Automatically classify customer conversations using AI-powered custom labels tailored to your business workflows.
Overview
AI Custom Auto-Labeling allows merchants to create their own labels and automatically categorize customer conversations based on their unique business workflows. This helps teams organize tickets more accurately, improve automation, enhance reporting, and avoid relying on generic predefined labels.
AI Custom Auto-Labeling
Now merchants can:
Create their own labels
Define label descriptions
Test labels before activating them
Build hierarchical (nested) label structures
Enable or disable AI detection per label
This allows merchants to create labels that match their actual business workflows rather than adapting to predefined categories.
How It Works
AI Custom Auto-Labeling uses merchant-defined training examples to identify conversation intent and automatically apply labels.
Step 1: Create a Label
Merchant defines:
Label Name
Label Information
Parent Label (optional for nested structures)
Example:
Label Name: Premium Support
Information: Customers requesting assistance under premium support plans.
Step 2: Add Sample Messages
Provide example customer messages that should match the label.
Up to 10 sample messages can be added per label.
Examples:
"I need urgent help with my enterprise account."
"Can someone from premium support contact me?"
"My dedicated account manager asked me to raise this ticket."
The AI learns the context and patterns from these examples.
Step 3: Test the Label
Before activation, merchants can validate label behavior using Test Mode.
Enter a sample message and review:
Predicted Label
Confidence Score
AI Reasoning
This helps ensure the label performs as expected before going live.
Step 4: Activate AI Detection
Once enabled, AI automatically evaluates incoming conversations and applies the appropriate labels.
A conversation can receive multiple labels when relevant.
Key Capabilities
Custom Labels
Create labels specific to your business needs.
Each label supports:
Custom name
Custom description
Up to 10 training sample messages
Per-Label AI Detection Control
AI detection can be configured individually for each label.
Merchants can:
Enable AI detection
Disable AI detection
Keep labels for manual use only
This provides complete control over automation.
Multi-Label Detection
A single conversation may belong to multiple categories.
AI supports assigning up to 5 labels to the same conversation.
Example:
Customer message:
"My premium order arrived damaged and I need a refund."
Possible labels:
Premium Customer
Damaged Product
Refund Request
All relevant labels can be applied automatically.
Nested & Hierarchical Labels
Labels can be organized into parent-child structures.
Example:
Orders
Refund Request
Exchange Request
Delivery Delay
Support
Technical Issue
Account Access
Billing Issue
This helps teams maintain structured categorization and reporting.
Test Mode with AI Reasoning
Before enabling a label, merchants can validate AI behavior.
AI Reasoning
Returns which label will be selected.
Example:
Message:
"I received the wrong item in my shipment."
AI Result:
Order Issue
Label Retention for Returning Customers
Tickets can retain previously assigned labels for returning customers.
Benefits include:
Better customer context
Reduced repetitive classification
Faster ticket routing
Improved customer history visibility
Supported Use Cases
Product-Specific Classification
Automatically categorize conversations based on product lines.
Example:
Product A Support
Product B Support
Product C Support
Support Tier Routing
Differentiate customer priority levels.
Example:
Standard Support
Premium Support
Enterprise Support
Business Workflow Automation
Organize conversations according to internal processes.
Example:
Escalation Required
Finance Review
Technical Investigation
Compliance Check
Reporting & Analytics
Create business-specific reporting categories.
Example:
Subscription Cancellation
Upgrade Requests
Shipping Issues
Partner Inquiries
Key Benefits
Labels Match Your Business
Create categories that reflect how your teams actually work rather than relying on predefined labels.
Higher Adoption
Teams are more likely to use labels when they align with existing workflows.
Better Reporting Granularity
Track business-specific metrics with meaningful categories.
Example:
Instead of:
Other
You can report on:
VIP Complaints
Warranty Requests
Product Feedback
Delivery Issues
Reduced "Other" Bucket
Custom labels significantly reduce uncategorized conversations.
This improves reporting accuracy and operational visibility.
Improved Automation
Labels can power:
Assignment rules
Routing logic
SLA workflows
Escalation processes
Analytics dashboards
Setup Process
Setup Type
Self-Serve
Steps
Create a custom label
Add label description
Add sample messages
Test AI detection
Enable AI detection
Monitor performance
Refine training samples if required
Smart Labeling Settings
Smart Labeling Settings determine when AI evaluates conversations and applies labels to tickets.
Before Agent Assignment
AI analyzes the conversation and applies relevant labels before the ticket is assigned to an agent. This ensures tickets are categorized correctly from the start, helping with routing, prioritization, and workflow automation.
Until the Ticket is Resolved
AI continues to evaluate conversations and update labels throughout the ticket lifecycle, even after an agent has been assigned. This is useful when customer intent changes during the conversation and labels need to stay up to date.
Label Limits
To maintain labeling quality and relevance, AI can apply a maximum of 5 labels to a single ticket. The most relevant labels are selected based on the conversation context.
Best Practices
Use Clear Label Definitions
Avoid vague labels.
Good Example:
Refund Request
Poor Example:
Customer Issue
Add Diverse Training Examples
Include different customer phrasings for the same intent.
Example:
"I want a refund."
"Can I get my money back?"
"Please cancel and refund my order."
This improves detection accuracy.
Avoid Overlapping Labels
Ensure labels have distinct meanings.
Example:
Instead of:
Order Problem
Order Issue
Use:
Delivery Delay
Damaged Product
Regularly Test Labels
Use Test Mode whenever:
New labels are created
Sample messages are updated
Business workflows change
Limitations
Sample Quality Impacts Accuracy
AI performance depends heavily on the quality and diversity of training examples provided.
Poor or limited samples may reduce detection accuracy.
Maximum Sample Messages
Each label supports up to:
10 sample messages
No Bulk Import
Bulk upload or import of labels is not supported in the current release.
Labels must be created individually.
AI Confidence May Vary
Detection confidence can differ based on:
Message complexity
Training quality
Similarity between labels
Testing before activation is strongly recommended.
System Labels
Under Settings → Labels → System Labels, some labels are provided by default by the platform.
The following system labels are tied to predefined platform workflows:
Refund
CTWA
Spin The Wheel
Product Info
Why is Smart Labeling disabled for these labels?
These four labels are system-defined labels and are not designed to be AI-driven. Therefore:
The Smart Labeling toggle is disabled.
Smart Labeling cannot be enabled for these labels.
AI cannot automatically assign these labels based on conversation content.
The Smart Labeling configuration for these labels cannot be modified.
If you want AI to automatically identify and apply a similar label, create a Custom Label under My Labels and enable Smart Labeling for that label.
Custom Labels allow you to:
Enable Smart Labeling.
Add label information to guide AI classification.
Add sample messages to improve AI accuracy.
Configure labels based on your business requirements.
Note: This restriction applies only to the system labels Refund, CTWA, Spin The Wheel, and Product Info. If you need AI-driven labeling for these use cases, create equivalent custom labels and enable Smart Labeling on them.
Chat Activity on Label detection in Helpdesk ticket:
Frequently Asked Questions (FAQs)
1. What is AI Custom Auto-Labeling?
AI Custom Auto-Labeling allows merchants to create their own labels and train AI to automatically classify conversations using merchant-defined examples.
2. How is this different from previous AI labeling?
Previously, AI could only apply predefined BIK labels.
Now merchants can create and train their own labels for custom business workflows.
3. How many sample messages can be added?
Each label supports up to 10 sample messages.
4. Can a conversation receive multiple labels?
Yes.
AI supports assigning up to 5 labels to a single conversation.
6. Can I disable AI detection for specific labels?
Yes.
Each label has its own AI detection toggle.
7. Can I create parent and child labels?
Yes.
Hierarchical and nested label structures are supported.
8. How do I know if a label will work correctly?
Use Test Mode before activation.
The system provides:
Predicted labels
Confidence score
AI reasoning
9. What happens if AI applies the wrong label?
Merchants can review results, update sample messages, retrain labels, and improve future accuracy.
10. Do returning customers retain previous labels?
Yes. If enabled
The platform supports label retention for returning customers, helping maintain historical context.
11. Is setup technical?
No.
The feature is fully self-serve and can typically be configured within 5 - 10 minutes.
12. Can labels be imported in bulk?
No.
Bulk import is not supported in the current version.
Labels must be created manually.
13. What is the biggest factor affecting accuracy?
The quality and diversity of sample messages used to train the AI.
Providing clear and representative examples significantly improves detection performance.
For further assistance or to raise feature requests related to AI Labels, please contact [email protected].