Skip to main content

Chatbot Basics

Msghub's AI chatbot is knowledge-base grounded, meaning it answers questions from your data—not hallucinations. It works across all 6 channels (SMS, WhatsApp, Email, RCS, Instagram, Web Chat) and resolves 80% of customer queries without human intervention.

How the Chatbot Works

1. Customer Asks a Question

A customer messages you on WhatsApp, web chat, or any channel:

Customer: "Where's my order?"

2. Chatbot Receives the Message

Msghub receives the message and routes it to the chatbot.

3. Chatbot Searches Knowledge Base

The chatbot searches your knowledge base for relevant information:

  • FAQs
  • Product documentation
  • Policies
  • Order data (if integrated)

4. Chatbot Responds

The chatbot generates a response based on your knowledge base:

Chatbot: "I found your order! Here's the latest: Order #7842, Status: Out for Delivery, ETA: Today by 4:00 PM"

5. Customer Satisfied or Escalates

  • Satisfied: Conversation ends, customer is happy
  • Not satisfied: Customer asks for human agent, chatbot escalates

6. Human Agent Takes Over

A human agent picks up the conversation with full context:

  • Conversation history
  • Customer profile
  • Order data
  • Chatbot's summary

Chatbot Models

Msghub supports three AI models. Choose the one that fits your needs:

Claude (Anthropic)

  • Best for: Nuanced conversations, complex reasoning
  • Cost: Moderate
  • Speed: Fast
  • Accuracy: Very high

GPT-4 (OpenAI)

  • Best for: General-purpose, versatile
  • Cost: Moderate to high
  • Speed: Fast
  • Accuracy: Very high

Gemini (Google)

  • Best for: Cost-effective, fast
  • Cost: Low
  • Speed: Very fast
  • Accuracy: High

Switch Models

You can switch models anytime:

  1. Go Settings → AI Chatbot → Model
  2. Select your preferred model
  3. Click Save

No vendor lock-in—switch whenever you want.

Knowledge Base

The chatbot answers from your knowledge base. It doesn't make things up.

What Goes in the Knowledge Base

  • FAQs — Frequently asked questions and answers
  • Product Docs — Product features, specs, usage
  • Policies — Return policy, shipping policy, privacy policy
  • Guides — How-to guides, tutorials, troubleshooting
  • Company Info — About us, contact info, hours

Upload Knowledge Base

  1. Go Settings → AI Chatbot → Knowledge Base
  2. Click Upload Documents
  3. Choose files:
    • PDF
    • Word documents
    • Text files
    • Web pages
  4. Click Upload

Msghub will extract text and index it for the chatbot.

Add Knowledge Manually

  1. Go Settings → AI Chatbot → Knowledge Base
  2. Click Add Article
  3. Enter:
    • Title — Article title
    • Content — Article text
    • Category — Optional category
  4. Click Save

Knowledge Base Quality

The better your knowledge base, the better the chatbot:

  • Be specific — "Our return policy is 30 days" not "We have a return policy"
  • Include examples — "You can return items within 30 days of purchase. For example, if you bought on April 1st, you can return until May 1st"
  • Cover common questions — What are customers asking most?
  • Update regularly — Keep information current

Chatbot Capabilities

Answer Questions

The chatbot answers questions from your knowledge base:

Customer: "What's your return policy?"

Chatbot: "We offer 30-day hassle-free returns. You can return items within 30 days of purchase. Just pack the item and drop it off at any shipping location. We'll process your refund within 5 business days."

Look Up Order Data

If you integrate your order system, the chatbot can look up real order data:

Customer: "Where's my order?"

Chatbot: "I found your order! Order #7842, Status: Out for Delivery, ETA: Today by 4:00 PM. Would you like me to send you a notification when it arrives?"

Qualify Leads

The chatbot can qualify leads for your sales team:

Customer: "I'm interested in your enterprise plan"

Chatbot: "Great! I'd love to help. Can you tell me:

  1. How many users do you need?
  2. What's your monthly message volume?
  3. What integrations do you need?"

Book Appointments

The chatbot can book appointments:

Customer: "I'd like to schedule a demo"

Chatbot: "Perfect! I can help with that. What date and time work best for you?"

Collect Information

The chatbot can collect customer information:

Customer: "I want to return my order"

Chatbot: "I can help with that. What's your order number?"

Customer: "ORD-4821"

Chatbot: "Got it. Can you tell me why you want to return it?"

Escalation to Human Agents

When the chatbot can't help, it escalates to a human agent.

Escalation Triggers

The chatbot escalates when:

  • Customer asks for a human agent
  • Chatbot can't find an answer in knowledge base
  • Conversation requires human judgment (refunds, complaints)
  • Customer is frustrated or angry

Escalation Message

Chatbot: "I understand you need help with a refund. Let me connect you with an agent who can help. One moment..."

Agent Handoff

The human agent receives:

  • Full conversation history
  • Customer profile
  • Order data (if available)
  • Chatbot's summary of the issue

The agent can see:

Conversation History:
- Customer: "I want to return my order"
- Chatbot: "I can help with that. What's your order number?"
- Customer: "ORD-4821"
- Chatbot: "I found your order from April 1st for $89.99. Why would you like to return it?"
- Customer: "The product is damaged"

Customer Profile:
- Name: Rahul
- Email: rahul@example.com
- Phone: +919876543210
- VIP: Yes

Order Data:
- Order #7842
- Date: April 1, 2024
- Amount: $89.99
- Status: Delivered
- Items: Sony Headphones

AI Summary:
Customer wants to return order due to damaged product. This is a legitimate return request. Recommend processing refund immediately.

Chatbot Across Channels

The same chatbot logic works across all 6 channels.

WhatsApp

Customer: "Where's my order?"

Chatbot: "I found your order! Order #7842, Status: Out for Delivery, ETA: Today by 4:00 PM"

Web Chat

Customer: "What's your return policy?"

Chatbot: "We offer 30-day hassle-free returns!"

Email

Customer: "Do you have this product in blue?"

Chatbot: "Yes! We have it in blue, red, and black. Which color would you like?"

SMS

Customer: "How do I track my order?"

Chatbot: "Reply with your order number and I'll send you the tracking link"

Instagram DM

Customer: "Is this product available?"

Chatbot: "Yes! It's in stock in all sizes. Would you like me to add it to your cart?"

RCS

Customer: "What's the price?"

Chatbot: "The Sony Headphones are $349. Would you like to buy now?"

Chatbot Performance

Resolution Rate

Msghub's chatbot resolves 80% of customer queries without human intervention.

This means:

  • 80% of conversations end with the chatbot
  • 20% escalate to human agents
  • Agents handle only complex issues

Response Time

  • Chatbot: 2 seconds average
  • Human agent: 5-30 minutes average

The chatbot provides instant responses 24/7.

Satisfaction

Customers are satisfied when:

  • The chatbot answers their question
  • The response is accurate
  • The chatbot escalates when needed
  • Human agents are available when needed

Best Practices

Knowledge Base

  • Be comprehensive — Cover all common questions
  • Be accurate — Keep information current
  • Be clear — Use simple language
  • Be organized — Group related topics

Chatbot Responses

  • Be helpful — Answer the question directly
  • Be concise — Keep responses short
  • Be friendly — Use a conversational tone
  • Know when to escalate — Don't pretend to know

Training

  • Review conversations — What questions does the chatbot struggle with?
  • Update knowledge base — Add answers to common questions
  • Test regularly — Ask the chatbot questions and check responses
  • Iterate — Continuously improve

Troubleshooting

Chatbot not responding

  1. Check if enabled — Is the chatbot enabled in Settings?
  2. Check knowledge base — Does it have answers?
  3. Check flow — Is the chatbot flow configured?

Low resolution rate

  1. Check knowledge base — Does it cover common questions?
  2. Check accuracy — Are the answers correct?
  3. Check escalation — Is the chatbot escalating appropriately?

Inaccurate responses

  1. Check knowledge base — Is the information correct?
  2. Check model — Try a different AI model
  3. Check training — Provide more examples

See Also