Power Automate + AI Builder

Your team processes hundreds of invoices manually. Employees copy data from emails into spreadsheets. Approval chains stall because someone forgot to forward a document. Power Automate + AI Builder eliminates these bottlenecks by combining workflow automation with ready-to-use AI models — no machine learning expertise required.

This guide shows you how to build AI-powered automations that read documents, classify emails, analyze sentiment, and route decisions automatically across your organization.

How Power Automate and AI Builder work together

Power Automate is Microsoft’s automation platform — it connects triggers (an email arrives, a file is uploaded, a form is submitted) to actions (send a notification, update a database, create a task). AI Builder adds intelligence to those flows by providing pre-built and custom AI models that you can drop into any automation step.

AI Builder model types

Building an invoice processing flow

Let’s build a complete automation: invoices arrive via email, AI Builder extracts the data, and the flow creates records and routes approvals.

Step 1: Create the flow trigger

1. Go to make.powerautomate.com.
2. Click Create → Automated cloud flow.
3. Choose the trigger: When a new email arrives (V3) from the Outlook connector.
4. Add a filter: Has attachments = Yes and Subject contains “invoice”.

Step 2: Add AI Builder invoice processing

1. Add a new step and search for “AI Builder”.
2. Select Extract information from invoices.
3. Set the Invoice file to the email attachment content.
4. AI Builder returns structured data: vendor name, invoice number, amount due, due date, line items, and confidence scores.

Step 3: Add business logic

Condition: Amount due > $5,000
  Yes → Start an approval (Manager)
    Approved → Create row in SharePoint "Approved Invoices"
    Rejected → Send email to vendor with rejection reason
  No → Auto-approve → Create row in SharePoint

Always: Update Excel tracker with extracted data

Step 4: Error handling

AI Builder returns a confidence score for each extracted field. Build conditions around it:

• Confidence > 0.85 → auto-process the invoice.
• Confidence 0.60 – 0.85 → route to a human for review.
• Confidence < 0.60 → flag as unreadable and notify the sender.

Building a custom text classifier

Pre-built models cover common scenarios, but what if you need to classify support tickets into your own categories? AI Builder lets you train custom models with your data.

1. Go to AI Builder → Build → Text classification.
2. Define your categories (e.g., “billing”, “technical”, “shipping”, “returns”).
3. Upload or connect your training data — a minimum of 10 tagged examples per category, though 50+ produces better results.
4. Click Train. AI Builder trains and evaluates the model automatically.
5. Review the accuracy metrics and publish the model when satisfied.

Once published, use it in Power Automate just like any pre-built model:

Trigger: When a new item is created in SharePoint "Support Tickets"
Step 1: AI Builder → Classify text (your custom model)
  Input: Ticket description
Step 2: Condition on classification result
  "billing" → Assign to finance team
  "technical" → Assign to engineering team
  "shipping" → Assign to logistics team
  "returns" → Start return process flow

Using GPT in Power Automate

AI Builder includes a GPT-based action (powered by Azure OpenAI) that lets you add generative AI to any flow without managing API keys or deployments.

Summarize long emails automatically

Trigger: When an email arrives with body length > 500 characters
Step 1: AI Builder → Create text with GPT
  Prompt: "Summarize this email in 3 bullet points: {{email body}}"
Step 2: Post summary to Teams channel

Generate customer response drafts

Trigger: New support ticket created
Step 1: AI Builder → Create text with GPT
  Prompt: "Draft a professional response to this customer complaint.
   Be empathetic and offer a concrete next step: {{ticket description}}"
Step 2: Create draft email (not sent — human reviews first)
Step 3: Send Teams notification to support agent for review

Governance and monitoring

In enterprise environments, you need to control who uses AI Builder and track usage:

• AI Builder credits — each AI action consumes credits. Monitor usage in the Power Platform Admin Center.
• Data Loss Prevention (DLP) policies — restrict which connectors can be used together to prevent data leakage.
• Environment controls — isolate production flows from development using Power Platform environments.
• Flow analytics — track run history, failure rates, and processing times in the admin center.

Licensing

One invoice extraction costs approximately 1 credit. One GPT prompt costs 1-3 credits depending on token usage. Monitor your consumption to avoid surprises.

Next steps

1. Start a trial at make.powerautomate.com and explore the AI Builder templates.
2. Build an invoice flow — connect your Outlook and let AI Builder extract data from a real invoice.
3. Train a text classifier — use 50 support tickets to build a custom routing model.
4. Read the docs: learn.microsoft.com/ai-builder