14.2 Consuming Prompts in Apps and Cloud Flows

A tested prompt in AI hub > Prompts does nothing until something calls it. The AB-410 exam tests three consumption surfaces — canvas apps, model-driven apps, and cloud flows — and each has its own mechanics.

Canvas apps: prompts as Power Fx functions

In a canvas app, a custom prompt behaves like a callable data source. Open the app in Power Apps Studio (make.powerapps.com), go to Data > + Add data, and select your prompt — it now appears in the app like any other connection. You invoke it with the .Predict method, passing a value for each input you defined in the prompt builder:

Set(result, 'Task identifier'.Predict(TextInput1.Text))

The call returns a record; for a text-output prompt the generated string is on the .text property, so a label displays it with result.text. A realistic maker scenario: a field-service app where a technician types job notes into a text input, a button's OnSelect runs the summarization prompt, and a label shows the customer-ready summary before it is saved to Dataverse.

Practical guidance for the exam:

• One argument per prompt input — if the prompt defines two inputs (for example Text and Tone), .Predict takes two arguments in order
• Handle the wait — model calls take seconds; disable the triggering button or show a spinner so users do not double-submit
• Plan for failure — the call can fail due to throttling or regional availability, so check whether the result is blank before writing it to a data source
• Prompts that use JSON output format return structured data you can dot into, rather than a single blob of text

Exam trap: custom prompts in canvas apps are available only in some regions and can be subject to usage limits or capacity throttling — "call the prompt synchronously in a tight loop over thousands of rows" is never the correct architecture.

Model-driven apps: prompt columns

In model-driven apps you do not call prompts from formulas; instead you add a prompt column to a Dataverse table. A prompt column binds a custom prompt to the table, and its value is generated automatically from other columns on the row (for example, a Summary column on the Case table that runs a prompt over the Description column). The generation details and configuration steps are covered in section 5.2 — for this chapter, remember that prompt columns are the model-driven consumption pattern and that they are a Dataverse column type, not a control you drag onto a form.

Cloud flows: the Run a prompt action

In Power Automate (make.powerautomate.com), prompts are invoked with the AI Builder action Run a prompt. Exam trap: the action was called Create text with GPT using a prompt before May 2025, and older question banks still use that name — both refer to the same action.

Building the flow:

1. Add the Run a prompt action from the AI Builder connector
2. In the Prompt dropdown, select your prompt — or choose New custom prompt to author one inline without leaving the designer
3. Every input defined on the prompt appears as a field on the action card; fill each with dynamic content from previous steps (for example, the body of a When a new email arrives trigger)
4. Use the action's Text output variable in downstream actions — post it to Teams, save it to a Dataverse row, or email it

When the prompt produces JSON output format, feed the Text output into a Parse JSON action with a schema generated from a sample response; downstream steps then reference typed fields instead of slicing strings. For error paths, add a parallel branch or a scope whose Configure run after settings include has failed and has timed out, and route failures to a notification or a retry-with-backoff pattern — model calls can throttle under load. For quality control, insert Start and wait for an approval of text after the prompt action so a human reviews the suggested text before it is used; the approver's accepted or edited text flows downstream through the approval outcome.

Licensing awareness

You do not need pricing math for the exam, but you do need the shape: prompts executed in Power Apps and Power Automate consume AI Builder credits first, and once those are exhausted they consume Copilot Credits; prompts used inside Copilot Studio consume Copilot Credits only. An environment with no AI Builder credits and no Copilot Credit capacity will fail prompt calls at runtime — a common root cause in troubleshooting scenarios.

Testing and deployment notes

Always test the prompt itself in the prompt builder before wiring it into an app or flow — the test pane lets you supply sample values for each input and inspect the raw response, which isolates prompt-design problems from formula or flow problems. Prompts are solution-aware: add the prompt, the app, and the flow to the same solution so the whole package exports and imports between environments together, with the consumers reconnecting to the same prompt after import. If a prompt works in the prompt builder but fails when called, check region availability, credit capacity, and whether the calling user shares access to the prompt — those three cover nearly every exam troubleshooting scenario.