📚 Filter incoming emails automatically with AI

Reading Time: 4 minutes

Hello AI Enthusiast,

Today, we're sharing how we automated the sorting of emails received at AI Academy's general inbox. We'll show you how we built a system that uses ChatGPT to analyze, categorize, and route incoming messages to the right team members automatically.

The Problem

Managing a busy general email inbox (you know, that [email protected] address) can be overwhelming. At AI Academy, we were spending a lot of time manually sorting through partnership requests, support tickets, questions about our programs, and other general inquiries.

Sound familiar?

How We Do It: A Step-by-Step Guide

Let's walk through how we built our AI-powered email filtering system using Gmail, ChatGPT, and Make. We'll show you how we automated this process while maintaining the human touch where it matters.

Step 1: Set Up the Classification System

First, we defined our email categories:

• Partnerships

• Support requests

• Course inquiries

• Marketing/PR

• Other

Step 2: Build the Automation Flow

Here's how we connected everything using Make:

Our workflow uses four main modules:

a. Gmail Module: Monitors our general email inbox

b. OpenAI Module: ChatGPT analyzes the email and returns structured data in JSON format.

Before building our automation, we create a prompt and test it out. Only when satisfied with the outputs we add it to the OpenAI module. Here’s the prompt we used:

Act as a helpful assistant for [Company Name], a company specializing in [Company Description]. I will give you an email (subject and content), you will have to categorize it according to the category description I will give you here.

CATEGORIES:
Partnerships: Any business collaboration, corporate training, or partnership requests
Support: Technical issues, access problems, or student assistance
Courses: Questions about courses, pricing, or enrollment
Marketing: Press inquiries, marketing collaborations, or PR requests
Other: Anything that doesn't fit the above categories

Respond in JSON format with these fields:
{
  "category": "chosen category",
  "confidence": "high/medium/low",
  "reasoning": "brief explanation",
}

EMAIL SUBJECT: {{1.subject}}
EMAIL CONTENT: {{1.text}}

OUTPUT:

Our prompt extracts 3 key information from emails:

1. Category

2. Confidence of the model (to ensure that every email will always go to the correct team member)

3. Reasoning (to debug and improve the automation)

The answer, formatted as JSON, helps us structure the information consistently:

c. JSON Module: Parses the structured data.

This structured format acts as a bridge between ChatGPT's analysis and our automation, ensuring a correct filtering process.

d. Filtering requests: we set up different filter to match the correct email with each team member.

e. Slack module: to send the information needed to answer to the right team member

Your Turn

Ready to set up your own email filtering system? Here's a simple exercise to get started:

1. List your common email categories (aim for 4-6)

2. Write one classification prompt for ChatGPT

3. Test it with 5 sample emails

4. Start building your automation

Maybe start automating just one category first and then expand.

Want to get even more practical? Explore hands-on AI learning with AI Academy:

• Generative AI Project Bootcamp: Create AI prototypes to solve real business problems.

• Corporate Training: Tailored AI training for your team.

• Practical Introduction to ChatGPT: Free course on effectively using ChatGPT.

We'll be back with more AI tips soon!