Kai Resolution Agent

Kai is Kaana's AI-powered resolution agent that analyzes System Feed signals and helps fix issues directly in your connected billing and CRM systems.

Overview

When you encounter a signal in the System Feed, Kai can analyze the underlying problem, identify root causes, and propose specific actions to resolve it. For automated actions, Kai can execute fixes directly in your connected systems.

How Kai Thinks About Problems

Kai uses intelligent anomaly detection to focus on issues that truly matter to your business. Rather than flagging every small error, Kai identifies patterns that indicate systemic, abnormal, or high-risk behavior.

What Kai Looks For (Anomalies)

Anomaly Type	Example
Failure rate spikes	Payment failures jump from 2% to 20%
Volume changes	Invoice generation drops 50% overnight
Multi-entity patterns	Same error affecting multiple customers
System inconsistencies	Usage recorded but no charges created
Post-change failures	Billing errors starting after a config update

What Kai Ignores (Not Anomalies)

• A single customer's card decline due to insufficient funds

• Isolated expired card errors

• One-off data entry mistakes

This approach ensures Kai helps you address root causes rather than chasing individual symptoms.

Understanding Root Causes

Kai understands how billing systems work and fixes problems at the right level. Billing operations flow through a sequence:

Usage → Rating → Billing → Invoices → Payments

If a billing run fails, Kai knows that no invoices were created yet, so it will recommend retrying the billing run - not individual invoices. This prevents wasted effort on symptoms while the root cause remains.

For example:

• If usage processing fails, invoices can't be generated - Kai fixes the usage first

• If credentials are invalid, payments will fail - Kai recommends refreshing credentials before retrying payments

• If a billing run fails, affected invoices don't exist yet - Kai retries the billing run, not invoices

Starting a Resolution Session

1. Navigate to the System Feed

2. Open any signal that requires action

3. Click Resolve with Kai to start an agent session

4. Kai will analyze the signal and present its findings

The Resolution Process

Step 1: Analysis

Kai analyzes the signal and provides:

• Root Cause - What's actually causing the problem

• Impact Assessment - How this affects your business

• Affected Systems - Which integrations and entities are involved

• Confidence Level - How certain Kai is about the diagnosis

Step 2: Proposed Actions

Based on the analysis, Kai proposes specific actions to resolve the issue. Each action includes:

• Description - What the action does

• Type - Automated (Kai executes) or Manual (you execute)

• Target System - Which integration will be affected

• Risk Level - Low, Medium, or High

• Expected Outcome - What should happen after execution

• Detailed Steps - Specific operations that will be performed

Step 3: Confirmation

Review each proposed action before execution. You control what Kai does:

• Select which actions to approve

• Skip actions you want to handle manually

• Cancel the session if you prefer to address it differently

Step 4: Execution

For approved automated actions, Kai:

1. Connects to your integration

2. Executes the specific API calls

3. Records the system response

4. Reports success or failure with details

Step 5: Results

After execution, you see detailed results:

Successful Actions:

• Green checkmark with "Action Completed"

• System response data (invoice numbers, IDs, status, amounts)

• Confirmation of what changed

Failed Actions:

• Red X with "Action Failed"

• Error message explaining what went wrong

• Full API response for troubleshooting

Understanding the Action Timeline

Kai presents actions in a visual timeline format:

Stage	Color	Description
Issue	Red	The problem being addressed
Action	Amber	The resolution action to take
Steps	Blue	Detailed steps involved
API Call	Purple	The specific system API being called
Expected Result	Green	What should happen after execution

Action Types

Actions are categorized by execution method:

Automated Actions

Kai can execute these directly in your connected systems:

• No manual intervention required

• Real-time execution with immediate feedback

• Full audit trail of changes made

Manual Actions

Some situations require your intervention:

• Actions that need human judgment

• Steps that must be performed in the external system's UI

• Escalations to support teams or customers

For manual actions, Kai provides detailed instructions on what to do.

Available Resolution Actions

Kai can perform a variety of actions across your connected billing systems:

Integration Actions

Action	Description
Refresh Credentials	Update API tokens or OAuth connections
Verify Integration	Test that the connection is working

Usage & Rating Actions

Action	Description
Reprocess Usage Batch	Resubmit failed usage records
Replay Usage Events	Resend usage events for rating
Recalculate Charges	Recalculate pricing for charges

Billing Run Actions

Action	Description
Retry Billing Run	Re-execute a failed billing run
Cancel Billing Run	Cancel a stuck billing run

Invoice Actions

Action	Description
Retry Invoice	Regenerate a failed invoice
Backfill Invoices	Create missing invoices
Void Invoice	Cancel an erroneous invoice

Payment Actions

Action	Description
Retry Payment	Attempt payment collection again
Retry Payment Batch	Re-run a failed payment batch
Pause Payment Runs	Temporarily stop payment collection

Subscription Actions

Action	Description
Sync Subscription State	Align subscription status with source
Repair Entitlements	Fix entitlement mismatches

Configuration Actions

Action	Description
Review Configuration	Manual review of system settings
Rollback Configuration	Revert recent configuration changes

Escalation Actions

Action	Description
Escalate to Support	Create a support ticket
Contact Customer	Reach out to affected customers

Kai prioritizes actions based on root cause analysis. For example, if there's an authentication issue, Kai will recommend fixing credentials before trying to retry payments.

Activity Tracking

All resolution activity is recorded in the signal's Activity section:

Entry	Icon	Meaning
AI Resolution Completed	Green robot	All actions succeeded
AI Resolution Failed	Red robot	One or more actions failed
Marked as In Progress	Play icon	Agent session started
Resolved	Checkmark	Signal automatically resolved after success

When all automated actions complete successfully, Kai automatically marks the signal as "Resolved."

Permissions

Access to Kai Resolution requires specific permissions:

Permission	What It Allows
view_agent_remediation	View resolution sessions and proposed actions
update_agent_remediation	Start sessions and approve/execute actions

Administrators can configure these permissions through role management.

Plan Requirements

• Advanced Plan - AI resolution requires the Advanced plan or higher

• The agentAI plan feature must be enabled

Integration Requirements

For Kai to execute actions:

• Active Integration - The connected system must be active and healthy

• Valid Credentials - API credentials must be current and have write permissions

• Sufficient Permissions - The integration's API key must have permission for the specific operations

Best Practices

Before Starting a Session

• Review the signal details and affected entities

• Ensure your integration credentials are valid

• Understand the potential business impact

During Confirmation

• Read each proposed action carefully

• Consider the risk level for high-impact actions

• Start with low-risk actions if you're uncertain

After Execution

• Review the results for each action

• Check the affected records in your source system

• Add comments to the signal documenting what was done

Troubleshooting

"Session failed to start"

• Check that your integration is active

• Verify API credentials are still valid

• Ensure you have the required permissions

"Action execution failed"

• Review the error message for specific details

• Check if the entity still exists in the source system

• Verify the integration has write permissions

• Try the action manually in the source system

"No actions proposed"

• Kai may not have enough data to propose actions

• The issue may require manual investigation

• Check if the signal has sufficient context

Data Privacy & Security

• Kai only accesses data necessary for analysis and execution

• All actions are logged in the audit trail

• Execution results are stored securely within your tenant

• Kai respects your integration's permission boundaries