Not All Course Feedback Is Equal: The Kano Model and Why Chasing Every Low Score Wastes Effort

The short version: course-evaluation dashboards quietly assume that satisfaction is additive and symmetric — that raising any item's score by half a point does the same amount of good, and that a 4.0 means the same thing wherever it appears. The Kano model, developed by Noriaki Kano and colleagues in 1984, shows this is false. Some attributes are must-be qualities whose absence enrages students but whose presence earns no credit; others are attractive qualities that delight when present but are not missed when absent; only a middle band behaves linearly. A 4.0 on a must-be attribute (say, fair marking turnaround) is a quiet emergency, because must-bes have to be near-perfect. A 4.0 on an attractive attribute (a guest industry speaker) is a genuine win. The mean score cannot tell these apart. Kano can.

The linear fallacy baked into your reporting

Almost every course-evaluation report treats attributes as interchangeable units of satisfaction. Ten items, each scored 1-5, often averaged into a single index or ranked into a "to-improve" list. The arithmetic only makes sense if satisfaction is linear (each point of improvement adds equal value) and symmetric (doing well and doing badly on an attribute are mirror images). Neither assumption holds for how humans actually form satisfaction judgements.

Consider two attributes. First, "the timetable was published and accurate." If it is, no student writes a glowing comment about it — they expected nothing less. If it is not, the cohort is furious and the whole module's ratings suffer. Now consider "the module included an optional talk from a practitioner." Its absence is barely noticed; its presence produces disproportionate delight. These two attributes contribute to satisfaction in opposite, non-linear ways. Averaging them into one number, or ranking them on a single scale, throws that structure away.

This is closely related to why almost every course scores around 4 out of 5: must-be attributes are usually satisfied, so they pile up near the ceiling and compress the distribution, while the genuinely differentiating attractive attributes are rarer and more variable.

What the Kano model says

Kano's insight, first published as "attractive quality and must-be quality" in 1984, was to ask about each attribute twice — once functionally ("how do you feel if the course has feature X?") and once dysfunctionally ("how do you feel if it does not?") — and to classify the attribute from the pattern of the two answers (overview of Kano in higher education, MDPI 2019). The categories:

• Must-be (basic). Expected. Presence is taken for granted; absence causes strong dissatisfaction. Examples in a course: accessible materials, an accurate timetable, marking returned within the promised window, a lecturer who turns up. You cannot win points here — you can only lose them. Must-bes must be reliably met, not optimised.
• One-dimensional (performance). The linear ones. More is straightforwardly better and less is worse. Examples: usefulness of feedback, clarity of explanations, responsiveness to questions. These are where "raise the score" advice actually works, and where investment scales with return.
• Attractive (delighters). Their presence delights; their absence is not held against you. Examples: a practitioner guest lecture, an unexpectedly personalised piece of feedback, an industry field trip. A systematic review of Kano in education found such attributes — "the interaction of the course with the industry, such as technical trips and invited speakers" — repeatedly classified as attractive needs that "increase student satisfaction by creating delight" (Kano systematic review in education, 2024).
• Indifferent. Students do not care either way. Effort here is wasted.
• Reverse. Some students actively prefer the absence of the feature — a signal to segment rather than standardise.

The practical payoff is a re-reading of your dashboard. A middling score on a must-be is the most urgent thing on the page, because must-bes are supposed to be effectively solved; a middling score on an attractive attribute is an opportunity, not a failure. Two identical 3.6s can mean "fix this before next week" and "consider adding this if you have capacity."

The kinship with Herzberg — and why it is not the whole story

Readers from a management background will recognise the shape: Kano's must-be/attractive split echoes Herzberg's distinction between hygiene factors (whose absence dissatisfies) and motivators (whose presence satisfies). The parallel is useful shorthand, but Kano is more operational: it gives you a classification procedure per attribute and a way to detect the asymmetry directly in your data, rather than a fixed list of factor types. That procedural specificity is what makes it usable inside an evaluation cycle.

The strongest counterarguments, taken seriously

"Kano categories drift, so any classification is a snapshot." This is the most important limitation. Attractive qualities migrate over time into must-be qualities as expectations rise — lecture recordings were a delighter a decade ago and are now, post-pandemic, an expectation whose absence generates complaints. Any Kano classification is valid for a cohort at a moment, not forever. The honest response is to re-classify periodically and to read migration itself as intelligence about rising student expectations, rather than to treat the categories as permanent.

"The paired-question method is burdensome and ambiguous to code." Asking two questions per attribute lengthens the survey and the standard evaluation table (functional × dysfunctional) sometimes yields "questionable" responses that are hard to classify. Where survey length is already a data-quality risk, adding a full Kano battery may cost more than it returns. A lighter alternative — inferring asymmetry from the data via penalty-reward or asymmetric-impact analysis — sidesteps some of the burden at the cost of some precision.

"Stated reactions may not predict behaviour." As with all self-report, what students say would delight or enrage them need not match how they actually respond. Kano organises stated preference structure; it does not certify that structure translates into engagement or learning. And satisfaction is not the target anyway — student ratings explain at most about 1% of the variance in actual learning (Uttl et al., 2017), so a Kano map of satisfaction attributes must always be read as one lens, never the whole evaluation.

Where a conversational platform earns its place

Kano's classic weakness is the burden and ambiguity of the double question. This is precisely where an AI-native, conversational instrument helps — not by replacing the model but by gathering its inputs more naturally.

Koji for Education operationalises the Kano logic without a rigid paired-question grid:

• Probing the "why" behind a rating. Koji's AI-moderated conversational interviews can ask, in the moment, whether a missing feature would have genuinely bothered a student or whether a present feature was an unexpected highlight — the exact functional/dysfunctional signal Kano needs, gathered conversationally rather than as a fatiguing double battery.
• Thematic clustering into hygiene vs delight. Koji's automatic thematic analysis groups open-text feedback into the attributes students actually raise, and the emotional framing of those comments helps distinguish anger-at-absence (must-be) from delight-at-presence (attractive) — an insight a sentiment percentage alone cannot deliver.
• Catching must-be failures mid-cycle. Because must-bes are dissatisfiers, discovering one at the end of term is too late. Koji's formative, mid-cycle collection surfaces a broken must-be (say, chaotic assessment logistics) while there is still time to fix it — which is also the difference between collecting feedback and closing the loop on it.
• Pairing with priorities. Kano tells you what kind of attribute you are looking at; Importance-Performance Analysis tells you which to act on first. Used together inside Koji's programme-level reporting, they turn a flat ranked list into a genuine strategy: guarantee the must-bes, invest in the linear performers, and add delighters selectively.

Teams running non-teaching research — alumni, employer, or general user studies — can apply the same conversational engine on the main Koji platform, where the must-be/attractive distinction is just as useful for product and service feedback.

The takeaway

Stop reading your evaluation dashboard as a single ladder of scores to climb. Some rungs are must-bes you must simply never drop; some are linear performers worth steady investment; a few are delighters worth adding when you can. A number cannot tell you which is which — the structure of student satisfaction can. See how Koji for Education captures that structure in students' own words.

Koji surfaces and classifies the drivers of satisfaction; deciding which delighters are worth the cost remains an academic and resourcing judgement.