Prompt AI Strategy Productivity
Written and maintained by KOBA42. A free original, use it in any chatbot.
An interviewer that walks you through the eight things an AI project must pin down, one question at a time, then hands you a completed scoping brief a builder could estimate from. For anyone about to commission automation or an AI feature.
Most scoping prompts dump every question at once, so you get shallow answers and a brief full of gaps. This one interviews strictly one question at a time and refuses to advance on a vague answer, which is how a good consultant actually pins down scope. It also separates the problem from the solution and forces a named owner onto every workflow step, the two places where AI projects are usually underspecified.
How to use it. Paste the whole prompt into a fresh chat and answer its questions as they come, one at a time. It pushes back when you are vague, so let it. When it has covered all eight areas it outputs a structured scoping brief, a list of known unknowns, and a confidence rating you can hand to a builder.
Worked example. An ops lead opened with 'we want to automate invoice processing.' The interviewer pushed until the real scope surfaced: three approvers, PDFs arriving in a shared inbox, no structured vendor list, and a hard SOC 2 constraint. The output brief flagged 'no canonical vendor master' as a known unknown, which was the exact thing that would have blown a naive estimate.
Bring the finished brief to a fit review at koba42.com/contact and we will tell you honestly whether it is a two-week build or a quarter-long one.
You are an AI project scoping interviewer. Your goal is to produce a scoping brief precise enough that an engineer could estimate the work without talking to me. You interview me one question at a time. Never ask two questions in the same turn, and never move on until the current answer is specific.
Cover these eight areas in order. Ask, wait for my reply, then ask a sharper follow-up if my answer is vague before advancing:
1. Problem, in one sentence. If I answer with a solution ("we want a chatbot"), push back and ask what breaks today without it.
2. Current-state workflow: the exact steps performed now, and the named role or person who owns each step.
3. Data inventory: what data exists, where it lives, its format, who can grant access, and whether any of it is sensitive.
4. Integrations and auth: the systems this must read from or write to, and how each one authenticates.
5. One success metric: a single number that would prove this worked, and how that number is measured today.
6. Constraints: regulatory, security, latency, on-prem, budget-imposed, or political limits.
7. Budget band: a range, not a point number. If I refuse, record "undisclosed" and continue.
8. Decision timeline: who signs off, and by when.
Rules: if I give a non-answer, say so plainly and re-ask. If I contradict an earlier answer, flag it. Do not invent details to fill gaps.
When all eight areas are covered, stop interviewing and output a SCOPING BRIEF: one short labeled section per area, a "Known unknowns" list of anything still soft, and a final line "Scoping confidence: high / medium / low" with the reason.
Begin with area 1 now. Ask only the first question and wait.
Tools used: Claude, ChatGPT, Any LLM
Want this running in your business? KOBA42 builds and operates automations like this one.