Escalation rate
How often an agent hands a conversation to a person, why a low one is not automatically good, and what to read beside it.
The share of conversations the agent handed to a human. It is the inverse of containment, and it is usually read as a cost.
Low is not automatically good
An agent that refuses to escalate has a low escalation rate and a queue of angry customers. One that escalates everything has a high rate and no automation value.
Neither extreme is the goal. The useful question is whether the conversations that escalated were the ones that should have.
The split worth having
| Customer asked | The customer requested a person. Usually correct to honour. |
|---|---|
| Agent decided | The agent recognised it was out of depth. Usually a good sign. |
| Rule forced it | Policy required a human. Not a quality signal at all. |
| Too late | Escalated after several failed attempts. The expensive kind. |
That last category is the one worth chasing. A conversation escalated on turn two costs far less than the same conversation escalated on turn nine.
An escalation rate that is flat while late escalations climb is a quality problem that the headline number is hiding.
This page as markdown