AI Product Concept Testing Before You Build

What decision should a concept test help you make?

A concept is more than a feature list. It connects a specific customer, a situation they struggle with and a reason to change their current behavior. Before testing, write one decision: should we build a prototype, narrow the audience, change the promise or stop investigating this idea?

For an illustrative example, imagine a founder considering an AI reporting assistant for independent agencies. The initial pitch is “automated reporting.” That leaves important questions unanswered: who checks the numbers, whether clients need custom explanations and what happens when a connected source is incomplete. A concept test should expose these questions before a polished demo conceals them.

Build a concept-testing packet from actual evidence

Prepare a short concept description and a separate evidence file. The concept should explain the user, triggering situation, proposed outcome, expected effort and alternative they use today. Give competing concepts comparable detail; otherwise the longer description may appear more persuasive simply because it contains more information.

For evidence, use permitted interview excerpts, recurring support questions and descriptions of existing workflows. Remove personal identifiers and confidential client details. Label each statement as an observation, an interpretation or an assumption. A founder’s belief that agencies dislike spreadsheets belongs in the third category until the research supports it.

• Customer: who experiences the problem and who pays?
• Trigger: what event makes the problem urgent?
• Alternative: how is the task completed today?
• Trade-off: what effort, cost or risk does switching introduce?
• Unknown: what could make the whole concept unnecessary?

Compare concepts without changing every variable

One option might generate a complete report automatically. Another might prepare a draft that an account manager approves. Hold the target customer and reporting task constant while comparing those two operating models. If you also change price, audience and promised outcome, you will not know which difference explains a reaction.

Mirror is a multi-agent simulation platform for exploring scenarios from supplied material. Use the evidence packet as input and review the generated entities and relationships before running the scenario. Explore how a report preparer, agency owner and client approver might respond to the same proposal. These simulated roles are analytical perspectives, not recruited research participants.

A brief for testing an agency reporting concept

Turn reactions into a decision record

Keep a simple worksheet outside Mirror with five columns: concept, possible benefit, strongest objection, supporting evidence and next test. An objection such as “I need to verify the source” suggests a prototype task around traceability. “I already have a template” suggests investigating whether preparation time is important enough to justify switching.

Do not count simulated supporters and convert the result into a percentage of buyers. Several agents may share assumptions from the same model and input material. Instead, look for arguments that change your next research step. The useful output may be a narrower concept, such as draft preparation for recurring monthly reports, rather than an overall winning idea.

Check the concept with real behavior

Ask potential customers to describe their most recent reporting cycle before presenting your concept. Then show a neutral prototype and give them a realistic task. Observe whether they can complete it, what they verify and what existing tool they would still need. Avoid asking only whether the idea sounds useful.

Before recruiting participants, define the evidence your team needs to proceed. A paid pilot, a completed workflow or access to a decision-maker is a different signal from a polite expression of interest. Record disagreements and revise the concept brief accordingly. Your research should be able to overturn the simulation, not merely decorate it.

Prepare your first concept in Mirror

Bring one concept, one clearly defined customer role and the evidence you already have to Mirror. Run a focused scenario to identify the most consequential adoption questions, then take those questions into customer research. Start with a decision small enough that your team can act on the result this week.

Common questions