Automate with Escalation

Escalation is the process where the AI agent forwards a conversation to a human supervisor when configured trigger conditions are detected.

In practical terms, escalation helps when the user is upset, requests a complex exception, reports a critical incident, or asks something outside policy and the AI should not decide alone.

Why Escalation Is Needed

Without escalation, an AI agent can keep responding in loops, miss emotional urgency, or fail to route sensitive requests to the right owner.

Escalation solves this by giving you:

1. Customer protection - angry users can be handled by a real person quickly.
2. Operational control - supervisors can claim, resolve, or release cases in a clear workflow.
3. Risk reduction - sensitive situations are routed to humans instead of forced AI answers.
4. System integration - webhook events can notify external tools (ticketing, CRM, Slack, etc.).

Scenario Used in This Tutorial

To keep this concrete, we use a customer support scenario where a user is angry because their order was charged twice.

Practical Setup Flow Before Configuration

Before opening escalation settings, prepare a simple implementation checklist so your first test is meaningful:

1. Define trigger phrases and boundaries - list examples that should escalate (angry, legal risk, refund dispute) and examples that should stay with AI.
2. Map urgency levels to response targets - decide what counts as low, medium, and high urgency, including expected response time.
3. Confirm supervisor coverage - make sure at least one active supervisor can claim cases during your operating hours.
4. Prepare escalation metadata - standardize reason, topic, and area labels so filtering and reporting stay consistent.
5. Plan external notification behavior - if webhook is enabled later, decide which downstream system receives events first (ticketing, CRM, chat ops).

This preparation keeps your escalation rules concise, your queue easier to triage, and your audit history easier to analyze.

How Escalation Works in qlar

At runtime, the escalation lifecycle looks like this:

1. AI detects a trigger condition from your configured escalation rules.
2. A new escalation record is created with urgency, reason, topic, and area.
3. Supervisor workspace receives a real-time update (SignalR push).
4. A supervisor claims the escalation and reviews the conversation context.
5. The supervisor resolves or releases the escalation and audit history is updated.
6. Optional webhook event is sent to external systems.

What You Will Configure Next

Use these pages in order:

1. Configure Escalation
   Start by enabling escalation and defining trigger logic, urgency mapping, and assignment defaults.
2. Receive and Manage Escalation in Real Time
   Next, verify the supervisor workflow: live updates, claim flow, context review, and resolve/release actions.
3. Send Escalation Events to Webhook
   Finally, connect escalation events to external systems so operations and compliance teams are notified automatically.

Following this order ensures each stage builds on the previous one: rule quality first, operator handling second, outbound integration last.

Prerequisites

• You already created an agent in CMS.
• You have Owner or Contributor access to configure escalation.
• At least one supervisor candidate exists in your workspace members.

Recommended before testing end-to-end:

• Prepare 2 to 3 conversation scripts that should trigger escalation and 2 scripts that should not.
• Decide who will act as supervisor during testing so claim and resolution can be validated immediately.
• If webhook is part of rollout, make sure a test endpoint is available to capture payloads.

After prerequisites are ready, continue to Configure Escalation.