Automate Email Responses with GPT-4o-mini and Human Review in Gmail

This n8n workflow automates the process of drafting email responses using an AI model (GPT-4o-mini) and provides an optional human review step before sending. It helps streamline customer support, sales inquiries, or any scenario requiring quick, intelligent email replies.

What it does

This workflow performs the following steps:

1. Triggers on new emails: Listens for new incoming emails in a specified Gmail account.
2. Classifies email intent: Uses a Text Classifier node to determine the intent or category of the incoming email.
3. Generates a draft response: If the email intent is suitable for AI response, a Basic LLM Chain node, powered by an OpenAI Chat Model (GPT-4o-mini), generates a draft reply.
4. Parses structured output: A Structured Output Parser extracts the generated response in a defined format.
5. Conditional routing for review: An If node checks a condition (likely based on the classified intent or a confidence score) to decide if human review is needed.
6. Sends draft for human review (optional): If human review is required, the draft email is sent to a specified Gmail address (or another system for review). This step is currently represented by a "No Operation" node, implying a placeholder for a human-in-the-loop (HITL) process.
7. Sends automated response: If no human review is needed, the AI-generated draft is automatically sent as a reply via Gmail.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n instance: A running n8n instance.
• Gmail Account: Configured with credentials in n8n for both triggering on new emails and sending replies.
• OpenAI API Key: For the OpenAI Chat Model (GPT-4o-mini) used in the Basic LLM Chain. This will be configured as an n8n credential.
• Langchain Nodes: Ensure the @n8n/n8n-nodes-langchain package is installed and enabled in your n8n instance, as it provides the AI-related nodes (Basic LLM Chain, OpenAI Chat Model, Structured Output Parser, Text Classifier).

Setup/Usage

1. Import the workflow: Download the JSON provided and import it into your n8n instance.
2. Configure Credentials:
   • Set up your Gmail OAuth2 credentials for both the Gmail Trigger and Gmail nodes.
   • Set up your OpenAI API Key credential for the OpenAI Chat Model.
3. Activate the workflow: Once all credentials are set and configurations are done, activate the workflow.
4. Customize:
   • Gmail Trigger: Adjust the mailbox and query settings in the Gmail Trigger node to specify which emails should initiate the workflow.
   • Text Classifier: Configure the Text Classifier node with your desired categories and examples to accurately classify incoming email intent.
   • Basic LLM Chain & OpenAI Chat Model: Fine-tune the prompt in the Basic LLM Chain to guide GPT-4o-mini in generating appropriate responses. Adjust the model parameters in the OpenAI Chat Model if needed.
   • Structured Output Parser: Ensure the schema in this node matches the expected output format from your LLM chain.
   • If Node: Modify the condition in the If node to define when a human review is necessary (e.g., based on classification confidence, specific keywords, or a flag from the AI).
   • Human Review (No Operation, do nothing): Replace the "No Operation, do nothing" node connected to the False branch of the If node with your preferred human review process. This could involve:
     • Sending an email to a human reviewer with the draft.
     • Posting to a Slack channel or other communication platform.
     • Creating a task in a project management tool.
     • Using n8n's built-in "Wait for Webhook" node for a manual approval step.
   • Gmail (Send Reply): Configure the Gmail node to send the final response, ensuring it replies to the original sender and includes the AI-generated content.