Kano Model Analysis

The idea behind the Kano model

In 1984 Professor Noriaki Kano and colleagues showed that product features do not all affect satisfaction in the same way. Some features are taken for granted: customers are not delighted when they are there, but very unhappy when they are missing. Others work linearly: the better they are, the happier customers become. Others are unexpected delighters that raise satisfaction but are not missed if absent. A satisfaction rating or an importance score cannot separate these patterns, but a pair of simple questions can. The Kano model is now a standard tool in product management, UX research, service design and quality management.

The Kano questionnaire

For every feature, respondents answer two questions on the same five-point scale:

• Functional question: “How would you feel if the product had this feature?”
• Dysfunctional question: “How would you feel if the product did not have this feature?”

The answer options are: 1 = I like it, 2 = I expect it (it must be that way), 3 = I am neutral, 4 = I can live with it, 5 = I dislike it. Some studies number the scale the other way round (5 = I like it); choose the matching coding in the options.

The Kano evaluation table

Each respondent’s pair of answers is classified with the standard evaluation table. For example, “I like it” if present and “I dislike it” if absent means the feature is One-dimensional (performance) for that person. “I expect it” or “neutral” if present but “dislike” if absent means Must-be. “Like” if present and anything from “expect” to “live with” if absent means Attractive. Answers in the middle of both questions mean Indifferent. Answers that contradict the feature (dislike it when present, like it when absent) mean Reverse, and logically inconsistent pairs (like it both present and absent) are Questionable and usually signal a misunderstanding.

• Must-be (M): basic expectations, such as order tracking for an online shop. Investing beyond the basic level brings little extra satisfaction.
• One-dimensional (O): performance attributes, such as delivery speed. Satisfaction rises with performance; these are where companies compete.
• Attractive (A): delighters, such as an augmented-reality try-on. They differentiate the product, and over time they tend to become performance and then must-be features as customers get used to them.
• Indifferent (I): customers do not care; investment here is usually wasted.
• Reverse (R): some customers actively prefer not to have the feature.

What the tool calculates

For each feature, the tool counts the share of respondents in each category and assigns the feature to its most frequent category. It then computes the customer satisfaction coefficients of Berger et al. (1993):

• Better (CS+) = (A + O) / (A + O + M + I): how much satisfaction rises when the feature is present, from 0 to 1.
• Worse (CS−) = −(O + M) / (A + O + M + I): how much satisfaction falls when the feature is absent, from 0 to −1.

The better-worse diagram plots each feature with |Worse| on the horizontal axis and Better on the vertical axis. Features in the top right are one-dimensional, top left attractive, bottom right must-be, and bottom left indifferent. Because the coefficients use all four main categories, the diagram also shows features that are a mix of types.

When the two most frequent categories are close, the most-frequent rule can be misleading. Fong’s test checks whether the difference between the top two category counts is statistically significant: if |a − b| < 1.65 √((a + b)(2N − a − b) / 2N), the feature is reported as a mix (for example “M / O”), meaning the population is split and the feature may play different roles for different segments.

Worked example with the built-in data

The example simulates 160 online shoppers answering Kano questions about eight features of an e-commerce site. Order tracking and easy returns come out as must-be features, with Worse coefficients around −.8: if they are missing, customers are very dissatisfied, but their presence adds little (Better ≈ .2). Free delivery and loyalty points are one-dimensional, high on both coefficients. Live chat and AR try-on are attractive: Better ≈ .75–.80 but Worse close to zero. Gift wrapping and dark mode are indifferent. The resulting roadmap is clear: make sure tracking and returns work flawlessly, compete on delivery and loyalty, and use live chat or AR try-on as differentiators.

Running a Kano study

• Select 10 to 25 features; more makes the questionnaire long because every feature needs two questions.
• Describe features in the customer’s language and focus on the benefit rather than the technology.
• Keep the functional and dysfunctional questions close together and phrase them consistently. Randomise the order of features.
• Add an importance question if you want to prioritise within categories.
• Sample size: 50 respondents per segment gives a usable classification; 100–200 gives stable percentages.
• Arrange the data with two columns per feature, for example Feature_F and Feature_D, and select them in the same order in the two lists. The tool strips _F and _D suffixes to name the features.

Interpreting and prioritising

The classic order of priority is M > O > A > I: fix missing basic features first, because they cause dissatisfaction and churn, then invest in performance attributes, then add delighters. Within a category, use the coefficients: a must-be feature with Worse = −.85 matters more than one with −.55. A high share of questionable answers for a feature (above about 5%) suggests the description was unclear. A large reverse share signals a segment that does not want the feature; consider making it optional.

Remember that Kano categories change over time. Today’s delighter becomes tomorrow’s expectation, so repeat the study periodically, especially in fast-moving categories like apps and consumer electronics.

Limitations

• The classification describes the sample as a whole. If customer types differ, split the data by segment and run the analysis for each.
• Kano says nothing about willingness to pay or about trade-offs between features; use conjoint analysis or MaxDiff for those questions.
• Hypothetical questions about features people have never used can be hard to answer; show a short demo or description.

Reporting Kano results

Report the number of respondents, the category percentages for each feature, the final category (with mixes from Fong’s test), and the Better and Worse coefficients, usually with the better-worse diagram. Example: “Order tracking was classified as a must-be attribute (63.7% M; Better = .25, Worse = −.85), whereas live chat was attractive (73.1% A; Better = .79, Worse = −.03).”

Checklist for a Kano study

• Define features clearly: describe the benefit in plain language, one feature per question pair.
• Use the standard answer scale: like, expect, neutral, live with, dislike, identical for both questions.
• Pilot the questionnaire: check that respondents understand the dysfunctional wording; questionable answers above 5% signal confusion.
• Collect enough responses: at least 50 per segment, ideally 100 to 200 in total.
• Analyse segments separately: new and existing customers, or different user types, often classify features differently.
• Report category shares and coefficients: not only the final category, so readers can see mixed features.
• Repeat over time: delighters become expectations; rerun the study as the market matures.

Key references: Kano, Seraku, Takahashi and Tsuji (1984), Journal of the Japanese Society for Quality Control; Berger et al. (1993), Center for Quality Management Journal; Fong (1996), Center for Quality Management Journal.