Our escalation policy technically allowed anyone to page anyone. In practice, junior engineers waited an average of 22 minutes before escalating to a senior leader. Here's why, and what fixed it.

The policy document said escalation was hierarchy-blind. The data said otherwise: engineers with under a year of tenure waited nearly four times longer than senior engineers to escalate a stuck incident to a VP-level responder, even when the underlying severity was identical.

Why the policy alone didn't work

A written policy doesn't erase the social cost of paging someone senior at 3am and being wrong about the severity. We interviewed a dozen junior engineers and heard a consistent theme: they weren't afraid of the technical mistake, they were afraid of being seen as someone who "couldn't handle it," and that fear scaled directly with how senior the person they'd be paging was.

What we changed

We made two changes. First, we published real (anonymized) examples of senior leaders escalating incidents that turned out to be minor, specifically to normalize false alarms. Second, we changed the framing of the escalation button itself.

escalation_ui_copy — before and after
BEFORE: "Escalate to VP on-call (use only if truly necessary)"
AFTER:  "Escalate to VP on-call — false alarms are expected and fine.
         An unnecessary page costs 2 minutes. An unescalated real
         incident costs the company far more."

leadership_commitment (published internally):
  Every VP-level responder logs their own false-alarm rate publicly,
  quarterly, as a badge of a healthy escalation culture — not a metric
  to minimize.
"Seeing our CTO's own page log with three false alarms on it, posted without any embarrassment, did more than any policy memo could have."

What moved

Median time-to-escalate for engineers under one year of tenure dropped from 22 minutes to 6 within a quarter — still slightly higher than senior engineers' 4 minutes, but the gap closed by more than 80%. More tellingly, the false-alarm rate for junior-initiated escalations barely changed at all, which told us they'd never been bad at judging severity. They'd been bad at feeling permitted to act on that judgment.

  • A hierarchy-blind escalation policy on paper doesn't remove the social cost of paging someone senior — address that cost directly.
  • Publishing leaders' own false-alarm rates, without shame, normalizes escalation for everyone below them.
  • If junior escalation is slow, check whether it's a judgment problem or a permission problem before you build more training.