Back to Community
M

Marcus T.

@macro_marcus ·

Building a Customer Support Bot with n8n and AI: My Journey So Far

Hey fellow automation enthusiasts! I'm excited to share my experience of building a customer support bot using n8n and AI. I've been working on this project for a few weeks now, and I'm keen to get your feedback and advice on how to take it to the next level.

The Goal

My goal is to create a bot that can handle basic customer inquiries, such as answering FAQs, providing order status updates, and routing complex issues to human support agents. I've chosen to use n8n as the workflow automation tool, as it provides a flexible and customizable way to integrate with various APIs and services.

The Tech Stack

Here's a breakdown of the tech stack I'm using:

  • n8n as the workflow automation tool
  • Dialogflow (formerly known as API.ai) as the AI-powered chatbot platform
  • HubSpot API for customer data and support ticket integration
  • Slack API for notifications and routing complex issues to human support agents

The Workflow

Here's a high-level overview of the workflow I've set up:

  1. Customer Inquiry: The customer sends a message to the bot via a chat interface (e.g. website chat widget, Facebook Messenger, etc.)
  2. Intent Detection: The message is sent to Dialogflow for intent detection, which determines the customer's intent (e.g. FAQ, order status, etc.)
  3. n8n Workflow: Based on the intent, the corresponding n8n workflow is triggered, which may involve:
  • Retrieving customer data from HubSpot API
  • Checking order status from an external API
  • Sending a response back to the customer via the chat interface
  1. Human Intervention: If the bot is unable to resolve the issue, the workflow routes the issue to a human support agent via Slack, along with relevant customer data and context.

Challenges and Next Steps

While I've made good progress so far, I'm facing a few challenges:

  • Contextual Understanding: The bot sometimes struggles to understand the context of the conversation, leading to inaccurate responses.
  • Entity Extraction: Extracting relevant entities (e.g. order numbers, customer names) from customer messages is proving to be a challenge. I'd love to hear from you - have you built a similar customer support bot using n8n and AI? What were some of the challenges you faced, and how did you overcome them?

Example n8n Workflow

{"nodes": [...], "connections": {...}}

(Note: I'll be happy to share the full workflow JSON if anyone's interested!)

+9
3 comments

Add a comment

S
signal_sean2d ago

This is exactly what I needed! I've been trying to build a similar bot for my company, and I'm excited to learn from your experience. What AI model are you using for intent detection?

A
api_ace_andy2d ago

Regarding the AI model for intent detection, I'm curious to know more about the options you've considered. Have you looked into using a pre-trained model like Dialogflow or are you training your own model from scratch?

B
botmaker_chris2d ago

I've also built a support bot using n8n, but I used a different approach for routing complex issues. Instead of using a traditional AI model, I used a decision tree node in n8n to determine when to escalate issues to human support. Would love to hear more about your approach and how it's working out!