Kano Analysis

Goal

A survey that has every function rated from 1 to 5 flattens all requests onto a single importance scale. It ranks the login that has to work level with the function nobody has thought of yet: the first earns nothing when it is there and costs everything when it is missing, the second is missed by no one and wins the product decision the moment it appears.

The deliverable is a classification: one row per feature, how the respondents divide across the grid's six letters and the category retained. Four letters name the categories; the other two are diagnostics, one for the contradictory answer, the other for the customer who prefers the feature absent. The Agile Extension states the limit: the technique identifies customer satisfaction only, and prioritizing a backlog calls for other factors.

Usage

When to use it

• A release scope to settle: separating what has to ship before launch from what can wait.
• Differentiation to find in a regulated market: spotting the feature competitors do not offer yet and customers do not ask for.
• A backlog full of requests all declared equally important: telling dissatisfaction avoided from satisfaction created.
• Implicit requirements never voiced: the dysfunctional question surfaces what the customer takes for granted.
• A team that disagrees about what the market wants: replacing internal opinion with a count of customer answers.

When not to use it

• No customer panel within reach: the classification would fall back on the team's opinion; settle it with a multi-criteria decision analysis on stated criteria.
• A feature imposed by regulation: customer satisfaction decides nothing here; go through business rules analysis.

Description

Kano and his co-authors published the distinction between must-be quality, one-dimensional quality and attractive quality in 1984, together with the two questions and the evaluation grid that put it to work. Charles Berger and his co-authors carried it outside Japan in 1993, fixing the vocabulary practitioners have used ever since, an arbitration rule between adjacent categories and two coefficients drawn from the counts. The Agile Extension to the BABOK Guide catalogues the technique under the names threshold, performance, excitement and indifferent and gives it its framing for the business analyst.

The two questions

Each feature is put twice to the same respondent. The functional question asks: "How do you feel if this feature is present in the product?" The dysfunctional question asks: "How do you feel if this feature is absent from the product?" Both questions offer the same five answers, in this order: I like it that way, it must be that way, I am neutral, I can live with it, I dislike it that way.

The wording of a feature has to name something the team can build or leave out, at a level where its absence can be pictured. "A faster app" describes a benefit: everyone likes it present and dislikes it absent, the feature comes out as performance and the survey has learnt nothing. "The balance displayed in under a second after the app opens" either gets built or does not, and both questions mean something.

The evaluation grid

Each pair of answers falls into one cell of the grid, which returns a letter for that respondent and that feature. The row carries the answer to the functional question, the column the answer to the dysfunctional one.

The grid is read from its edges. One cell alone combines two clear and consistent reactions, "I like it" to presence and "I dislike it" to absence: that is performance, the case where the more there is, the better. The "I dislike it" column carries threshold across its three middle cells, where the respondent treats presence as normal and absence as unacceptable. The "I like it" row carries excitement everywhere the absence causes no trouble. The block of nine cells at the centre is indifference, where neither presence nor absence draws a reaction.

The five answers are ordered without being mutually exclusive: a customer can expect a feature and enjoy it at the same time. Someone who settles on "I like it" pushes their answer towards excitement or performance, someone who settles on "it must be that way" pushes it towards threshold or indifference, on the same opinion. This ambiguity comes from the original instrument and no later processing catches it; it argues for a broad panel rather than a selected one.

The four categories

Threshold (Kano's must-be quality). The feature is indispensable before a stakeholder will consider adopting the solution. Its absence causes sharp dissatisfaction; its presence, being the acceptable minimum, adds little satisfaction. Elicitation, the work of getting needs expressed, hits its limit here: nobody thinks to ask for what they take for granted, and the dysfunctional question exists to go and fetch it.

Performance (Kano's one-dimensional quality). The more of the feature is delivered, the higher satisfaction climbs, in a roughly straight line. Speed, ease of use, capacity: these are the requests that fill a requirements workshop with nobody having to prompt them.

Excitement (attractive quality). The feature goes well beyond expectation or covers something the customer did not know was possible. With nothing comparable on the market, nobody will write a requirement describing it: it is found by observation, by competitor analysis or by trial.

Indifferent. The Agile Extension describes these features as of no value to the customer and unused, then adds that they are the ones "the customer does not want", which describes the reverse case; the same guide's grid places indifference on its nine central cells, where the respondent reacts to neither question. The guide leaves them off the graph. The satisfaction they produce does not move with the degree of fulfilment, so there is no curve to draw; the classification keeps them, because a feature a third of the panel is indifferent to is a budget not worth committing.

The two diagnostic letters

Letters Q and R judge the pair of answers itself, and the Agile Extension keeps them out of its list of four categories. Q, questionable, comes from a contradictory pair: liking both presence and absence or disliking both. Above a few per cent on one feature, a Q rate points at the wording of the question or at the respondent's attention. R, reverse, comes from the respondent who prefers the feature absent. An isolated R is noise; a majority of R is a result, that of a feature the target market refuses.

Aggregating the answers

The tally produces one letter per respondent per feature, and the most frequent category becomes the feature's classification. For close gaps, Berger and his co-authors propose an arbitration rule that keeps the most demanding category: threshold before performance, performance before excitement, excitement before indifference. It breaks the tie without excusing anyone from cross-reading the panel by age, by channel or by tenure when the gap comes down to a handful of respondents.

Berger and his co-authors also draw two coefficients from the same counts. The satisfaction coefficient is (E + P) divided by (T + P + E + I) and measures what the presence of the feature gains. The dissatisfaction coefficient is (P + T) over the same denominator, counted negatively, and measures what its absence costs. Letters Q and R stay out of the denominator. The two coefficients keep both effects, which classification by majority category reduces to one.

The graph

The three categories that vary with effort are drawn on one set of axes: degree of fulfilment on the horizontal, customer satisfaction on the vertical, the origin at the neutral point. The performance curve is a rising straight line through the origin. The threshold curve starts low, climbs fast then flattens below the neutral axis, because delivering more of an obligation stops paying early. The excitement curve starts at the neutral point, stays flat for a long stretch then rises steeply, because an attractive feature half built produces nothing. These three shapes explain why a budget split evenly across three features from different categories gives three unrelated results.

What makes the technique fail

The classification expires

The Agile Extension names drift as one of its two limits: a feature's category moves over time, as customers get used to finding it there. The guide takes Google search as its example, a novelty when it appeared and the default expectation a generation later. Kano documented that cycle in 2001: the attractive becomes one-dimensional, then must-be and sometimes indifferent once the usage disappears. The question that decides whether to run the survey again: since the classification was drawn up, has a competitor shipped one of the features classified as excitement?

A panel too small or badly chosen

The classification takes a relative majority across six possible outcomes. When the top two categories are a handful of respondents apart, the gap sits inside the sampling noise and the classification would change if the survey were run again with another group of the same size. A panel of a few dozen people gives an indication to cross-check. The choice of panel weighs as much as its size: the company's current customers read threshold features accurately, since they are the ones who suffer their absence, but they say nothing about what made the others leave.

What respondents say and what they do

The survey measures a stated reaction to a described feature, one the respondent has often never used. The gap bites hardest on excitement, where the respondent does not know the running cost of the idea being reacted to, and on indifference, where they under-report what they would use every day. Where the feature is cheap to try, an A/B test on a fraction of the users settles what the survey leaves open. Where it is expensive, the Kano classification remains the best information available before committing, provided it is read as a declaration.

After the classification

The classification feeds into a prioritization technique that carries an axis of cost, effort, risk or dependency. Grouping into MoSCoW categories is its most direct outlet, threshold features becoming Must haves. A multi-criteria decision analysis makes the Kano category one of the weighted criteria, alongside the cost of building.

AI considerations

A language model earns its place on the questionnaire and on the tally. On the questionnaire, it checks that the pairs are symmetrical: the dysfunctional question must negate exactly what the functional one puts forward. It spots the items that describe a benefit instead of a feature. It also translates: a Swiss panel answers in French, in German and sometimes in Italian, and the five answers have to carry the same gradation in all three versions; otherwise the letters shift from one language to the next. On the tally, it codes the free-text comments and groups by motive the respondents classified as reverse, whose comments are the most informative and the most laborious part of the survey to read.

What no model supplies is the answers. Asked to simulate a panel, it will produce a plausible distribution unrelated to the target market, averages drawn from the literature where the classification is empirical. The raw answers carry customer identifiers and segment data covered by the Federal Act on Data Protection (FADP). They are anonymised before anything goes to a public model, and the counts are all the tally needs.

Examples

A Swiss retail bank is preparing the next release of its mobile app. Five candidate features were put to 240 e-banking customers, two questions each.

The push notification is an obligation the dysfunctional question surfaces when no workshop would have named it. Peer-to-peer payment without leaving the app rewards effort continuously. The carbon footprint estimate is the only candidate for differentiation, with a third of the panel showing no reaction to the feature, present or absent (79 respondents out of 240). Its build cost, put at CHF 180'000, fits into no cell of the grid.

The biometric login row is the one that teaches most. Six respondents separate threshold from performance, a gap that decides nothing. Cross-read by age, the panel splits cleanly: 71 of the 118 respondents under 40 classify it as threshold, against 26 of the 122 respondents of 40 and over. The feature is an obligation for one segment and a line of progress for the other. The decision that follows is therefore taken segment by segment.

Sharing an expense on a social network is the only candidate the panel actively rejects: 101 respondents prefer it absent. What decides is not the gap to indifference, five respondents, but the level reached: two respondents in five refuse the feature. A single-question importance survey would have given this feature a low score like any other.

Berger's coefficients add the reading by both effects. For the push notification, the satisfaction coefficient is (9 + 44) / 235 = 0.23 and the dissatisfaction coefficient is -(44 + 163) / 235 = -0.88. For the carbon footprint, on a denominator of 232, the same formulas give 0.61 and -0.21. The notification gains little and ruins everything if it is missing; the carbon footprint gains a lot and its absence is barely noticed.

Visualisations

Two tables and two graphs are all the technique needs. The first table is the evaluation grid, where the position of the four categories can be read; the second is the classification. The first graph carries the three curves on one set of axes, and the second reuses those axes to show the drift: one feature marked at three moments of its life on the market, a novelty on the excitement curve, widespread on the performance line, expected on the threshold curve.

Cost

Tooling

A survey platform (LimeSurvey, Qualtrics, SurveyMonkey and their equivalents) carries the questionnaire. Two functions decide the choice: presenting the two questions of a pair separately, so that the respondent does not offset one answer against the other, and where the data is hosted, LimeSurvey installing on infrastructure the organisation controls when the answers carry customer data.

The spreadsheet is enough for the tally. The twenty-five-cell grid becomes a lookup table, a two-way lookup formula turns each pair of answers into a letter, a pivot table produces the count per feature. Berger's two coefficients take one formula each.

The backlog management tool (Jira, Azure DevOps, GitLab and their equivalents) takes the classification as a field on each item. Without that connection, the classification lives in a spreadsheet nobody opens when the release is being put together.

That leaves the panel recruitment channel: the secure e-banking message box, a banner in the app or a segmented customer file. It determines who answers and therefore what the survey can say, and the choice is documented with the classification itself.

Sources

• IIBA, Agile Extension to the BABOK Guide, §7.5 Kano Analysis: the purpose of the technique, the four categories and their definition, the two questions and the five possible answers, the evaluation grid and its legend (Table 7.5.1), the exclusion of indifferent features from the graph, the two limits stated (drift of the categories over time among them) and the classing of the technique as product management with stakeholders outside the team (Table 7.0.1).
• Noriaki Kano, Nobuhiko Seraku, Fumio Takahashi and Shin-ichi Tsuji, Attractive Quality and Must-Be Quality, Journal of the Japanese Society for Quality Control, vol. 14, no. 2, 1984: the source of the technique, which sets out the distinction between must-be quality, one-dimensional quality and attractive quality, the indifferent and reverse categories, as well as the two-question survey and the evaluation grid.
• Charles Berger, Robert Blauth, David Boger et al., Kano's Methods for Understanding Customer-Defined Quality, Center for Quality Management Journal, vol. 2, no. 4, 1993, pp. 3-36: the spread of the method outside Japan in the form practitioners use, the terminology taken up since, the arbitration rule between adjacent categories and the two satisfaction and dissatisfaction coefficients.
• Noriaki Kano, Life Cycle and Creation of Attractive Quality, Proceedings of the 4th International QMOD Conference, Linköping, 2001, pp. 18-36: the life cycle of an attractive feature, which becomes one-dimensional and then must-be as the market adopts it.
• American Society for Quality, What is the Kano Model?, Quality Resources: the model and its current terminology as taken up by a professional quality body, along with the shape of the graph plotting satisfaction against degree of fulfilment.