Methodology

The platform is opinionated about research method on purpose. These are the frameworks it applies, and where a human stays in the loop.

Problem structuring

Briefs are framed with TOSCA — Trouble, Owner, Success criteria, Constraints, Actors — then decomposed into a MECE issue tree or an issue pyramid. A null-hypothesis branch is always generated, so a favourable answer is never the only one tested.

Ready — 75 and aboveNeeds refinement — 50 to 74Too vague — below 50

Measurement & scaling

Measurement level (nominal, ordinal, interval, ratio) is enforced before analysis is chosen. Comparative scaling — paired comparison, rank order, constant sum — and non-comparative scaling — continuous rating, Likert 3/5/7/10-point, semantic differential, Stapel — are offered with an explicit recommendation and an override at each step.

Questionnaire design intelligence

Every question is audited against Tourangeau's four cognitive stages — comprehension, retrieval, judgment, response formulation — plus eleven wording dimensions covering double-barrelled, leading and ambiguous phrasing. Sensitive topics get mitigation techniques with a plain-language rationale.

AI governance

Every AI-generated output carries the model and version that produced it, an explanation citing the methodological principle behind it, and an override control. Overrides require a reason and are logged. Reports carry a disclosure label on every page:

AI-Generated Analysis — Reviewed by a Research Lead

Simulated persona output and scenario interpretations are labelled separately, because neither is a research finding:

⚠ AI-Simulated Responses — Not Real Consumers

⚠ Scenario Interpretation — Not a Research Finding