How to use it
One idea. One short test cycle.
Work through each stage in order. Mark an exit gate only when the evidence supports it. Run no more than one or two ideas at once.
It is to decide, cheaply and quickly, whether the next level of effort is justified.
Frame the idea
Name the real job.
Do not start with a feature list. Start with a specific person, a repeated job, the measurable result they need and the frustration they already tolerate.
Framing statement
For [specific person] who needs to [job they are already trying to do], this helps them achieve [measurable outcome] without [current frustration or cost].
Research existing pain
Find behaviour, not agreement.
Use AI to accelerate discovery, never as the judge. Look for traceable evidence that people are already struggling, improvising or spending.
Evidence sources and search patterns
"[competitor]" review "wish" · "too expensive"site:reddit.com "[job]" "how do you"site:linkedin.com/posts "[problem phrase]""[job]" "responsibilities"Search the category; read low-star reviewsType the job, problem and alternatives"[competitor]" community · helpAI research prompt
Act as a market-research analyst. For the audience [persona] and job [job], identify repeated complaints, current workarounds, paid alternatives and the exact language people use. Separate direct evidence from inference. Return a table with source type, insight, problem phrase, willingness-to-pay clue, and community or channel to investigate. Do not invent quotes or sources.Locate communities
Go where help is already exchanged.
You do not need a large audience. Find concentrated places where your specific people already share recommendations, workarounds and complaints.
Community scorecard
Score each criterion from 0–2. Prioritise communities at 7/10 or above, then read the rules before participating.
No community has reached the 7/10 priority threshold yet.
Pilot post template
Three-question form
- When did you last deal with this problem?
- What did you do instead, and what did that cost in time, money or risk?
- If a tool solved it reliably, what would need to be true for you to try or pay for it?
Make a testable offer
One audience. One outcome. One action.
Create a single page, not a full product. Make the offer concrete enough that the right person can accept or reject it without decoding vague product language.
Offer preview
[Audience]: achieve [outcome] without [current friction]. From [price]. [Commitment action].
Avoid “revolutionary”, “AI-powered” and “all-in-one” unless the words clarify something concrete. Never ask only, “Would you use this?”
Test for commitment
Measure what people give up.
Interest is not equal to intent. Signals become more useful as prospects invest time, data, trust or money.
Pre-commit the decision threshold
Set the gate before promotion to prevent excitement from changing the rule after the results arrive.
Run a concierge pilot
Deliver before you automate.
Provide the promised result manually or with lightweight tools. The pilot reveals the real workflow, exceptions and value drivers before software hardens your assumptions.
Written feedback
Decide deliberately
Make the evidence decide.
End every short test cycle with an explicit outcome. Preserve rejected ideas and the evidence behind them so you do not repeat the same assumptions later.
Reusable summary
One-page worksheet
A compact record for comparing ideas or handing the evidence to your future self.
Weekly cadence
Guardrails
- Do not represent a prototype as a finished product.
- Follow community rules and never scrape private spaces.
- Treat AI-generated claims as hypotheses until evidence supports them.
- Do not over-value compliments from people outside the target audience.
- Keep an evidence log so decisions remain reversible and explainable.