Skip to main content

Configure AI Agent actions and guardrails

Configure the actions the Claimlane AI Agent may use and add guardrails for approvals, escalation, and safe automation.

Written by Support team

Actions define what the Claimlane AI Agent can suggest or do on a ticket. Activating only the actions you want to allow helps you keep the AI Agent within clear operational boundaries.


Before you start

  • Set up and test the relevant AI Agent rules.

  • Decide which actions should require human approval.

  • Define situations that must be handed to a person.

  • Confirm that connected systems are ready for any action that depends on an integration.

Action types

Hand-off actions

Hand-off actions pause the AI flow and prevent it from continuing until the required condition is met. Examples include:

  • Include agent

  • Wait for customer response

  • Suggest ticket type change

Use hand-off actions when a person needs to review the case, the customer must provide more information, or the request appears to belong in another workflow.


System actions

System actions let the AI Agent interact with tickets and workflows. Depending on your setup, examples can include:

  • Archive a ticket

  • Write a message

  • Forward to a supplier

  • Arrange shipping

  • Set ticket status

  • Assign a member

  • Generate a PDF

Custom actions

Custom actions connect the AI Agent to workflows configured for your organization. Add a clear purpose so the AI Agent can determine when the action is appropriate.


Configure an action

  1. Go to Manage company → AI Agent → Actions.

  2. Open the action you want to configure.

  3. Describe its purpose in plain language.

  4. State when the action should be used.

  5. State when the action must not be used.

  6. Add any approval, value, product, supplier, or risk constraints.

  7. Activate the action when it is ready for testing.

  8. Test it on representative tickets before allowing broader use.

Note: An action must be activated before the AI Agent can use it.


Example: Archive

When to use: The issue is fully resolved, the ticket is a confirmed duplicate, or the message is spam or irrelevant.

When not to use: The customer may need follow-up, an agreed action is still pending, or the issue is not fully resolved.


Recommended guardrails

  • Require human review for safety concerns, legal threats, suspected fraud, unusual high-value cases, and other sensitive situations.

  • Limit financial actions by amount, ticket type, product, supplier, or customer segment where appropriate.

  • Use handoffs whenever required evidence is missing or the AI Agent is uncertain.

  • Do not activate actions that depend on an incomplete or untested integration.

  • Keep customer communication aligned with your approved tone and policies.

  • Review declined or corrected plans and update the relevant rule or action purpose.

Roll out automation gradually

  1. Review mode: The AI Agent creates summaries and plans for a person to review.

  2. Approved execution: A person approves selected actions before they run.

  3. Targeted automation: The AI Agent completes well-defined, low-risk cases within your configured boundaries.

Move to the next stage only after testing shows consistent results for the relevant case type.


Check your result

Use test tickets to confirm that the AI Agent:

  • Selects the correct action.

  • Avoids the action when exclusion conditions apply.

  • Requests approval when required.

  • Hands the ticket to a person in sensitive or uncertain cases.

  • Produces the expected result in Claimlane and any connected system.

If an action is selected incorrectly, deactivate it while you refine its purpose, constraints, or related rules.

Did this answer your question?