Verified against ChatGPT · 2026-08-08
Build a training program from an actual skill gap, not a generic curriculum
Turns a specific, observed skill gap on a real team into a phased training program with measurable checkpoints, instead of a generic course outline that looks the same for every role.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are designing a training program to close one specific, observed skill gap on a real team — not a general onboarding curriculum, and not a list of course titles copied from an LMS catalog. TEAM AND ROLE 8-person support team, tier-1 agents handling billing disputes. OBSERVED GAP Agents can process a refund but can't de-escalate a customer who's already threatening to charge back before the ticket is even opened. EVIDENCE THE GAP IS REAL Chargeback rate on billing tickets is up 40% quarter over quarter, and QA flags de-escalation as the top miss. TIME AND BUDGET CONSTRAINTS No budget for outside vendors; can use 2 hours per week per agent for 3 weeks max. HOW SUCCESS WILL BE MEASURED Chargeback rate on billing tickets back under 15% within one quarter. PHASE 1 — DIAGNOSE Restate the gap as a behavior someone can currently do versus a behavior they need to be able to do, not as a vague trait like "needs more experience." If the evidence provided is thin or anecdotal (a single complaint, one bad quarter), say so explicitly and propose one cheap way to confirm the gap is real and shared across more than one person before building a program around it. PHASE 2 — DESIGN Build the program in three stages: a short foundational stage (what they need to know), a practice stage (a real task or simulation drawn from actual work, not a hypothetical case study), and a reinforcement stage (spaced follow-up, not a one-time session that decays within a month). For each stage, specify format, duration, and who delivers it — do not default to "a workshop" as the answer for every stage. PHASE 3 — MEASURE Define what improvement looks like in terms of the success measure given, at a specific checkpoint (30/60/90 days), and name one leading indicator that would show up before the lagging outcome metric does. WHAT NOT TO DO Do not propose a program that requires more budget or time than the stated constraints allow — if the ideal program doesn't fit, say what you cut and why, rather than silently ignoring the constraint. Do not recommend a certification or course by brand name as if you have verified it fits this exact gap; describe the type of content needed and let the requester source it. OUTPUT FORMAT 1. The gap restated as a specific behavior change. 2. A confidence note on the evidence, with a cheap validation step if evidence is thin. 3. The three-stage program as a table (stage / format / duration / owner). 4. The 30/60/90 measurement plan with one leading indicator.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Generic training-program prompts fail because they let the model default to a familiar three-part shape (intro session, workshop, quiz) regardless of what the actual gap is, since that shape is heavily represented in generic corporate-training text GPT-5.1 has seen. Forcing the gap to be restated as a specific behavior change before any design work happens blocks that shortcut, because a behavior-level restatement can't be answered with a stock curriculum — it has to reference the actual failure mode. Requiring a confidence check on the evidence matters mechanically because GPT-5.1, given a stated problem, will proceed to solve it as true by default rather than interrogating whether the premise holds; explicitly asking it to flag thin evidence and propose a validation step counteracts that eagerness-to-solve bias, which matters here because building a multi-week program around one anecdote wastes real manager and employee time. Naming a leading indicator alongside the lagging outcome metric closes a specific gap in how these prompts are usually answered: a model asked only for a success measure will name the outcome metric and stop there, but a training program that only gets checked at the 90-day lagging metric gives no early signal to adjust course, so the instruction has to explicitly ask for both. The constraint-fit rule exists because an unconstrained answer is a worse answer disguised as a better one — a program that would be great with unlimited budget but silently ignores the two-hours-a-week ceiling isn't actually usable, so the prompt forces the model to show its cuts rather than produce an aspirational plan nobody can execute.
What you get back
Gap restated: agents can execute a refund transaction but freeze or over-apologize when a customer is already hostile before a ticket exists, escalating faster than necessary. Evidence confidence: moderate — chargeback data is solid, but QA's "de-escalation miss" tag hasn't been validated by listening to actual calls; recommend pulling 10 flagged calls before finalizing. Program: Stage 1 (30 min self-paced video on de-escalation framework), Stage 2 (90 min live role-play using 3 real anonymized transcripts, led by the QA lead), Stage 3 (weekly 15-min call-review huddle for 3 weeks). 30-day leading indicator: QA de-escalation score; 90-day lagging: chargeback rate.
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — 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
