Automate Purchase Order Emails with AI Using n8n

If you've ever managed purchase orders through email, you know the drill. Attachments pile up. Excel files need opening. Data needs copying. Numbers need checking. One wrong digit and you're chasing vendors for corrections.

What if your inbox could handle all of this automatically? What if every purchase order that landed in your email got validated, extracted, and processed without you lifting a finger?

That's exactly what this n8n workflow does. It watches your inbox, reads Excel attachments, extracts purchase order details using AI, runs validation checks, and sends automatic responses. All while you focus on work that actually needs a human brain. Let me show you how it works.

What You'll Need to Get Started

Before diving in, make sure you have these accounts and tools ready:

API Connections Required

• Microsoft Outlook account – For receiving and responding to purchase order emails
• OpenAI API key – Powers the AI classification and data extraction (GPT-4o-mini works great)
• n8n instance – Either cloud or self-hosted version

Key Components in This Workflow

• Outlook Trigger – Monitors incoming emails with attachments
• Text Classifier – Uses AI to identify genuine purchase order submissions
• If nodes – Handle conditional logic for file format and validation checks
• Extract from File – Parses Excel spreadsheets into readable data
• Code node – Converts Excel data to markdown tables
• Information Extractor – AI-powered structured data extraction
• Edit Fields nodes – Transform and validate extracted data
• Microsoft Outlook reply nodes – Send automated responses

Building Your Purchase Order Automation: Step by Step

Step 1: Set Up Email Monitoring

Everything starts with the Outlook Trigger node. Configure it to watch a dedicated inbox like "purchase-orders@yourcompany.com" and make sure to enable attachment downloads.

Set it to poll every hour. That keeps things running smoothly without hammering the API. The trigger captures essential email details including sender information, subject line, body content, and most importantly, any attached files.

But here's the smart part. Not every email to that inbox is actually a purchase order. Someone might send a question about POs or reply to an existing thread. That's where the next step comes in.

Step 2: Filter Out the Noise with AI Classification

The Text Classifier node uses OpenAI to analyze the email content and determine intent. Is this person actually submitting a purchase order or just asking about one?

Feed it the sender's name, email address, subject line, and message body. The AI categorizes the email as either "is_purchase_order" or "other." Simple, effective, and it saves you from processing irrelevant messages.

Valid purchase order submissions move forward. Everything else goes to a "Do Nothing" node and stops there. No wasted processing power, no false positives.

image_1.png

Step 3: Convert Excel Files into AI-Readable Format

Here's a problem most people don't think about: Large language models cannot directly read Excel files. They need text. So we need a conversion step.

First, check if the attachment is actually an XLSX file. If not, the workflow sends an automatic rejection reply asking for the correct format.

When you do have a valid Excel file, the Extract from File node pulls every row and cell from the spreadsheet. Then comes the clever part. A JavaScript code node takes those rows and converts them into a markdown table.

Why markdown? Because AI models read it beautifully. Headers become clear. Rows stay organized. The structure remains intact. This conversion bridges the gap between how humans send purchase orders and how AI can understand them.

Step 4: Let AI Extract the Important Details

Now the Information Extractor node takes over. You give it a JSON schema that defines exactly what data you need:

• Purchase order number and date
• Total amount
• Vendor name, address, and contact information
• Delivery details and shipping method
• Line items with descriptions, part numbers, quantities, and prices

image_2.png

The AI reads through the markdown table and extracts every field into structured JSON output. No manual copying and pasting. No transcription errors. Just clean, organized data ready to use.

There's one quirk to handle though. Excel stores dates as serial numbers counting days since January 1, 1900. A separate Edit Fields node converts these weird numbers into proper JavaScript date objects.

Step 5: Run Smart Validation Checks

Having extracted data is great. Having correct data is better. The Run Checks node performs four critical validations:

• Does a purchase order number exist?
• Is the PO date valid and not in the future?
• Are there actual line items listed?

That last one catches surprisingly common errors. Someone types $1,250 but the line items only add up to $1,205. The workflow catches it immediately.

Step 6: Send Automatic Responses

Based on the validation results, the workflow sends one of two replies.

If everything checks out, the sender gets a friendly acceptance message confirming receipt of their purchase order. Quick, professional, and they know their submission went through.

image_3.png

If validation fails, they get a detailed rejection email listing exactly what went wrong. Missing PO number? They'll know. Math doesn't add up? The message tells them. This immediate feedback loop means vendors can fix issues and resubmit right away instead of waiting days for someone to manually review.

Why This Workflow Changes Everything

Manual purchase order processing eats time. Someone has to open every email, download the attachment, read through it, type details into a system, check calculations, and send a response. Multiply that by dozens or hundreds of POs per week and you're looking at serious hours.

This automation handles it all in seconds. Your team gets their time back. Vendors get instant feedback. Errors get caught immediately instead of after products ship. And you have structured data ready to feed into your ERP system or database without any additional work.

Real World Applications

This workflow pattern works for more than just purchase orders. Think about:

• Invoice processing and validation
• Customer order forms
• Expense report submissions
• Contract intake and review
• Any document-based approval process

The core pattern stays the same. Monitor email, classify intent, extract structured data with AI, validate, and respond. Swap out the schema and validation rules to fit your use case.