Build Your First Conversational Agent

In this guide, you’ll build a conversational AI agent that collects customer feedback through natural dialogue. Instead of a boring form, users chat with an AI that guides them through providing information. What you’ll build: A feedback collector that asks about satisfaction, collects suggestions, and thanks the user — all through conversation. Time: 15-20 minutes

What You’ll Learn

How to create a skillset for conversational data collection
Setting up the Ask a Person node
Configuring communication channels
Testing your conversational agent

Prerequisites

A CogniAgent account
Basic understanding of workflows

The End Result

Here’s what the conversation will look like:

AI: Hi! Thanks for being a customer. I’d love to hear your feedback. On a scale of 1-10, how satisfied are you with our service? User: I’d say about 8 AI: Great to hear! An 8 is solid. What’s one thing we could do to make it a 10? User: Faster shipping would be nice AI: That’s helpful feedback — faster shipping is definitely something we’re working on. Anything else you’d like to share? User: Nope, that’s it AI: Perfect! Thanks for taking the time. Your feedback helps us improve. Have a great day!

Behind the scenes, your workflow captures:

Satisfaction score: 8
Improvement suggestion: “Faster shipping would be nice”
Additional comments: None

Step 1: Create Your Application

Go to Applications

From the CogniAgent dashboard, click Applications in the sidebar.

Create a new application

Click New Application and name it “Customer Feedback Bot”.

Create a workflow

Inside your application, click New Workflow and name it “Feedback Collector”.

Step 2: Add the Start Trigger

Every workflow needs a trigger. For testing, we’ll use a manual start.

Add a Start node

Drag a Start node onto the canvas. This lets you manually trigger the workflow while testing.

Configure test data

In the Start node settings, add some test input data:

{
  "customerName": "Alex",
  "customerEmail": "[email protected]"
}

Step 3: Add the Ask a Person Node

This is where the magic happens. The Ask a Person node manages the conversation.

Add Ask a Person node

Drag an Ask a Person node onto the canvas and connect it to the Start node.

Name it

Set the display name to “Feedback Conversation” and key to feedback_conv.

Step 4: Configure Your Skillset

The skillset defines what information your AI collects and how it converses.

Open skillset configuration

In the Ask a Person node settings, find the Skillset section.

Create the skillset

Add a new skillset with this configuration:

{
  "name": "Customer Feedback",
  "description": "Collect customer satisfaction feedback through friendly conversation",

  "persona": {
    "agentName": "Feedback Assistant",
    "tone": "Friendly, appreciative, and genuinely interested in feedback"
  },

  "fields": [
    {
      "id": "satisfaction_score",
      "name": "Satisfaction Score",
      "description": "Customer's satisfaction rating from 1-10",
      "type": "number",
      "required": true
    },
    {
      "id": "improvement_suggestion",
      "name": "Improvement Suggestion",
      "description": "What could we do better to improve their experience",
      "type": "text",
      "required": true
    },
    {
      "id": "additional_comments",
      "name": "Additional Comments",
      "description": "Any other feedback they want to share",
      "type": "text",
      "required": false
    }
  ],

  "greeting": "Hi {{customerName}}! Thanks for being a customer. I'd love to hear your feedback.",
  "confirmationMessage": "Perfect! Thanks for taking the time. Your feedback helps us improve. Have a great day!"
}

Understanding the Skillset

Let’s break down what each part does:

Section	Purpose
`persona`	Defines the AI’s personality and tone
`fields`	The information to collect (like form fields, but conversational)
`greeting`	The opening message (uses `{{customerName}}` from workflow data)
`confirmationMessage`	The closing message after all data is collected

Understanding Fields

Each field has:

id — How you reference the data later ({{feedback_conv.satisfaction_score}})
name — Human-readable name for summaries
description — Tells the AI what to ask for (the AI uses this to phrase questions naturally)
type — Data type (text, number, selection, etc.)
required — Whether to skip if user can’t/won’t answer

Step 5: Configure the Channel

Choose how users will communicate with your agent.

Select a channel

For testing, select Web Widget as the channel. This gives you a chat interface to test with.

Optional: Add more channels

You can also enable Email or Slack if you have those integrations connected.

Step 6: Handle the Completion

When the conversation finishes, you’ll want to do something with the collected data.

Connect the event handle

The Ask a Person node has an Event Handle output. This triggers when the conversation completes.

Add an Update Variable node

Connect an Update Variable node to store the feedback:

{
  "variableName": "collected_feedback",
  "value": {
    "score": "{{feedback_conv.collectedData.satisfaction_score}}",
    "suggestion": "{{feedback_conv.collectedData.improvement_suggestion}}",
    "comments": "{{feedback_conv.collectedData.additional_comments}}",
    "collectedAt": "{{timestamp}}"
  }
}

Optional: Add notifications

Add an App Action node to send the feedback to Slack or add it to a Google Sheet.

Your workflow should now look like:

Step 7: Test Your Agent

Save the workflow

Click Save to save your workflow.

Run the workflow

Click Run to start the workflow. This will initiate the conversation.

Open the chat

A chat window will appear (or check the channel you configured). You’ll see the greeting message.

Have a conversation

Respond naturally! Try:

Giving a satisfaction score
Providing a suggestion
Saying you have nothing else to add

Check the results

After the conversation ends, check the Execution History to see the collected data.

What’s Happening Behind the Scenes

When your agent runs, here’s the flow:

Workflow starts → Passes customer data to the skillset
AI sends greeting → Uses the persona and greeting template
User responds → AI analyzes the response
AI decides what to do → Accept answer? Ask for clarification? Move on?
Repeat → Until all required fields are collected
AI confirms and closes → Sends confirmation message
Workflow continues → Event handle triggers with collected data

The AI makes intelligent decisions at each step. If the user says “probably like a 7 or 8”, it picks up that 7-8 is the satisfaction score without asking again.

Experiment and Extend

Now that you have a working agent, try these variations:

Add Field Validation

Make the satisfaction score require a 1-10 range:

{
  "id": "satisfaction_score",
  "name": "Satisfaction Score",
  "description": "Customer's satisfaction rating, must be a number from 1 to 10",
  "type": "number",
  "required": true,
  "maxAttempts": 3
}

The maxAttempts limits how many times the AI will ask for clarification before moving on.

Add a Conditional Field

Only ask about specific issues if the score is low:

{
  "id": "issue_details",
  "name": "Issue Details",
  "description": "If satisfaction is 5 or below, what specific issues did they experience",
  "type": "text",
  "required": false,
  "condition": "satisfaction_score <= 5"
}

Connect to a Real Trigger

Replace the Start node with:

App Trigger (Email) — Start feedback collection when a customer emails
Webhook — Trigger from your website after a purchase
Scheduled Trigger — Send feedback requests on a schedule

Send Results Somewhere

Add nodes after the conversation completes:

App Action (Google Sheets) — Log feedback to a spreadsheet
App Action (Slack) — Alert your team about low scores
App Action (HubSpot) — Update the customer record

Troubleshooting

The AI keeps asking the same question

Check your field descriptions. If they’re vague, the AI might not recognize valid answers. Be specific about what you’re collecting.

The conversation feels robotic

Adjust the persona.tone in your skillset. Add personality descriptors like “warm”, “casual”, or “empathetic”.

Users are getting stuck

Add maxAttempts to fields so the AI moves on after a few tries. Also ensure optional fields are marked required: false.

I'm not getting the collected data

Make sure you’re referencing the correct path: {{node_key.collectedData.field_id}}. Check your node key matches.

Next Steps

You’ve built your first conversational agent! Here’s where to go next:

Lead Qualification Guide

Build a more complex agent that qualifies leads and syncs to your CRM.

Skillset Engine Reference

Deep dive into all skillset configuration options.

Channels Documentation

Learn about all available communication channels.

E-commerce Returns Guide

See a real-world example with integrations.

Guides

​Build Your First Conversational Agent

​What You’ll Learn

​Prerequisites

​The End Result

​Step 1: Create Your Application

​Step 2: Add the Start Trigger

​Step 3: Add the Ask a Person Node

​Step 4: Configure Your Skillset

​Understanding the Skillset

​Understanding Fields

​Step 5: Configure the Channel

​Step 6: Handle the Completion

​Step 7: Test Your Agent

​What’s Happening Behind the Scenes

​Experiment and Extend

​Add Field Validation

​Add a Conditional Field

​Connect to a Real Trigger

​Send Results Somewhere

​Troubleshooting

​Next Steps