does not establish a guaranteed accuracy level.
  • Prompt: Consider it for a task such as summarizing, classifying, translating, or generating text. Prompt use is subject to model, region, licensing, and usage constraints, and consumption can vary with model tier and token volume.

Microsoft’s AI Builder overview describes prebuilt and custom-model capabilities, while its prebuilt prompts guidance covers prompts and their prerequisites.

Do I need Dataverse to use AI Builder prompts?

For the prebuilt prompts described in Microsoft’s guidance, yes: the stated prerequisites include a Dataverse database, a supported region, a Power Apps or Dynamics 365 license, and an AI Builder add-on. Confirm the current prerequisites for the particular prompt and environment before designing around it; availability and licensing can change. Microsoft’s prebuilt prompts documentation is the relevant feature-specific reference.

How much do AI Builder credits cost?

Credits are capacity units, not a universal cash price per successful task. Microsoft distinguishes AI Builder credits from Copilot Credits, and the applicable capacity depends on the feature and available entitlement. The rates below are Microsoft’s published consumption figures in its licensing guidance, updated September 8, 2026; they are not a quote for a subscription or a prediction of task accuracy.

Capability and model tier Copilot Credits AI Builder credits
Prompt, basic LLM 0.1 per 1,000 tokens 1 per 1,000 input tokens and 3 per 1,000 output tokens
Prompt, standard LLM 1.5 per 1,000 tokens 20 per 1,000 input tokens and 60 per 1,000 output tokens
Prompt, premium LLM 10 per 1,000 tokens 140 per 1,000 input tokens and 560 per 1,000 output tokens
Document/content processing 8 per image 32 per image

The document/content-processing row carries a preview qualification in Microsoft’s rate table. Check that table for the exact capability and current conditions before estimating usage. Prompt totals vary by model tier and token volume; an image-based task’s consumption should likewise be checked against the capability definition rather than assumed to equal a whole-document price.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Microsoft’s credit-management guidance lists monthly seeded entitlements including 500 AI Builder credits for Power Apps Premium and 5,000 for Power Automate Premium, subject to tenant caps and other rules. These are not stable future entitlements: Microsoft says that beginning November 1, 2026, new or renewed Power Platform or Dynamics 365 licenses will no longer receive seeded AI Builder credits. Eligible existing credits continue through the applicable contract term; new customers must purchase Copilot Credits to run AI Builder features. Check contract timing and tenant allocation against Microsoft’s AI Builder licensing overview, credit-management guidance, and credit transition notice.

Training a model does not itself consume credits under Microsoft’s licensing guidance; running AI capabilities does. Credit consumption is separate from the license needed to run a flow: AI Builder actions are premium connectors, and applicable Power Automate licensing can also apply. See Microsoft’s Power Automate licensing FAQ.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

How to plan and validate an AI workflow

  1. Choose one bounded task. Define the input, the result you need, and what a person or downstream step will do with that result.
  2. Choose where it belongs. Put the AI interaction in Power Apps if a user needs the result in the app; put it in Power Automate if it is a step in a cloud workflow.
  3. Check the feature and prerequisites. Confirm whether the task uses a prebuilt capability, custom model, or prompt, then check region, Dataverse, license, connector, credit, and preview requirements that apply.
  4. Test with representative inputs. Inspect outputs using examples that reflect the real range of documents or text you expect. Decide how to handle missing fields, unclear results, and failures.
  5. Set review and exception rules. Determine which outputs need human approval and what the workflow should do when the model returns incomplete or unsuitable information.
  6. Estimate and monitor runtime usage. Use Microsoft’s current rate table and your environment’s available capacity. Account for model tier, token volume, images or documents processed, and flow licensing.

Microsoft’s published materials document capabilities and licensing, but do not provide a universal validation protocol, head-to-head accuracy benchmark, or ROI comparison. Evaluate the workflow against your own data and acceptance criteria before depending on its outputs.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.