How it works — and where it stops.

The five steps, in plain terms

When you start a study, your idea and decision go through a quick specificity check — a lightweight AI pass that flags vague inputs before you spend anything. From your description we infer a plausible target audience, which you then correct: markets, ages, roles, and how skeptical the panel should be.

A short structured survey is generated for your specific decision. Each simulated persona answers it independently, shaped by the audience profile you set, and can be probed with follow-up questions. Their responses are aggregated into a decision memo: a headline verdict, the numbers behind it, the strongest objection, and a recommended next test.

Why a panel beats a single answer

Ask one model one question and you get one confident answer that shifts every time you re-ask. A panel does something different: it holds the survey and the audience constant, varies the personas, and reports the distribution — how many were positive, how skeptics differed, where a segment breaks from the average. Consistency is the feature. It lets you compare two ideas, two prices or two messages on the same footing.

What this is good for

• ✓ Killing weak ideas early, before you build
• ✓ Finding which segment cares most
• ✓ Comparing messaging, framing or names
• ✓ Sniffing out a price ceiling
• ✓ Surfacing the objection you'll have to answer
• ✓ Deciding what's worth testing for real

What it is not

• ✕ A substitute for talking to real customers
• ✕ A source of true market-size numbers
• ✕ A predictor of exact conversion rates
• ✕ A reason to skip a live experiment
• ✕ Reliable for niche B2B where personas are thin
• ✕ The final word on anything you're betting the company on

How to read the confidence signals

Every memo carries a verdict and a confidence level. Treat a Go as “worth building a real test around,” a Conditional go as “works for a segment under a condition — confirm the condition,” and a No as “reshape the idea or the audience.” The percentages are directional, not precise: a 58% is meaningfully different from 30%, but not from 55%. Watch the gap between segments more than any single number.

Where the persona behaviour comes from

Personas are grounded in the audience settings you provide and any reference pages you attach for extra context. Their responses reflect broad, well-documented patterns in how people react to new offers — enthusiasm tempered by cost, trust and switching effort — not the private opinions of any real, named individual. Simulated respondents can share the blind spots of the models behind them, which is exactly why we frame the output as a starting point, not a verdict on reality.