Prompts that sharpen an idea into a concept
"A repair service for our power tools" is an idea. A concept says for whom, why now, what must be true and what it costs to find out. An AI assistant can carry half of that thinking if you prompt it well. Here are eight prompts, in the order to run them.
One rule first: the AI sharpens, it does not invent facts. Every number, customer and competitor it produces is a placeholder until a person replaces it with a source.
When to use them
After the workshop, not in it. The idea's owner works through the prompts in one session and ends with a draft concept one-pager the sponsor can read in five minutes. For AI during the workshop itself, see what AI can and cannot do in a workshop.
Paste in the idea line, the workshop notes about it, and who argued for and against it and why. Nothing confidential about customers by name. Without the notes, the same prompt gives a generic answer about any company.
The eight prompts
Replace the square brackets and keep the rest of the wording.
1. The problem restated
Use it first, before anyone falls in love with the solution.
Here is an idea from our workshop: [idea line]. Here are the notes: [notes]. Rewrite the idea as a problem statement for a named customer: who has the problem, when, and what it costs them. Three variants with different customers. No solutions, no adjectives.
A good answer names three different people with a concrete moment of pain; if all three describe the same customer, the notes were thin.
2. The five whys on the idea
Use it when the team agrees the idea is good but cannot say why.
Take variant [number]. Ask "why would this customer want this?" and answer it. Then ask why of that answer, five times in total, one line per level. Stop and say so if an answer would be a guess.
A good answer reaches a motive the customer would recognise by level three; if it goes vague at level two, the problem statement is wrong.
3. The assumption list
Use it before you plan anything.
List everything that must be true for this idea to work, as sentences starting with "We believe that". Cover customer, organisation, economics and timing. Sort from what we know best to what we know least, and name the evidence from the notes for each item, or write "no evidence".
A good answer has at least eight sentences and marks most "no evidence"; take the list into assumption mapping with the team.
4. The sceptic
Use it before you write the concept, not after.
Argue against this idea three times: as the finance director, as a customer who tried something similar and stopped, and as the operations lead who would run it. One paragraph each, in their voice, using the notes. End each paragraph with the one question that person would ask in the decision meeting.
A good answer hurts a little and leaves three questions you cannot yet answer; they go into the one-pager as open questions.
5. The cheapest test
Use it when the riskiest assumption is clear.
The riskiest assumption is: [assumption]. Propose three tests that could prove it wrong within two weeks and under 2.000 euro. For each: what we do, with whom, what we count, and the number below which we stop. Prefer tests that need no software and no new hires.
A good answer includes one test you could start on Monday; if all three need a prototype, ask again with "no prototype".
6. The concept in two sentences
Use it after the sceptic, when you know what the idea has to withstand.
Write the concept in two sentences: what it is, for whom, and how it differs from what the customer does today. Five variants, mark the clearest. Avoid the words innovative, seamless, unique and solution.
A good answer has five variants that differ in substance, not wording; pick the one a new colleague would understand without the notes.
7. The one-pager draft
Use it when prompts 1 to 6 are done.
Fill the nine fields of a concept one-pager from everything above: problem, concept, what we know, what we assume, the first test, kill criterion and date, what success looks like, full-scale effort, the ask. Mark every number "(estimate)" and every fact without a source in the notes "(to verify)". One page.
A good answer is full of "(estimate)" and "(to verify)"; each one is a task for a person before the page goes to the sponsor.
8. The no
Use it last, when the draft feels ready.
In one sentence a new colleague would understand, write the most likely reason this idea would be rejected.
A good answer names something you can still fix; if not, put the sentence at the top of the one-pager and let the sponsor decide with open eyes.
Run prompt 4 before prompt 6. A concept written before it has been attacked reads like a sales text; written after, it already answers the first three objections.
Do not paste customer names, prices from contracts or personal data into a public AI assistant. Use the tool your organisation has approved, or anonymise the notes first. When in doubt, check with your data protection officer.
When the AI output looks better than it is
- Letting the AI invent the customer. Prompt 1 will happily describe a "busy urban professional" nobody has met. If the notes name no customer, the honest answer is "we do not know yet".
- Accepting the first variant. The first version is the average of everything the model has read. Ask for five and choose.
- Prompts without the workshop context. Without the notes, every idea becomes the same idea. Tools like IdeaHone keep the context next to the idea, ready to paste.
- Treating the one-pager draft as finished. It is a set of placeholders in the shape of a decision document. Replace every "(estimate)" with a source or an honest "we do not know", then send it.
AI that suggests, people who decide
IdeaHone turns what is said into cards, scores ideas against your criteria with written reasons and drafts the summary. Every suggestion can be edited; no decision is made by a machine.
Can AI help turn an idea into a concept?
Yes, as a sparring partner: it restates the problem, lists assumptions, argues against the idea and drafts a one-pager. It cannot supply facts about your customers; every number it produces is a placeholder until a person checks it.
What should I not paste into an AI assistant?
Customer names, contract prices, personal data and anything under an NDA. Use the tool your organisation has approved or anonymise the notes, and check with your data protection officer if unsure.
How do you prompt an AI to challenge an idea?
Give it a role and the workshop notes, ask it to argue against the idea in that voice, and have it end with the question that person would ask in the decision meeting. Run it before you write the concept.
