AI Workflows

Create structured conversation flows for consistent handling of specific scenarios.

What are Workflows?

Workflows define how your AI agent handles specific situations:

Customer: "I want to return my order"
           ↓
    Return Workflow Triggered
           ↓
    • Check order eligibility
    • Collect return reason
    • Generate return label
    • Confirm with customer

Creating Workflows

Natural Language Definition

Define workflows in plain English:

When a customer asks about returns:
1. Ask for their order number or email
2. Look up the order
3. Check if it's within 30-day return window
4. If eligible:
   - Ask for return reason
   - Generate return label
   - Send confirmation
5. If not eligible:
   - Explain the policy
   - Offer alternatives

Fyncall converts this to an executable workflow.

Visual Builder

Use the drag-and-drop builder:

1. Add trigger conditions
2. Build decision trees
3. Configure actions
4. Set fallback paths

Workflow Components

Triggers

What starts the workflow:

Trigger Type	Example
Intent	"return", "refund", "send back"
Keyword	"track order", "order status"
Event	Order placed, shipping update
Manual	Agent triggers workflow

Conditions

Decision points in the flow:

IF order.created_at > 30 days ago
  THEN explain_policy
ELSE IF order.status = "delivered"
  THEN start_return
ELSE
  THEN explain_status

Actions

What the workflow does:

Action	Description
Send Message	Reply to customer
Ask Question	Collect information
Call API	Fetch or update data
Create Task	Assign follow-up
Escalate	Transfer to human

Variables

Store and use data:

{{customer.name}}
{{order.number}}
{{product.name}}
{{custom.return_reason}}

Example Workflows

Order Status Workflow

Trigger: Intent matches "order status" or "track order"

Steps:
1. Ask: "Could you provide your order number or email?"
2. Collect: order_identifier
3. Look up order in system
4. If found:
   - Display order status card
   - Show tracking if available
5. If not found:
   - Ask for verification
   - Offer alternatives

Return Request Workflow

Trigger: Intent matches "return" or "refund"

Steps:
1. Identify customer
2. Find recent orders
3. Let customer select order
4. Check return eligibility
5. Collect return reason
6. Generate return label
7. Send confirmation email
8. Create return case

Appointment Booking Workflow

Trigger: Intent matches "book appointment" or "schedule"

Steps:
1. Ask for preferred date/time
2. Check availability
3. Present available slots
4. Confirm selection
5. Collect contact info
6. Create appointment
7. Send confirmation

Workflow Best Practices

Keep Flows Simple

• Limit to 5-7 steps
• Avoid deep nesting
• Provide escape routes

Handle Errors Gracefully

IF api_call fails
  THEN apologize and escalate

Provide Context

Give AI enough context at each step:

"We're helping {{customer.name}} with order #{{order.number}}"

Test Thoroughly

• Test happy path
• Test edge cases
• Test error scenarios
• Test with real data

Managing Workflows

Workflow Status

Status	Description
Draft	Not yet active
Active	In production use
Paused	Temporarily disabled
Archived	No longer needed

Version Control

• Workflows are versioned
• Roll back to previous versions
• Compare version differences

Analytics

Track workflow performance:

Metric	Description
Trigger Rate	How often triggered
Completion Rate	Successful completions
Escalation Rate	Handoffs to human
Avg Duration	Time to complete

Debugging Workflows

Test Mode

Run workflows in test mode:

1. Open workflow
2. Click Test
3. Simulate conversation
4. View step-by-step execution

Logs

Review workflow execution logs:

• Input received
• Conditions evaluated
• Actions taken
• Errors encountered