AI agents become easier to understand when you build something that solves a real problem.
In this tutorial, we'll design a beginner-friendly lead qualification workflow for an education business. The goal is to process an enquiry, identify useful information, classify the lead, and prepare a follow-up draft for human approval.
This is a learning project. It uses fictional data and does not send real messages.
What Are We Building?
Imagine an education provider receives enquiries about AI, digital marketing, and data analytics courses.
A team member needs to identify the course of interest, understand the enquiry, and decide on the next step.
Our workflow will help organise this process.
Workflow overview
Sample Enquiry
|
v
Validate Input
|
v
Extract Relevant Fields
|
v
Classify Using Defined Rules
|
v
Prepare Follow-Up Draft
|
v
Human Review
The important design choice is that AI prepares a recommendation, while a person approves the final communication.
Step 1: Define the Input
Start with a simple JSON object:
{
"name": "Alex",
"course": "AI Agents",
"message": "I want to learn AI automation. Please share course details.",
"preferred_contact_time": "Evening"
}
Use fictional information while developing and testing your workflow.
Before processing the enquiry, validate that required fields exist and have the expected format. If the course or message is missing, return a validation error instead of guessing.
Step 2: Extract Structured Information
Ask the AI model to return a predictable structure, such as:
{
"course_interest": "AI Agents",
"intent": "Requesting course information",
"missing_information": [],
"requires_human_review": true
}
Use a schema or equivalent validation step to check the output. AI-generated JSON can still be malformed or contain unsupported values.
The workflow should not invent a budget, qualification, urgency, or personal preference that the person did not provide.
Step 3: Classify the Enquiry
For a beginner project, keep the rules simple.
If the course field matches an available programme, record that interest.
If required information is missing, flag the enquiry for clarification.
If the message is unclear, send it to human review.
Never treat the model's confidence alone as proof that a classification is correct.
You can use ordinary conditional logic for predictable rules and an AI model for interpreting free-text messages. Not every step needs an AI agent.
Step 4: Prepare a Follow-Up Draft
Using the validated fields, prepare a message such as:
Hi Alex, thanks for your interest in the AI Agents programme. We can share the available course details and help you understand the learning options. Would you prefer to receive the information by email or discuss it with our team?
Store the message as a draft. Do not send it automatically in this learning project.
Step 5: Test the Workflow
Test more than the ideal scenario.
Test case
Expected behaviour
Complete enquiry
Extract fields and prepare a draft
Missing course
Ask for clarification or flag for review
Empty message
Return a validation error
Unknown course
Flag for human review
Malformed AI output
Reject or retry safely
Prompt-injection attempt
Treat enquiry text as untrusted input
Duplicate enquiry
Flag or handle according to a defined rule
Also check that logs do not expose unnecessary personal information.
Step 6: Extend the Project
Once the basic version works, consider connecting a form, spreadsheet, CRM sandbox, or visual automation tool. Add authentication, error handling, access controls, and monitoring before working with real systems.
Measure useful outcomes such as classification accuracy, the number of cases requiring correction, and the time taken to process an enquiry. Do not assume automation improves performance until you test it.
What Can You Learn From This Project?
This small workflow introduces structured data, prompt design, conditional logic, AI integrations, validation, and human-in-the-loop automation.
For more business-oriented examples, read AI agent workflow examples for professionals, which explores five practical workflows and the skills behind them.
What would you add to this project? If you've built a lead qualification workflow, share your preferred tools, testing strategy, or lessons learned in the comments.
Top comments (1)
Really like the practical approach here. Instead of treating AI as a replacement for the sales team, this shows how it can take care of the repetitive qualification work while humans stay in control of the important conversations. Great beginner-friendly example of where AI workflows can actually create value.