Your invoice‑reading flow just got a bill: AI Builder credits are gone

Scenario: AP team, 4k invoices a month, and a quiet deadline

Picture a small finance/IT team with a Power Automate flow:

• Vendor emails → shared mailbox
• Flow triggers on new email
• Attachment sent to AI Builder – extract information from invoices (prebuilt model)
• Output written to Dataverse or a SharePoint list

They built it on seeded AI Builder credits that came with Power Apps or Power Automate licences. Until now, they never had to think about capacity.

From 1 November 2026, those seeded AI Builder credits stop. There is no automatic transition to Copilot Credits. If they do nothing:

• The flow still triggers
• The AI Builder invoice action fails once credits are exhausted or no longer available
• Downstream actions don’t run, or run with partial data depending on error handling

The rest of this article uses this AP scenario to show:

• What actually changes in Power Automate
• How to see if your flows are at risk
• Concrete options: pay for AI Builder, move to Copilot Studio, or redesign

What exactly is changing with AI Builder credits

Microsoft has been moving AI workloads from AI Builder credits to Copilot Credits. The key points that matter for your flows:

• Seeded AI Builder credits end 1 November 2026

  • These are the bundled credits that came with some Power Apps/Power Automate licences
  • After that date, only paid AI Builder capacity continues to work for AI Builder models

• No automatic migration to Copilot Credits

  • AI Builder and Copilot Credits are separate meters
  • An AI Builder invoice-processing action does not start using Copilot Credits just because you have a Copilot licence

• AI Builder document processing still exists

  • Prebuilt models (Invoices, Receipts, Identity documents, etc.) and custom form processing models continue
  • They simply require available AI Builder capacity to execute

For our AP flow, this means the invoice action keeps its connector and configuration, but:

• If you have paid AI Builder capacity: it keeps working, consuming that capacity
• If you only had seeded credits: execution eventually fails once any remaining credits are gone

How to see if your flows are about to break

You need to know two things:

1. Which flows use AI Builder?
2. What capacity do you actually have?

1. Find flows using AI Builder

In the AP scenario, the critical action is usually:

• Action name: Extract information from invoices (or another AI Builder model)
• Connector: AI Builder

To find all such flows:

• Go to Power Automate portal
• Check:
  • My flows → filter by environment
  • Team / Solution flows in the relevant environments
• Open each flow and look for actions from the AI Builder connector, such as:
  • Extract information from invoices
  • Extract information from documents
  • Classify text / Predict (custom AI Builder models)

If you have many flows, you can use the Power Platform Admin center:

• Navigate to Environments → [your environment] → Resources → Flows
• Review flows and search within definitions for AI Builder actions to find AI Builder usage

Any flow that calls AI Builder for document processing is at risk once seeded credits stop.

2. Check AI Builder capacity

In the Power Platform Admin center:

1. Go to Capacity
2. Look for AI Builder (sometimes under AI capacity or similar grouping)
3. Check:
   • Total capacity assigned to your tenant
   • Consumption trends

Key distinctions:

• Paid AI Builder capacity

  • Purchased as an add-on and assigned to environments
  • Continues to work after 1 November 2026 until you hit the purchased limit

• Seeded credits

  • Bundled and expiring; after the cutoff, they no longer apply to usage

If you only see seeded credits and no paid add-on, your invoice flows are running on borrowed time.

What happens to flows when credits are gone

When your AI Builder capacity is exhausted or no longer available:

• The flow still triggers normally
• The AI Builder action attempts to run and then fails
• Downstream actions behave according to your flow design:
  • If you didn’t handle errors: the run is marked Failed, and nothing after the AI step executes
  • If you used Configure run after with has failed: you may log the error or send a notification, but you won’t get invoice data

In the AP scenario, the symptoms you’ll see:

• Finance complains that no new invoices appear in the SharePoint list / Dataverse table
• Flow runs show red exclamation on the AI Builder action
• Error messages reference quota, capacity, or credits for AI Builder

This is not a transient service issue; it’s a capacity/licensing problem.

Option 1 – Stay with AI Builder and pay for it

If your invoice automation is business‑critical and already working well, the most conservative option is to:

1. Buy AI Builder capacity

   • Work with your licensing owner to purchase AI Builder add‑on capacity
   • Assign it to the environment where the AP flows run

2. Confirm the environment assignment

   • In Power Platform Admin center → Capacity → AI Builder
   • Ensure the AP environment shows non‑zero capacity

3. Add basic error handling to flows

   • For the AI Builder action, use Configure run after on the next step
   • Add a branch for has failed that:
     • Logs the error (e.g. to a Dataverse table or SharePoint list)
     • Notifies the AP team with the invoice ID and error message

Pros:

• Minimal change to your flows
• No retraining or re‑wiring of document processing

Cons:

• You now have a real cost per unit of AI usage
• You must monitor capacity and adjust as volume grows

For many teams, this is the right call if invoice automation is stable and they don’t want to redesign today.

Option 2 – Move to Copilot Studio for document processing

Microsoft is steering new AI workloads towards Copilot Studio. For document scenarios, the relevant capabilities are:

• Prompt‑based document understanding using Copilot
• Integration with Power Automate via actions that call a Copilot

This is not a drop‑in replacement for AI Builder invoice models, but for some scenarios you can:

1. Build a Copilot in Copilot Studio that:

   • Takes a document (or text extracted from the document)
   • Uses prompts to extract key fields

2. Expose that logic as an action or API

3. Call it from Power Automate instead of the AI Builder invoice action

Important constraints:

• Copilot Studio uses Copilot Credits, not AI Builder credits
• Prompt‑based extraction is more flexible but less structured than AI Builder’s trained invoice model
• You may need additional logic to validate and normalize outputs (dates, amounts, tax codes)

In the AP scenario, this path makes sense if:

• You already have Copilot licences and credits
• Your invoices are varied and AI Builder’s rigid schema has been a pain

If you take this route, plan a parallel run:

• Keep the AI Builder flow as reference while credits still exist
• Compare results from the Copilot‑based extraction for a subset of invoices
• Only cut over once you’re comfortable with accuracy and coverage

Option 3 – Reduce or remove AI Builder usage

Not every AP process needs full AI invoice parsing. If budget is tight, you can reduce your dependence on AI Builder or remove it entirely.

A. Use simpler extraction where possible

For invoices that follow strict templates, you may be able to:

• Require vendors to send structured files (e.g. CSV, XML, EDI, or standard e‑invoice formats)
• Use Power Automate + Power Query / basic parsing to extract fields without AI

This shifts work from AI Builder to deterministic parsing.

B. Use AI Builder only for exceptions

Instead of running AI Builder on every invoice:

1. Identify invoices that are hard to parse deterministically (scanned PDFs, non‑standard layouts)
2. Run AI Builder only on those exceptions

This reduces AI Builder consumption and therefore required capacity.

C. Fall back to manual entry with automation around it

For some teams, the trade‑off is:

• Use Power Automate to route and track invoices
• Let humans key the data into the system

You still gain:

• Automated capture of emails and attachments
• Consistent logging and approvals

But you avoid AI capacity costs entirely.

Hardening your flows before the cutoff

Regardless of which option you choose, harden your flows so they fail predictably instead of silently breaking.

1. Add explicit error handling around AI Builder actions

For each flow that uses AI Builder:

• Add a parallel branch or subsequent step with Configure run after set to:
  • is successful → normal processing
  • has failed → error handling path

In the AP flow, the error path should:

• Write a record to a log table (invoice ID, vendor, error message, timestamp)
• Send a notification to the AP mailbox or Teams channel

2. Monitor capacity and usage regularly

Set up a simple admin flow (in an admin environment) that runs daily and:

• Calls the Power Platform admin APIs or uses the admin connectors to read AI Builder capacity
• Sends a summary to the platform owner or IT ops team

If you don’t want to build a flow, at least schedule a monthly manual review in the Power Platform Admin center.

3. Document which invoices depend on AI

For the AP scenario, create a short note that covers:

• Which suppliers send structured vs unstructured invoices
• Which flows use AI Builder and in which environments
• What the fallback is if AI capacity is unavailable

This makes it easier to justify budget or redesign later.

Practical checklist for this week

Given the 1 November 2026 cutoff, here’s a concrete checklist you can run through now:

1. Inventory flows

   • List all flows using the AI Builder connector, especially invoice/document actions

2. Check capacity

   • In Power Platform Admin center, confirm whether you have paid AI Builder capacity assigned

3. Decide on a path per flow

   • Business‑critical and working well → plan for paid AI Builder capacity
   • Experimental or low‑value → consider retiring or redesigning without AI
   • Candidates for Copilot → schedule a prototype in Copilot Studio

4. Add error handling

   • Ensure every AI Builder action has a failure branch that logs and alerts

5. Communicate with owners

   • Tell finance / operations that invoice automation depends on paid AI capacity from now on

One concrete takeaway

Treat “Your invoice‑reading flow just got a bill: AI Builder credits are gone” as a licensing change you must design for, not a temporary outage: this week, identify every flow that calls AI Builder for document processing, decide which ones deserve paid capacity or a Copilot‑based redesign, and add error handling so the first sign of trouble is a planned alert—not a stack of unpaid invoices.