Escalation
Escalation is the process of moving a request, case, or task to another level of expertise, authority, or human involvement when the current workflow cannot or should not complete it.
Escalation is an important part of both traditional support and AI-powered service workflows.
What Is Escalation?
Escalation happens when the current support path is no longer sufficient.
The request may move to:
- A specialist
- A supervisor
- Another department
- A more capable AI agent
- A human support team
- A higher-authority workflow
The purpose is to place the request with the resource most capable of resolving it.
Common Reasons for Escalation
Complexity
The issue requires more specialized knowledge.
Policy Exception
The customer requests something outside the normal policy.
High-Risk Action
The action requires additional review.
Customer Request
The customer asks to speak with a person.
Low Confidence
The AI system does not have enough evidence to continue.
Tool Failure
A required system or integration is unavailable.
Sensitive Issue
The situation requires human judgment or empathy.
Repeated Failure
The workflow has attempted to resolve the issue unsuccessfully.
Types of Escalation
Functional Escalation
The issue moves to someone with more specialized expertise.
Hierarchical Escalation
The issue moves to someone with more authority.
Technical Escalation
The case moves to a technical specialist.
Human Escalation
Automation transfers the request to a person.
Agent-to-Agent Escalation
A general AI agent transfers the task to a specialized AI agent.
Escalation vs. Human Handoff
Human Handoff is one form of escalation.
Escalation describes the decision to move the request.
Human handoff describes the actual transfer from automation to a person.
A request can also be escalated to another automated workflow or specialized agent.
Escalation vs. Routing
Agent Routing determines where a request should go.
Escalation is usually triggered when the current route is not sufficient.
Routing can happen at the beginning.
Escalation usually happens after the workflow has already started.
Escalation in AI Systems
An AI system should not try to resolve every possible request.
Useful escalation rules can include:
- Unsupported intent
- Missing information
- Low-confidence answer
- Restricted action
- Policy exception
- Sensitive topic
- Customer preference
These rules act as part of the system's guardrails.
Escalation and AI Guardrails
AI Guardrails can define when escalation is mandatory.
For example:
- High-value refunds require approval.
- Legal questions go to a person.
- Unsupported account actions are blocked.
- Low-confidence answers trigger review.
This helps keep the AI within its intended role.
Escalation and Customer Experience
Poor escalation creates friction.
Common problems include:
- Repeating information
- Being transferred to the wrong team
- Long delays
- Losing conversation history
- Repeated automation attempts
A good escalation preserves relevant context.
What Should Transfer During Escalation?
Useful context can include:
- Conversation history
- Customer details
- Identified intent
- Information already collected
- Actions already attempted
- Tool results
- Search results
- Reason for escalation
This helps the next person or system continue efficiently.
Escalation Rate
Escalation Rate measures how frequently requests move from the current automated or frontline workflow to another level.
A high escalation rate may indicate:
- Complex request mix
- Limited automation scope
- Poor knowledge access
- Weak routing
- Tool limitations
It is not automatically negative.
Escalation in AskHandle
AskHandle can route conversations between workflows and transfer requests through Human Handoff when a person is needed.
Different nodes can handle distinct responsibilities, allowing escalation rules to remain explicit rather than hidden inside a single model prompt.