Power Automate + AI: Build Smart Workflows Guide

Operations managers waste 15-20 hours per week chasing approval emails, manually routing documents, and reconciling data between systems. Power Automate with AI Builder can cut that time by 70% by handling approvals, extracting data from documents, and routing work intelligently. This guide walks you through building your first AI-powered approval workflow that learns from your process patterns.

What You'll Learn

Create a cloud flow with approval actions that route based on conditions (amount, department, urgency)
Use AI Builder to extract invoice data from PDFs or photos automatically
Add adaptive cards so approvers can approve directly from Teams or email without opening another app
Build approval history tracking in Dataverse so you have full audit trails
Set up error handling and escalation rules for approvals stuck more than 48 hours
Measure time savings with Power Automate analytics dashboards

Prerequisites

Microsoft 365 account with Power Automate license (included in E3/E5 or standalone at $15/user/month)
Access to create flows in your Power Platform environment (check with IT if you're blocked by DLP policies)
Sample documents or emails that represent your current manual approval process
15 minutes to build and test your first flow

Step 1

Map your current approval bottleneck

Open a spreadsheet and document one approval process end-to-end: trigger (email arrives, form submitted), who approves, what data they need to decide, where approved items go next, and how long each step currently takes. Be specific about conditions—does anything over $5,000 need a second approver? Do certain departments bypass steps? This 10-minute exercise saves hours of rework because Power Automate mirrors your actual business logic, not an idealized process.

💡 Tip: Interview the people who actually do approvals today. Their workarounds reveal the conditions you need to automate.

Step 2

Create a cloud flow with the Approvals connector

Log into make.powerautomate.com, click Create > Automated cloud flow, and choose a trigger that matches your process—common ones are "When a new email arrives" (Outlook), "When an item is created" (SharePoint), or "When a new response is submitted" (Forms). Add the "Start and wait for an approval" action from the Approvals connector. Choose "Approve/Reject - First to respond" if one approver is enough, or "Approve/Reject - Everyone must approve" if you need consensus. This action creates the approval request and pauses the flow until someone responds.

💡 Tip: Use dynamic content from your trigger (like email subject or form fields) to populate the approval title and details—approvers need context to decide fast.

Step 3

Add AI Builder to extract data from documents

If your approvals involve invoices, receipts, or forms, add an AI Builder action before the approval step. Choose "Extract information from invoices" (pre-built model) or "Extract information from forms" (custom model you train with 5 sample documents). AI Builder pulls vendor name, amount, date, line items into variables you can use in the approval request or route to Dataverse. This eliminates the 5-10 minutes someone spends manually typing invoice data into your system.

⚠ Watch out: Pre-built invoice models work for standard invoices but may miss custom fields. Plan to train a custom model if your documents have unique layouts.

Step 4

Build conditional routing based on business rules

After the AI Builder step, add a Condition action to check extracted values—for example, "If Amount is greater than 5000" route to a senior manager approval, otherwise route to a team lead. You can nest conditions (if amount > 5000 AND department equals 'IT', send to CIO). Each branch can have its own approval action with different approvers. This mirrors your manual escalation rules but executes instantly instead of waiting for someone to read the amount and forward the email.

💡 Tip: Start with 2-3 conditions max for your first flow. You can add complexity after you validate the basic logic works.

Step 5

Design the approval request with adaptive cards

In the "Start and wait for an approval" action, expand the Advanced options and toggle "Enable adaptive card customization." Use the Adaptive Cards Designer (adaptivecards.io/designer) to build a card that shows key data (amount, vendor, date), attaches the original document, and includes Approve/Reject buttons. Approvers see this card in Teams, Outlook, and the Power Automate mobile app—they click a button and add a comment without opening a separate system. This cuts approval time from 2 days (waiting for someone to check email) to 2 hours.

💡 Tip: Include a "View full document" link in the card that opens the PDF or photo in SharePoint—approvers often need to see line items before deciding.

Step 6

Store approval results in Dataverse

After the approval completes, add a "Add a new row" action for Dataverse and create a new record in a custom table (you'll create this in Step 7). Map the outcome (Approved/Rejected), approver name, timestamp, comments, and the original document link into table columns. Dataverse gives you a permanent audit trail, enables reporting (average approval time, rejection reasons), and feeds dashboards so leadership sees bottlenecks. This replaces digging through email archives to find who approved what.

⚠ Watch out: Dataverse requires a Power Apps or Power Automate per-user license—it's not included in basic Microsoft 365 plans. Budget $20/user/month if you don't have it.

Step 7

Create a Dataverse table for approval history

In make.powerapps.com, go to Dataverse > Tables > New table and name it "Approval Records." Add columns: Approval Status (choice: Approved, Rejected, Pending), Approver (lookup to Users), Amount (currency), Request Date (date/time), Document Link (text), Comments (multiline text). Enable auditing on the table so Dataverse logs every change. Save and publish. This table becomes your single source of truth for all approvals, replacing scattered email folders and shared drives.

💡 Tip: Add a "Days to Approve" calculated column (Request Date minus Approval Date) so you can measure process speed improvements over time.

Step 8

Add escalation for stuck approvals

Create a second scheduled cloud flow that runs daily, queries Dataverse for approval records where Status = Pending and Request Date is more than 48 hours ago, then sends a reminder email to the approver (use "Send an email" action) and notifies their manager if still pending after 72 hours. Include a direct link to the approval in the email. This automates the nagging that operations managers currently do manually, freeing up 3-5 hours per week.

💡 Tip: Use the "Get rows" Dataverse action with a filter query: Status eq 'Pending' and Microsoft.Dynamics.CRM.OlderThanXHours(PropertyName='createdon',PropertyValue=48)

Step 9

Test the flow with real data

Trigger your flow manually with a test document or email. Watch it run step-by-step in the flow run history (you can see inputs/outputs for every action). Verify the AI Builder extraction is accurate, the approval shows up in Teams or Outlook with the correct data, and the Dataverse record is created when you approve or reject. Fix any errors (common issues: missing permissions, incorrect column names, wrong condition logic) and re-test. Do not skip this—flows that fail silently cause more chaos than manual processes.

⚠ Watch out: Test both approval and rejection paths. Many flows handle approvals fine but break on rejections because no one defined what happens next.

Step 10

Set up flow analytics and monitoring

In Power Automate, go to your flow > Analytics to see run success rate, average duration, and failure points. Create a Power BI report (or use the built-in Power Automate analytics dashboard) that shows approvals per day, average time to approve, top rejection reasons, and which approvers are slowest. Share this dashboard with your team monthly. Typical results after 30 days: 70% reduction in approval time (from 2 days to 4 hours), 40% fewer follow-up emails, and complete audit trail for compliance reviews.

💡 Tip: Set up flow failure alerts (in flow settings > Edit > Configure run after) to email you if a step fails, so you catch issues before users complain.

Summary

You've built an AI-powered approval workflow that extracts data from documents, routes intelligently based on your business rules, and stores a complete audit trail in Dataverse. Your approvers now respond in hours instead of days because they get clear requests in Teams or email with all the context they need. Operations managers save 15+ hours per week that used to go to chasing approvals and reconciling spreadsheets.

Next Steps

Replicate this flow for 2-3 more approval processes (PTO requests, expense reports, vendor onboarding) to compound time savings across your team
Enroll in PL-7002: Create and Manage Automated Processes to learn advanced patterns like parallel approvals, custom connectors for your ERP system, and RPA with Power Automate Desktop
Schedule a 30-minute Power Platform assessment with Scott Hay to identify which manual processes in your operation would benefit most from automation—typical ROI is 10:1 in year one
Set up governance policies (DLP, environment strategy, CoE Starter Kit) before approvals scale to dozens of flows—prevents Shadow IT chaos while enabling citizen developers

Need Power Platform Built Right the First Time?

Power Platform is flexible—which means it's easy to build poorly. I help SMBs architect Power Apps, Power Automate, and Power BI solutions that scale. Governance included. 90-day delivery, you own the IP.

Book Power Platform Consultation

Scott Hay Microsoft Certified Trainer & AI Solutions Architect Microsoft Certified Trainer (MCT) • Delivers 12 Microsoft Copilot courses (MS-4002 through MS-4023) plus Azure AI, Power BI • Azure AI Agents, Semantic Kernel, Power BI (PL-300), Power Platform certified • Former Microsoft and Amazon — 30+ years building production systems • Builds custom AI solutions for SMBs with 90-day delivery