Minds treats a research method as a product contract, not a label placed on a chat. Each registered method defines its required inputs, supported sources, output types, configuration schema, pipeline stages, artifact contracts, availability, version, and calculator where applicable.
This page describes the catalog present in the Minds research system on 21 August 2026. Availability can depend on plan, configuration, and supported workflow. Customer-facing feature descriptions and plan treatment remain governed by the Minds feature catalog and applicable agreement.
| Method | Primary job | Pipeline shape | Core estimate or artifact |
|---|
| Custom research | Multi-question exploration | Collect → synthesize | Responses and qualitative evidence |
| Focused question | One scoped question | Collect → synthesize | Response and qualitative evidence |
| Questionnaire | Structured question set | Collect → synthesize | Responses, distributions, themes |
| Qualitative exploration | Reasons, objections, language | Collect → synthesize | Qualitative themes and recommendations |
| Ranked preferences | Ordered priorities | Collect → estimate → synthesize | Versioned rank estimate |
| Segment comparison | Difference between audience groups | Collect → compare → synthesize | Segment comparison estimate |
| MaxDiff | Relative priority across items | Collect → estimate → validate → synthesize | Counts, diagnostics, ranked evidence |
| Conjoint | Trade-offs across attributes and levels | Design → collect → estimate → validate → simulate → synthesize | Utilities, diagnostics, simulated shares |
| NPS | Promoter, passive, and detractor distribution | Collect → calculate → synthesize | NPS estimate |
| Top/bottom box | Scale concentration | Collect → calculate → synthesize | Top/bottom-box estimate |
| Key drivers | Relationship between drivers and outcome | Collect → calculate → synthesize | Key-driver estimate |
| TURF | Incremental reach across combinations | Collect → calculate → synthesize | TURF estimate |
| Gabor-Granger | Response across price points | Collect → calculate → synthesize | Price-response estimate |
| Van Westendorp | Price-sensitivity thresholds | Collect → calculate → synthesize | Van Westendorp estimate |
| Kano | Feature expectation categories | Collect → calculate → synthesize | Kano estimate |
Custom research, focused questions, questionnaires, and guided qualitative exploration use panel collection and evidence synthesis. They are appropriate when the research objective is to discover language, reasons, objections, themes, or candidate hypotheses.
The output can include prose, distributions, qualitative themes, and recommendations depending on the method. A qualitative result should remain traceable to the accepted responses. It should not be converted into a statistical claim merely because many synthetic audience members participated.
Ranked preferences prepares structured tasks at execution, records responses, applies the versioned ranking calculator, and produces ranked evidence. Segment comparison administers the configured measure across audience groups and applies a versioned comparison calculation before qualitative synthesis.
These methods are useful bridges between open exploration and advanced experimental designs. They still require a well-defined audience and unambiguous items.
The available MaxDiff pipeline requires a registered item set. At execution it creates controlled best/worst collection tasks, records responses, applies the maxdiff-count-v1 estimator, runs maxdiff-diagnostics-v1, and synthesizes ranked evidence.
Artifacts include the response record, estimate, diagnostics, and evidence. Diagnostics are essential because a ranked output without task-completion and response-quality checks can create false precision.
The available conjoint pipeline requires attributes, levels, and choice tasks. It uses:
conjoint-design-v1 for the design;panel-conjoint-collection-v1 for controlled choices;multinomial-logit-v1 for estimation;conjoint-validation-v1 for diagnostics;conjoint-share-simulation-v1 for configured simulations; and- evidence synthesis for interpretation.
Artifacts preserve the design, responses, estimate, diagnostics, simulated shares, and evidence. This supports reproducibility of the analysis path. It does not turn a synthetic choice into observed market demand.
NPS, top/bottom-box, key drivers, TURF, Gabor-Granger, Van Westendorp, and Kano use a shared collection-to-calculation-to-synthesis structure with method-specific configuration and calculators. The exact research question must fit the method.
For pricing, synthetic outputs are directional. They can help compare scenarios, diagnose language, and improve a live instrument, but they should not be represented as exact willingness to pay without appropriate behavioral or human evidence. Read the pricing research methods guide.
Registered methods declare which artifacts are required and which method can serve as a fallback when execution is unsafe or configuration is insufficient. A fallback must be visible. A questionnaire result should never be mislabeled as a completed conjoint or MaxDiff study.
Research outputs also depend on answer contracts and classification. Deterministic estimation starts only after accepted responses meet the method's required shape. The artifact trail makes it possible to reconcile the summary with the underlying response and calculation.
The catalog documents supported, versioned pipeline definitions in the research system. Availability can depend on plan, workspace configuration, and method readiness. It does not prove universal external validity or that synthetic inputs equal recruited-human evidence. Validation has two layers:
- implementation validation: the design and calculator behave as specified;
- evidence validation: synthetic responses track a relevant human or behavioral reference for the decision.
Both layers matter. See quantitative pipelines vs persona chat and synthetic audience validation compared.
- Use qualitative exploration when the problem or language is unclear.
- Use ranked preferences or MaxDiff when the decision is relative priority.
- Use conjoint when bundles and multi-attribute trade-offs matter.
- Use Gabor-Granger or Van Westendorp for different pricing questions.
- Use TURF when incremental reach across combinations matters.
- Use Kano for feature expectation categories.
- Use segment comparison when the decision depends on audience differences.
- Use a live human or behavioral study when the decision requires external proof.
Pair this catalog with the Minds methodology, Minds research hub, validation checklist, and synthetic respondent platform hub.
