Escalation Accelerator: Page the Right Expert Prompt
Decide whether and to whom to escalate by matching the narrowed incident scope to the team or owner most likely to resolve it — with a ready-to-send escalation message — so the team stops escalating late or to the wrong people and cuts the dead time between 'we're stuck' and 'the right person is here.'
- Target user
- On-call SREs and incident commanders deciding escalations
- Difficulty
- Intermediate
- Tools
- Claude, ChatGPT, Cursor
The prompt
You are a senior incident commander who is good at the escalation call — knowing when the on-call team has hit its limit and exactly who to pull in. Help me decide. Recommend an escalation path and draft the page; I decide who actually gets paged. Inputs: - Narrowed scope so far: [FAILING SERVICE / PATH / SUSPECT AREA] - What we've tried: [INVESTIGATION + RESULTS] - Ownership map: [TEAM/SERVICE OWNERS, ON-CALL ROTATIONS, ESCALATION POLICY — PASTE WHAT YOU HAVE] - Time elapsed and trend: [HOW LONG / GETTING WORSE OR STEADY] Produce an escalation recommendation: 1. **Escalate-or-not** — a clear read on whether the current responders are still making progress or have plateaued, with the evidence (idle threads, repeated dead ends, scope beyond their expertise). 2. **Who to pull in** — match the narrowed scope to the most likely owner/team from the ownership map, with a one-line reason. Offer a primary and a backup, in case the first is unavailable. 3. **What they need on arrival** — the minimal context packet so the escalated expert can contribute in their first minute, not their tenth (current state, what's ruled out, the specific question for them). 4. **The escalation message** — a ready-to-send page/Slack draft: severity, what you need from them, the one question, and a link to the channel. 5. **The fallback path** — if the first expert can't be reached in [N] minutes, who's next. Output format: escalate? (yes/no + why) | primary + backup owner | context packet | ready-to-send message | fallback. Recommend only; do not page anyone; the human decides whom to escalate to and when.
Why this prompt works
Escalation timing is one of the most consequential and least examined levers on MTTR. Teams routinely lose long stretches in two ways: escalating too late, because nobody wants to admit they’re stuck, and escalating to the wrong person, because the ownership map lives in someone’s memory. Both extend the dead time between “we can’t crack this” and “the person who can is here.” This prompt makes the escalation call explicit — whether to escalate, to whom, and with what context — so that handoff happens fast and lands on the right expert.
The design front-loads the two decisions that actually save time. First, an honest read on whether the current responders are still progressing or have plateaued, grounded in evidence like idle threads and repeated dead ends rather than ego. Second, matching the narrowed scope to the most likely owner from the ownership map, with a backup, so a single unavailable expert doesn’t restart the search. Bundling a ready-to-send message and a minimal context packet means the escalated person contributes in their first minute instead of spending ten re-deriving the situation.
The guardrails keep the human in command of who gets paged. Ownership maps go stale, so the prompt insists the suggested owner be confirmed on-call before paging. And because escalation has a real cost — interrupting or waking someone — the prompt weighs genuine plateau against mid-progress rather than defaulting to “escalate.” The model recommends and drafts; the IC owns the decision to pull the trigger and on whom.
Related prompts
-
On-Call Context Handoff: No Re-Diagnosis Prompt
Compress a live incident's state into a tight handoff packet — confirmed facts, ruled-out hypotheses, live threads, and the next action — so the incoming responder picks up where the last one left off instead of re-diagnosing from scratch.
-
Parallel Investigation Planner Prompt
Split a live investigation across N responders into non-overlapping workstreams with clear owners and a sync point — so added hands shrink time-to-diagnose instead of duplicating each other's work.