Insights/Power Automate
Power Automate10 min readMarch 2026

Power Automate + AI Builder: Your First Intelligent Workflow

A step-by-step walkthrough for building your first AI-powered Power Automate flow — from trigger to action — using AI Builder models your organization already has access to.

01Pick the Right Use Case02Set Up AI Builder in Your Environment03Build Your AI Model (No Code Required)04Wire It Into a Power Automate Flow05Add Error Handling and Notifications06Test, Measure, and Expand

“We have Power Automate licenses but everything we've automated is just moving files between folders.”

Most Power Automate workflows are basic: move a file, send a notification, copy a row. They're useful but they're not intelligent. They can't read a document, understand what it says, or make decisions based on content.

AI Builder changes that. It adds AI capabilities — document reading, text analysis, image recognition, prediction — directly inside Power Automate. No coding. No external APIs. No data leaving your Microsoft tenant.

This guide walks you through building your first AI-powered workflow from scratch. We'll use invoice processing as our example (it's the most common use case), but the pattern applies to any document type: contracts, receipts, purchase orders, HR forms, or insurance claims.

By the end, you'll have a workflow that automatically extracts data from uploaded documents and routes it where it needs to go — no human data entry required.

01

Pick the Right Use Case

Most teams fail here. They pick something flashy ("build a chatbot") instead of something painful ("stop manually entering invoice data"). AI Builder works best on repetitive, structured tasks where humans are slow and error-prone.

What to do
  • Document processing — extracting fields from invoices, receipts, contracts, or forms
  • Email classification — routing support requests, flagging urgent items, categorizing feedback
  • Text analysis — summarizing long documents, extracting key entities, detecting sentiment
  • Object detection — identifying items in images for inventory, quality control, or compliance

Pro tip

Start with the task your team complains about most. If someone says "I spend two hours every week copying data from PDFs into a spreadsheet" — that's your first automation.

02

Set Up AI Builder in Your Environment

AI Builder is included in Power Automate Premium licenses and some Microsoft 365 E5 plans. Before you build anything, verify your licensing and permissions.

What to do
  • Go to make.powerautomate.com and sign in with your work account
  • Click 'AI Builder' in the left navigation — if you see it, you have access
  • Check that your environment has Dataverse provisioned (AI Builder requires it)
  • Verify you have the 'Environment Maker' role or higher in Power Platform admin center

Pro tip

If you don't see AI Builder, your admin needs to enable it. It's a 5-minute setting change in the Power Platform admin center — not a licensing purchase. Many organizations already have it enabled and don't know it.

03

Build Your AI Model (No Code Required)

AI Builder offers prebuilt models that work immediately and custom models you can train on your own data. For your first workflow, start with a prebuilt model — zero training required.

What to do
  • For invoices: use the 'Invoice processing' prebuilt model — it extracts vendor, amount, date, line items automatically
  • For business cards: use the 'Business card reader' to extract contact info into your CRM
  • For text: use the 'Sentiment analysis' or 'Category classification' prebuilt models
  • For custom needs: train a 'Document processing' custom model with 5+ sample documents

Pro tip

Prebuilt models handle 80% of common use cases. Only train a custom model if the prebuilt one doesn't extract the specific fields you need. Custom models need at least 5 sample documents and 15-30 minutes of labeling.

04

Wire It Into a Power Automate Flow

Now connect your AI model to an automated workflow. This is where the real value happens — the AI processes data automatically, without anyone clicking a button.

What to do
  • Create a new 'Automated cloud flow' in Power Automate
  • Set the trigger: 'When a file is created' in SharePoint (or OneDrive, or Outlook)
  • Add the AI Builder action: 'Extract information from invoices' (or your chosen model)
  • Map the extracted fields to your destination: Excel, SharePoint list, Dynamics 365, or email notification

Pro tip

The most common pattern: SharePoint folder → AI Builder extraction → Excel row or SharePoint list item. This replaces the "download PDF, open it, read it, type data into spreadsheet" cycle your team does dozens of times a month.

05

Add Error Handling and Notifications

Production workflows need guardrails. AI models aren't 100% accurate — especially on messy, handwritten, or unusual documents. Build in checkpoints so your team catches issues early.

What to do
  • Add a 'Condition' action to check the AI model's confidence score
  • Route high-confidence results (>85%) straight to your spreadsheet or system
  • Route low-confidence results to a Teams message or email for human review
  • Log every processed document to a SharePoint list for audit trail

Pro tip

Set the confidence threshold at 85% initially. Track how many documents get flagged for review over the first two weeks. If it's more than 20%, your source documents might need standardization — or you might need a custom-trained model.

06

Test, Measure, and Expand

Your first AI workflow is live. Now prove its value with numbers so you can justify expanding to more workflows across the organization.

What to do
  • Track time saved per document: manual processing time vs. automated processing time
  • Measure accuracy: spot-check 10% of AI-processed results for the first month
  • Calculate ROI: (hours saved × hourly cost) vs. Power Automate licensing cost
  • Document the workflow in a playbook so other teams can replicate the pattern

Pro tip

Most teams see 70-90% time reduction on document processing tasks. A workflow that saves 15 minutes per invoice × 200 invoices/month = 50 hours/month saved. That's your business case for the next automation.

The real unlock is the pattern

Your first AI workflow handles invoices. Your second handles contracts. Your third handles support tickets. The pattern is always the same: trigger → AI model → route by confidence → store results.

Once your team understands this pattern, they can build intelligent automations independently. That's the real value of this approach: not just one automated workflow, but a repeatable framework your organization can apply to dozens of processes.

The organizations that get the most value from AI aren't the ones with the fanciest models. They're the ones that systematically identify bottlenecks, build focused automations, measure the results, and expand.

Key takeaways

  • Start with your most painful manual process — not the most impressive one.
  • Use prebuilt AI models first. Custom training is only needed when prebuilt models miss your specific fields.
  • Always add confidence-based routing. High confidence goes straight through; low confidence gets human review.
  • Measure time saved per document to build the ROI case for your next automation.
  • Document the pattern. The goal isn't one workflow — it's teaching your team to build the next ten.
Need Help Building This?

We build the first workflow
so your team can build the rest.

Start with a $299 AI Readiness Audit to identify your best automation candidates, or go straight to a full Quickstart implementation.

See PricingMore Insights