How to Use AI Reply in ZazzyAgent Flow Builder
Not every customer message can be handled with a fixed keyword or button.
A customer might ask:
How much does your enterprise plan cost?
Another might say:
I'm looking for something for a team of about 50 people.
Another might ask:
Can you explain how this works?
These messages don't always match a predefined keyword.
That's where AI Reply can help.
What is AI Reply?
AI Reply allows a flow to use your configured AI training/knowledge to respond to customer questions.
Instead of matching only a specific keyword, the AI can interpret what the customer is asking and generate an appropriate response.
The current Flow Builder includes an AI Reply component for AI-powered responses.
AI Reply vs AI Agent
These features are related but shouldn't be confused.
AI Reply
Adds an AI-powered response inside a flow or fallback path.
AI Agent
A dedicated AI assistant with its own role, System Prompt, Knowledge, Actions and routing.
For example:
AI Reply
→ answer a question inside a specific flow.
AI Agent
→ continuously handle a broader conversation according to its configuration.
When should you use AI Reply?
AI Reply is useful when:
You already have a structured bot flow.
Most of the journey is predefined.
Some questions require natural-language understanding.
You want AI to answer questions that don't match your fixed paths.
You want AI as a fallback rather than the primary conversation engine.
Example
A flow starts with:
Welcome! What would you like to do?
Pricing
Book Demo
Support
The customer then asks:
Do you offer annual billing?
Instead of creating a separate keyword for every possible phrasing, an AI Reply can handle the question using your configured knowledge.
Add AI Reply to a Flow
Open Flow Builder.
Create or open the flow where you want AI to respond.
Add the AI Reply component.
Open its configuration.
The exact labels can change as the ZazzyAgent interface evolves, but the component's purpose is to connect the conversation to the configured AI training/knowledge.
Select the AI knowledge/training configuration
Choose the appropriate AI knowledge or training configuration for the reply.
This is important because AI Reply needs information to work from.
For example, your AI knowledge might contain:
Pricing
Services
FAQs
Business information
Policies
Product information
Add AI Reply after a customer question
A simple flow can look like:
Customer
↓
Question
↓
AI Reply
↓
Answer
Use AI Reply as a fallback
This is one of the most useful applications.
Suppose your bot already has:
Pricing
Support
Booking
Products
But the customer asks something unexpected.
You can let AI take over when there is no matching predefined response.
This creates:
Customer message
↓
Matches existing flow?
↙ ↘
Yes No
↓ ↓
Flow AI Reply
AI as Fallback Only
ZazzyAgent's AI configuration supports a fallback-only mode.
In this setup, AI responds when the customer's message does not match the existing predefined automation.
This is useful when you want your structured flows to remain in control while AI handles unexpected questions.
AI for All Queries
The platform also supports an AI mode where AI can handle all customer queries.
This is very different from fallback-only mode.
All Queries
AI becomes the primary response engine.
Fallback Only
Predefined flows remain primary and AI is used when those flows don't match.
Choose the mode based on how much control you want from your structured automation.
When should AI Reply handle all queries?
This can make sense when your chatbot is primarily conversational.
For example:
Customer support assistant.
The customer can ask questions naturally without navigating menus.
When should AI Reply be fallback-only?
This is useful when your business relies heavily on structured automation.
For example:
Book Appointment
→ structured booking flow.
Submit Application
→ structured form.
Check Order
→ structured process.
But if someone asks a general question:
Do you support Saturday appointments?
AI can answer it.
AI Reply needs good Knowledge
An AI component does not automatically know your business.
Your Knowledge should contain the information it needs.
See How to Train ZazzyAgent AI With Knowledge.
AI Reply can use contextual memory
The current AI configuration supports contextual memory, allowing the AI to consider previous conversation messages when generating responses.
This matters because:
Customer:
Do you offer the enterprise plan?
AI:
Yes.
Customer:
How much is it?
The second message depends on the first.
Without context, the conversation can feel disconnected.
Configure conversation context carefully
More context isn't always better.
You should use an appropriate context depth for your use case.
For a simple FAQ bot, a small amount of context may be enough.
For longer support conversations, more context can be useful.
Use AI Reply with structured flows
One of the strongest patterns is hybrid automation.
For example:
Structured Menu
↓
Customer asks question
↓
AI Reply
↓
Customer gets answer
↓
Returns to structured flow
This gives customers both predictable options and natural conversation.
AI Reply with User Input
You can also combine AI with User Input.
Example:
What's your budget?
Customer:
Around ₹75,000.
User Input stores the response.
AI can then use the available customer context to continue the conversation.
AI Reply with conditions
You can use structured conditions for deterministic tasks and AI for natural-language questions.
For example:
Customer
↓
Condition
↓
Known task?
↙ ↘
Yes No
↓ ↓
Flow AI Reply
This is often easier to maintain than trying to make AI control every single step.
AI Reply and No Match Reply
No Match Reply is one of the natural places to use AI Reply.
A customer sends something that doesn't match your predefined automation.
Instead of returning a generic:
Sorry, I didn't understand.
the platform can use your AI configuration to answer based on the available knowledge.
See No Match Reply in ZazzyAgent.
Don't use AI Reply to replace every deterministic action
Suppose a customer says:
Book an appointment.
If you have a structured booking process, don't necessarily make AI explain booking and then ask the customer to type every field manually.
A better approach can be:
AI understands request
→ Structured Booking Flow
This lets AI handle understanding while Flow Builder handles execution.
Test AI Reply with different phrasings
Don't test only:
Pricing
Try:
How much does it cost?
What's the price?
Do you have annual pricing?
How much would it cost for 20 users?
The AI should handle natural variations without relying on exact keywords.
Test questions outside your Knowledge
Also test something you don't support.
Your AI should not invent an answer.
A good fallback is:
I don't have that information available. I can connect you with our team.
Common problems
AI gives incorrect answers
Check your Knowledge first.
AI doesn't respond
Check whether AI Reply is actually connected to the flow or enabled through the appropriate AI settings.
AI answers but ignores the flow
Review whether you're using AI for all queries or fallback-only mode.
AI repeats the conversation
Check contextual-memory configuration.
AI invents information
Strengthen your System Prompt and Knowledge boundaries.
AI Reply vs keyword matching
Keyword bot:
pricing
AI Reply:
Can you tell me what your plans cost for a small company?
The second approach handles language variations much better.
Recommended architecture
For many businesses, this is a strong starting point:
Structured Automation
↓
Known Task → Flow
↓
Unknown Question → AI Reply
↓
Complex Task → AI Agent
↓
Needs Human → Human Handoff
The goal isn't to make every step AI.
It's to use AI where natural language is the problem.