Verified against ChatGPT · 2026-08-12
Turn ad hoc handling of a recurring issue into an SOP agents will actually follow instead of ignore
Documents how a recurring support issue should be handled as a real standard operating procedure, built from how it's actually being handled today and where that breaks down, rather than an idealized process nobody currently follows.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Write a standard operating procedure for a support issue that currently gets handled inconsistently, ad hoc, by whoever picks up the ticket. ISSUE TYPE Customers requesting a refund after a subscription auto-renewed despite them believing they had cancelled. CURRENT AD HOC HANDLING Some agents issue a full refund immediately, others check renewal history first and only refund if the cancellation attempt is confirmed in the logs, and a few just forward it to a manager every time. TOOLS AVAILABLE Access to the billing dashboard showing renewal and cancellation-attempt logs, and a refund tool with a $200 auto-approval limit. ESCALATION THRESHOLD Refund requests over $200, or any case where the cancellation-attempt log is ambiguous or missing entirely. AUDIENCE FOR SOP New Tier 1 agents within their first 60 days, who may not yet be familiar with reading the renewal log. PHASE 1 — WHAT'S ACTUALLY HAPPENING NOW Before writing the ideal process, state what's actually going wrong with the current ad hoc handling — where does it vary from agent to agent, and what does that inconsistency actually cost (repeat contacts, wrong resolutions, escalations that shouldn't have needed to happen). An SOP that doesn't first name the real failure of the current approach tends to get written as generic best practice that doesn't specifically fix what's broken here. PHASE 2 — THE STEPS Write the SOP as a sequence of concrete steps using only the tools actually available — do not reference a tool, dashboard, or macro that wasn't listed, since an SOP that assumes access nobody actually has will be abandoned the first time someone tries to follow it literally. Each step should state not just the action but the specific condition that tells the agent to move to the next step versus stop and escalate. PHASE 3 — THE ESCALATION LINE State precisely, using the given threshold, the exact condition at which this stops being something a frontline agent should keep working and becomes something that gets escalated — a vague "escalate if needed" instruction leaves the actual judgment call exactly as inconsistent as it was before this SOP existed. PHASE 4 — WHAT THIS SOP DOES NOT COVER State explicitly what variation of this issue falls outside this SOP's scope, so an agent doesn't try to force-fit a genuinely different problem into these steps just because it looks superficially similar. OUTPUT FORMAT 1. What's currently going wrong (Phase 1). 2. The numbered SOP steps with per-step escalate/continue conditions. 3. The explicit escalation threshold restated as a single clear rule. 4. What's explicitly out of scope.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Requiring Phase 1 to name what's actually going wrong with current handling, before any steps get written, matters because GPT-5.1 asked directly for an SOP will default to a generic best-practice process for the issue type named, drawing on common patterns for "refund SOP" or "escalation SOP" broadly rather than the specific inconsistency described — grounding it in the real, stated failure (some agents refund immediately, others check logs, others just forward everything) forces the SOP to actually resolve that specific variance instead of producing a plausible-sounding process that happens to leave the real inconsistency untouched. The instruction to use only the listed tools, and nothing else, addresses a reliability gap specific to process documentation: a model will often include a step that assumes a capability (an automated flag, a dashboard filter) that sounds like it should exist for this kind of workflow but wasn't actually confirmed as available, and the first agent who tries to follow that step literally and can't finds the whole document less trustworthy. Making the escalation threshold a single explicit, checkable rule rather than "escalate when appropriate" is the difference between actually standardizing behavior and just restating the current ambiguity in a slightly more formal document — the entire reason ad hoc handling was inconsistent in the first place was that the escalation judgment call was left to individual discretion, so an SOP that doesn't convert that into an explicit rule hasn't actually solved the stated problem. Naming what falls outside scope prevents the common failure of agents stretching a documented procedure to cover an adjacent situation it wasn't built for, simply because it's the only documented process that looks related.
What you get back
What's going wrong: refund decisions currently depend entirely on which agent picks up the ticket, meaning identical cases get different outcomes, and low-value tickets get needlessly forwarded to managers while some over-limit refunds get processed without proper review. Steps: 1) Pull the renewal and cancellation-attempt log for the account — if a cancellation attempt is clearly logged before the renewal date, proceed to step 2; if the log is ambiguous or missing, escalate immediately. 2) If the refund amount is under $200, process it directly using the refund tool. 3) If over $200, prepare a summary of the log findings and route to a manager rather than processing directly. Escalation rule: any refund over $200, or any ambiguous/missing cancellation log, goes to a manager — no exceptions based on how firmly the customer is asking. Out of scope: disputes where the customer claims they never signed up in the first place — that's a separate fraud-review process, not this SOP.
Verified against
ChatGPT GPT-5.1 · 2026-08-12
Changelog
- 2026-08-12 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
