Kano feature classification
What do people expect from your features?
Compare reactions to a feature’s presence and absence. See the complete Kano response profile, then check how consistently its leading category appears in resampled data.
Start with paired responses.
Import a study or explore the fictional example. Confirm your columns and answer coding before calculating. This release supports unweighted, independent respondents; it does not analyze repeated or clustered observations.
Ask about one feature in two states.
Describe the same feature when present and absent, with enough context to judge it. Avoid combining several benefits, adding a price change to just one state, or suggesting that one answer is preferable.
Pilot the wording and response options. A separate “don’t know enough to judge” option can be mapped as nonclassifiable. The five conventional choices are categories, not a numeric satisfaction scale.
Discuss your feature researchIllustrative question pair
Present: If the app lets you save a report for offline reading, how do you feel?
Absent: If the app does not let you save a report for offline reading, how do you feel?
Use the same choices for each: Like; Must / expect; Neutral; Can live with; Dislike.
Record the exact fielded wording and coding in the analyzer’s optional report fields. Feature names alone do not preserve the questionnaire.
A disclosed matrix, with limits on interpretation.
The conventional matrix below classifies each complete presence/absence pair. All six categories remain in the distribution and modal comparison. Exact ties stay tied; a leading share of 50% or less is described as mixed responses. That wording is descriptive, not a statistical test.
Attractive (A), One-dimensional (O), Must-be (M), and Indifferent (I) describe conventional reaction patterns. Reverse (R) can reflect a preference for the opposite state. Questionable (Q) combinations call for checking wording, understanding, and coding; they do not by themselves prove that a respondent failed a quality check.
| Present ↓ / absent → | Like | Must / expect | Neutral | Can live with | Dislike |
|---|---|---|---|---|---|
| Like | Q | A | A | A | O |
| Must / expect | R | I | I | I | M |
| Neutral | R | I | I | I | M |
| Can live with | R | I | I | I | M |
| Dislike | R | R | R | R | Q |
Better = (A + O) / (A + O + M + I).
Worse = −(O + M) / (A + O + M + I).
Reverse and Questionable are excluded from these two denominators only. With no A/O/M/I responses, both coefficients are undefined. The map uses the positive magnitude of Worse; reported coefficients retain the negative sign.
How are stability and coefficient ranges calculated?
For the selected audience, 1,000 bootstrap samples draw the original number of respondents with replacement. Each draw carries all that respondent’s feature pairs, including missing values. Unique modal wins, ties, and samples with no classifiable pairs are counted separately.
Coefficient ranges use the 2.5th and 97.5th percentiles with linear interpolation. Replications with zero coefficient bases are counted and omitted; a range is withheld unless at least 950 are defined. Seed 20260929 makes repeated runs with the same row order reproducible. With 1,000 replications, a win frequency near 50% has about 1.6 percentage points of Monte Carlo standard error.
These are pointwise empirical resampling ranges, not a population margin of error, simultaneous confidence guarantee, or validation of Kano theory. A sample containing only one pattern can produce a zero-width range and 100% stability. Sparse data and nonprobability recruitment still need research judgment. Audience tables are descriptive; no significance tests are performed.
How much should I trust a category?
Chapman and Callegaro’s 2022 experiments raised concerns about response reliability, dimensionality, and category stability. Their smartphone study found more consistent classifications around 200 respondents, but that is not a universal sample requirement or proof of validity.
Read the base, missingness, full distribution, top-two gap, and bootstrap frequencies together. More data cannot by itself repair unclear wording, selection bias, or an unsuitable measurement model. Category membership is not a development priority or a forecast of customer behavior.
What data is retained, and what is outside this release?
The analyzer processes records in a browser worker. It does not upload responses, store them in browser storage, or add them to contact links. Clear data removes the current file, paste, mappings, report, and results from the analyzer. Downloaded files remain on your device.
Weights, repeated observations, cluster sampling, longitudinal change, formal subgroup tests, importance/cost overlays, and alternative Kano scoring methods are not supported. XLSX reads one chosen sheet at a time, with size/complexity limits and no formula evaluation. Use CSV or a values-only workbook if validation fails.
Sources reviewed September 29, 2026
Match the next method to your decision.
Relative priorities
MaxDiff asks which items people prefer relative to others. Kano asks about presence and absence; the results are not interchangeable.
Plan a MaxDiff studyFeature and price trade-offs
Conjoint evaluates choices among configurations. Use a suitable design when features, price, and competing offers interact.
Plan a conjoint questionnaireChoose the research approach
Discuss the product decision, audience, and evidence you need. The method selector explains which approaches fit different research questions.
Choose a research method