Test the idea before you build it
Put a product concept in front of simulated buyers, sceptics, and competitors — and find out which objection kills it while changing it is still cheap.
MiroFish tests a concept by simulating the reaction to it rather than scoring its quality. You describe the idea, who it is for, and what it would replace; the engine models prospective buyers, incumbent users, and competitors as agents that evaluate, object, and influence each other. The report returns the dominant objection, which audience finds it compelling and which is indifferent, whether it reads as a new category or as a worse version of something familiar, and the single change most likely to improve reception. It estimates reaction, not demand — only real buyers measure demand.
How MiroFish simulates it
Concept testing is most valuable early, so the pipeline is tuned to surface objections rather than validate enthusiasm:
- 1
Describe the concept and its incumbent
What it does, who for, and critically what it displaces. Concepts are judged against whatever people use today, including a spreadsheet or doing nothing, and a concept described without its incumbent gets an unrealistically warm reception.
- 2
Model buyers, users, and sceptics separately
The person who pays, the person who uses it daily, and the person who has seen three similar tools fail all evaluate differently. Simulating them as one audience produces a flattering average that hides the objection that matters.
- 3
Objection rounds
Agents raise objections and react to each other’s, so concerns compound and rank themselves. The objection that spreads is the one you will actually face in sales calls, and it is not always the most rational one.
- 4
Reception and objection report
The dominant objection with an estimated likelihood, which segment finds it compelling, whether it is understood as novel or derivative, the comparison people reach for, and the highest-leverage change to the concept.
A worked example
A startup proposes an AI tool that writes and maintains API documentation automatically, aimed at engineering teams of 10–50 who currently write docs by hand.
What you give it
- Concept
- Auto-generated, self-maintaining API docs
- Audience
- Eng teams of 10–50
- Displaces
- Manual docs, often neglected
- Price idea
- $99/mo per team
What the report estimates
- Dominant objection
- "Generated docs will be subtly wrong"
- Most compelled
- Eng managers, not IC engineers
- Read as
- Derivative unless accuracy is proven
- Highest-leverage change
- Lead with verification, not generation
Illustrative output. Simulated interest is not demand — a concept that tests well can still fail to sell.
What it can’t do (honest limits)
- —Simulated enthusiasm is not demand. Agents have no budget, no procurement process, and nothing at stake in adopting your idea.
- —It is biased toward articulable objections. Concepts that fail for inarticulate reasons — it feels wrong, nobody has time — are harder for it to catch.
- —It cannot tell you whether the concept is technically feasible or what it would cost to build.
- —Use it to kill weak concepts and sharpen strong ones, not to green-light a build. Talking to ten real prospects still beats it.
Questions people ask
Can AI validate a product idea before building it?
It can surface the objections a product idea will face, which is the useful half of validation. MiroFish simulates buyers, users, and sceptics evaluating the concept and reports the dominant objection, who finds it compelling, and what change most improves reception. It cannot validate demand — simulated agents have no budget and nothing at stake, so a warm simulated reception is not evidence anyone will pay.
What is the biggest objection to my product concept?
The report ranks objections by how strongly they spread among simulated evaluators, which is usually more informative than how logical they are. The objection that dominates a comment thread is the one your sales team will keep hearing, and knowing it early lets you either address it in the product or lead with it in the positioning.
How is this different from a concept test with real users?
A real concept test gives you genuine reactions from a small sample, slowly and expensively. Simulation gives you a broader range of predicted objections in minutes, cheaply, with no sample at all. The sensible order is to simulate first, use the output to write better questions, then run the real test — you will get more from your prospect interviews for having done it.
Will my idea be seen as innovative or derivative?
The report estimates which existing product people compare it to, which is what determines that perception. Being read as derivative is rarely about the idea and usually about the framing — a concept described in the incumbent’s language invites comparison on the incumbent’s terms, and the simulation shows you when that is happening.
Can I test several concepts at once?
Test them as separate runs and compare. Running several concepts in one simulation encourages the model to rank them against each other rather than evaluate each against its real alternative, which is what actually determines whether it succeeds.
Should I include the price in a concept test?
Yes. A concept without a price gets evaluated on interest alone, which is the easiest bar to clear and the least informative. Including a price changes which objections surface — often from "is this useful" to "is this worth more than what we do now", which is the real question.
What if the simulation says my concept is bad?
Read the objections rather than the verdict. Most concepts that test poorly are fixable by reframing — leading with a different benefit, naming a different incumbent, targeting the segment that was compelled rather than the one that was not. The report identifies the highest-leverage change for exactly this reason.
Does a good simulated reception mean the product will sell?
No, and this is the limitation worth taking seriously. Simulation tests whether an idea is comprehensible and defensible against objections; selling depends on urgency, budget, timing, and trust, none of which a simulated agent has. Treat a good result as permission to go talk to real prospects, not as validation.
Go deeper on the method: read the full guide on the blog →
Run it on your own scenario
No install, no API keys. Monthly plans from $2.99 — results in about 30 seconds, cancel anytime.
Start predicting