Escalation judgment is the ability to recognise when a case should not be decided by you, or by an agent, and to pass it to the right person, at the right time, with what they need to act. It covers both directions: escalating what must go up, and not escalating what you are entitled and able to decide yourself.
Why does escalation judgment matter in AI agent work?
Agents are built to handle the routine. What reaches a person is, by design, the case that is unusual, risky or outside an agent's bounds. Much of the value of a human reviewer is deciding which of those cases they can settle and which must go further. Get it wrong one way and serious cases are decided too low down; get it wrong the other way and managers drown in cases that never needed them.
Agents are tested on the same skill. Escalation is one of the seven gates an agent must clear in Attest, Pryme's certification of agents: an agent that fails to escalate what it should, or escalates everything, does not earn its depth. Employers want the people around those agents to be at least as good at it. Why this skill is hard to hire for is argued in judgment is the AI skill employers cannot find.
How does escalation judgment work in practice?
It rests on three questions, asked of each case:
- Is this mine to decide? Check your authority: amount limits, case types, red lines.
- Do I have what I need to decide well? If evidence is missing and you cannot get it, the case may need someone who can.
- Who exactly should get it, and how fast? A vague escalation to "a manager" is slower and weaker than one to the named role, with a deadline.
The written routes help: an escalation path sets who handles what, and a matrix such as the escalation matrix for customer support agents puts it on one page. Judgment fills the gaps the matrix does not cover.
What is an example?
A support team member we will call Elif reviews agent-drafted replies for a broadband provider. One customer asks for a refund of 30 for an outage. Within her limit; she approves. Another customer's message mentions, almost in passing, that they rely on the line for a medical alarm. The agent has drafted a standard outage reply. Elif does not approve it: she escalates to the team handling vulnerable customers, marked urgent, with the customer's words quoted and the outage details attached. The first case did not need anyone else; the second could not wait for anyone less.
How is escalation judgment different from judgment under uncertainty?
Judgment under uncertainty is making a sound call when facts are incomplete, explained in what judgment under uncertainty is. Escalation judgment is a specific part of it: deciding whether the call is yours to make at all. Sometimes the right response to uncertainty is to decide; sometimes it is to escalate. Knowing which is the skill.
Which terms are related?
- Escalation path: the written route a case follows.
- Red line: an action that must never be approved, which always escalates or stops.
- Testing escalation: how teams check that agents escalate correctly, set out in how to test an AI agent's escalation behaviour.
For how employers test this and other skills, read AI skills for job seekers.
Where the Academy fits
The Operator track at Pryme Intelligence Academy includes modules on human review and on what to do when something goes wrong, and the practicum is judged to a standard of nineteen in twenty correct with no missed red line. Escalation is where that last condition is usually won or lost, so the practicum is a direct test of this judgment.
Questions people ask
How do I show escalation judgment on a CV?
Give one example of each kind: a case you escalated that turned out to matter, and a type of case you resolved yourself that used to be escalated. Both show you know where the line is.
Is it better to over-escalate when starting a new role?
Early on, leaning towards escalation is safer, but say why each time. Ask for feedback, and as you learn the limits, take on the cases you are entitled to decide.