Automate Email Replies with n8n and AI-Powered Summarization

What This Automation Does

This workflow reads incoming business emails, summarizes their content, drafts replies using AI plus company data, waits for human approval, then sends final answers automatically.

It solves long manual hours spent on reading and replying emails. It saves time by up to five hours daily and avoids errors from rushed reading or inconsistent answers.

Input is new emails from the inbox. Processing includes converting email to markdown, summarizing main points, and generating context-aware replies using AI with vector searches of company knowledge. Output is either draft emails for human review or final approved replies sent to customers.

Tools and Services Used

• Email Trigger (IMAP) node: Detects incoming emails.
• Markdown node: Converts email HTML into markdown.
• Email Summarization Chain node: Generates a short summary (max 100 words).
• Write Email agent node: Combines OpenAI GPT models and Qdrant vector search for reply drafts.
• Text Classifier node: Classifies human feedback as Approved or Declined.
• Send Email node: Sends final approved replies via SMTP.
• Gmail node: Sends drafts to team inbox for review.
• Qdrant vector database: Stores company data for AI context.

This mix of tools processes emails, summarizes content, drafts replies using AI and company information, and includes a step for team approval before sending answers.

Beginner Step-by-Step: How to Use This Workflow in n8n

Importing and Configuring

1. Open the n8n editor.
2. Use the Download button on this page to get the workflow file.
3. In n8n, click “Import from File” and select the downloaded workflow.
4. Add all required credentials such as IMAP, SMTP, OpenAI API key, Qdrant connection, and Gmail credentials.
5. Update any IDs, email addresses, folder names, or channels if necessary to match your setup.

Testing and Activating

6. Send a test email to the IMAP-triggered mailbox.
7. Check the workflow execution log to make sure each step runs without errors.
8. Review draft emails sent to Gmail and respond with “Approved” or “Declined” for testing.
9. Once verified, activate the workflow to run automatically.
10. If self-hosting n8n, ensure the workflow is running constantly on the server for uninterrupted email handling. See self-host n8n for help.

How This Workflow Works: Inputs, Processing, and Outputs

Inputs

• New business emails arriving in the configured IMAP inbox.
• Company knowledge documents stored in the Qdrant vector database.
• Human feedback replies classifying drafts as Approved or Declined via Gmail.

Processing Steps

• The Email Trigger (IMAP) node detects incoming emails.
• The Markdown node converts raw HTML content into markdown text.
• The Email Summarization Chain node creates a short and clear summary of the email content.
• The Write Email agent node generates a professional email reply. It uses OpenAI GPT language models combined with relevant info from the Qdrant vector store for accuracy.
• The draft reply is sent to the human team inbox through the Gmail node.
• Human feedback is classified by the Text Classifier node as Approved or Declined.
• If Declined, edits are made and sent again for review.
• If Approved, the Send Email node sends the final reply email via SMTP.

Outputs

• Draft email replies sent to Gmail inbox for human review.
• Final approved emails sent to original email senders.

Customization Ideas

• Modify summarization length or style in the Email Summarization Chain node prompt.
• Change the Send Email node fromEmail to use different sender addresses.
• Keep the Qdrant vector store refreshed with new company documents from Google Drive using the Google Drive nodes.
• Switch OpenAI models between GPT-3.5-turbo and GPT-4 for desired cost and result balance.
• Refine the Text Classifier node prompt and categories for more detailed human feedback.

Troubleshooting

• IMAP trigger not activating: Check IMAP credentials carefully and confirm server details with the email provider.
• AI-generated replies are vague: Make sure the Qdrant vector database has current company info and OpenAI API keys are valid.
• Feedback classification errors: Improve the classifier prompt and categories to better reflect human responses.

Pre-Production and Deployment Tips

• Confirm all credentials for IMAP, SMTP, OpenAI, Gmail, and Qdrant are correct.
• Test by sending emails to the trigger and checking each step’s output.
• Verify draft emails appear in Gmail and human feedback triggers correct workflow paths.
• Activate the workflow and monitor for errors using n8n’s execution logs.
• Use error handling within n8n to retry or alert if failures occur.

Summary and Results

✓ Saves many work hours by automating email reading, summarizing, drafting, and sending.

✓ Reduces mistakes by including human review before sending replies.

→ Produces clear, professional answers fast using AI plus company knowledge.

→ Frees up support teams to focus on harder tasks instead of routine email handling.