What is escalation rate?

Escalation rate is the share of customer conversations that an AI agent, or a front-line team, passes to a person because it cannot or should not resolve them itself. It is often treated as a cost to drive down. That is a mistake: the right escalation rate is the one that matches the cases that need a person.

Why does escalation rate matter once agents act?

Every escalation costs staff time, so there is constant pressure to lower the rate. But an agent's escalations are also its most important safety behaviour. When a customer mentions bereavement, threatens legal action, disputes a large charge or asks for something outside the agent's role, handing over is the correct answer.

An agent that escalates too little is resolving things it should not. An agent that escalates too much is passing on work it could have done. Both are failures, and the rate alone does not tell you which you have. Our piece on how high an AI support agent's escalation rate should be works through the reasoning.

How do you judge escalation in practice?

Look at the rate by reason, not as one number. Each escalation should carry a reason code: outside role, policy requires a person, customer asked for a person, agent uncertain, vulnerability indicator, complaint. Then ask two questions of a sample:

1. Should it have escalated? Review a sample of escalated conversations to find ones the agent could have handled.
2. Should it have escalated but did not? Review a sample of resolved conversations to find ones that needed a person.

The second question matters more and is checked less. An escalation matrix for customer support agents sets out which triggers require a handover.

An example from a building society

Take a building society whose mortgage support agent escalates a fairly high share of conversations. The operations team wants it lower. Reviewing escalations by reason, they find most are for payment difficulty and changes in circumstances, which policy requires a person to handle. A smaller group are rate-switch questions the agent could answer with better sources.

The society improves the rate-switch sources and recertifies that capability. Escalations fall a little. The payment difficulty escalations stay exactly as they are, because they are the agent doing its job.

How is escalation rate different from complaint escalation?

In support metrics, escalation means a conversation moved from the agent to a person, or from a front-line person to a specialist. It is an internal handover.

Complaint escalation is a different thing: a customer taking an unresolved complaint to the firm's senior complaints team or, in many regulated sectors, to an external body. That is an outcome measure, and a rising rate signals the firm is not resolving complaints fairly. Confusing the two in reports can make a healthy handover rate look like a complaints problem, or hide a real one.

Which terms sit next to escalation rate?

• Containment rate: conversations finished without a person, broadly the other side of escalation, though abandonment sits outside both.
• Deflection rate: contacts avoided before a conversation began.
• Warm transfer: a handover where the person receives the context, so the customer does not repeat themselves.

Where Pryme fits

In Pryme, escalation is one of the seven Attest gates. A support agent that knows when to say no is replayed against past cases where a person was required, and must hand each one over to clear the gate. Escalating less is never rewarded on its own.