High enough to cover every contact the agent should not finish, and no higher. There is no universal target. The right rate depends on your mix of contact reasons, what the agent is allowed to do, and how many of your customers need a person. Judge escalations by whether each was right, and plan staffing from the rate.
Why is a lower rate not automatically better?
Because the easiest way to lower an escalation rate is to stop escalating things that should go to a person. An agent pushed to escalate less will answer questions it should hand over, refuse requests a person could have granted, and make reaching a human harder. Each of those shows up later as recontacts, complaints and, in regulated businesses, conduct risk.
The same trap appears with containment, the mirror image of escalation. The argument against chasing it is made in containment rate is the wrong goal for a support agent. A definition of the measure itself is in what escalation rate is.
What drives the rate up or down?
Contact mix. Order tracking and password resets rarely need a person. Complaints, bereavements, fraud reports and disputes almost always do. An agent handling mostly the first kind will escalate rarely; one handling a mix will escalate often, and should.
Depth. An agent at draft depth sends everything for review, so "escalation" in the usual sense barely applies. An agent acting within limits escalates whatever falls above them. Raise the limits and the rate falls, for good reasons, if the agent earned the higher depth on evidence.
Customer base. A service with many older customers, customers in financial difficulty, or customers writing in a second language will see more legitimate handovers.
Knowledge coverage. Every gap in the knowledge base produces escalations. That is the right behaviour, and the fix is to fill the gap, not to stop the escalation.
Requests for a person. Some customers will ask for a human whatever the agent does. They should get one.
How should you plan for it?
| Step | What it gives you |
|---|---|
| Estimate expected escalations per contact reason from your own history | A baseline you can defend |
| Write the triggers that must escalate in an escalation matrix | A clear standard for "right" |
| Audit a sample of escalations and non-escalations each week | Whether the agent escalates the right things |
| Track the rate per contact reason, not overall | Changes in mix stop hiding changes in behaviour |
| Staff people for the expected rate, with headroom | Handovers that reach someone quickly |
The audit matters most. Two errors are possible: escalating what the agent should have handled, and handling what it should have escalated. The second is far more costly. If the audit finds even a few missed escalations of the serious kind, the agent should lose depth on that capability, whatever the overall rate looks like.
An illustrative example
Take an energy supplier whose agent handles meter readings, bill explanations, payment questions and moving home.
In its first month, the agent escalates about one conversation in three. Management asks why it is so high. The audit shows most escalations are right: customers in arrears asking about payments, which the supplier's own rules send to a person because affordability must be discussed. A smaller group are knowledge gaps about moving home, which are fixed. A handful are customers asking for a person straight away.
The audit also finds two conversations the agent should have escalated and did not: customers who mentioned medical equipment that needs power. Those are the serious finding. The trigger list gains the phrases, the cases join the test set, and the moving-home fixes bring the rate down a little. The arrears handovers stay, because they are correct.
Which measures sit alongside the rate?
Escalation rate on its own says little. Pair it with resolution that holds, meaning no recontact on the same issue, and with the quality of each handover. The difference between resolution and escalation is explained in what resolution rate is, and a fuller set of measures is in measuring a support agent beyond resolution rate.
Where Pryme fits
In Pryme, every escalation is a held verdict with the bound that triggered it recorded, so you can see whether handovers come from limits, knowledge gaps, distress triggers or customer requests. Attest's escalation gate checks the agent hands over every case it should before a capability is certified. More is in the AI support agent that knows when to say no.
Questions people ask
Should we set a target escalation rate for the agent?
Set targets for correct escalation, measured by audit, not for the rate. A rate target rewards the wrong behaviour.
Does a rising rate mean the agent is getting worse?
Not necessarily. Check whether the contact mix shifted, a policy changed, or a new trigger was added before drawing conclusions.
What is a missed escalation?
A conversation the agent completed that the escalation matrix says should have gone to a person. It is the most important error to find.