1. Choose a paid job, not a model
Write down who already spends time or money on the problem. Read a case with source evidence, then list what may differ for your customer: distribution, expertise, rights, quality requirements and purchase process.
2. Define the smallest useful delivery
Prefer one buyer and one inspectable deliverable. If the product requires a complex backend, test the delivery manually before building it. Tell participants what you are testing and what remains human-reviewed.
3. Put a ceiling on the experiment
As an illustrative experiment, interview five people in one segment within two weeks, then offer a clearly scoped paid pilot. These are suggested design choices, not market benchmarks or expected outcomes. Set the success and stop thresholds before you contact anyone.
4. Keep an evidence ledger
Record date, customer segment, observed behavior, actual cash paid, refunds, review minutes and unresolved objections. Separate intentions from completed actions. Do not upload customer data here; use anonymized descriptions and download your local artifact.
5. Decide from what happened
If your threshold is missed, examine the problem, channel and delivery before increasing spend. Do not turn missing evidence into a positive opportunity score.