Abstract
The Kano model has been widely applied in multiple disciplines, particularly in product or service development and optimization. Although the original categorization framework remains the most applied, modifications and new approaches have recently been developed. Yet, most of the hospitality and tourism studies continue to use the original Kano method when newer and superior methods are available. The new categorization methods also have flaws, which academics and practitioners should be aware of. The present study uses a systematic literature review approach to provide a comprehensive synthesis and critical analysis of the existing Kano model categorization methods structuring them in groups based on categorization procedures and pointing to the strengths, weaknesses, and fit of each method. The manuscript aims to provide the most current systematic roadmap and “how-to” of the existing methods. It also points to the existing knowledge gaps and directions for future research.
Highlights
This review provides a systematic roadmap and directory, organizing existing Kano categorization methods into clusters
It offers an in-depth critical analysis of the strengths and weaknesses of each Kano categorization method, allowing for a nuanced understanding of their limitations, key theoretical problems, terminological ambiguities, and conceptual issues
It highlighting areas requiring further research.
It facilitates the use of the most advanced categorization methods based on specific research needs and goals.
Introduction
Any product or service can be viewed as a bundle of product/service attributes or quality components that impact such outcomes as product/service selection, customer satisfaction (CS), loyalty, and so forth. (Oliver, 1997). The most common view on CS implies that it is a function of product/service attributes’ performance and better or worse performance at the attribute level impacts CS (Oliver, 1997). Several streams of research have investigated determinants of CS through theories and methods that aim at improving it through attribute performance and capturing the best combination of quality components (Bartikowski & Llosa, 2004; Chang et al., 2012; Gregory & Parsa, 2013; Gregory et al., 2015; Mikulić & Prebežac, 2011; Pandey et al., 2022; Reichenbach et al., 2022; Velikova et al., 2017).
Some of those methods utilize a summative approach, while others use a multiplicative approach. Summative methods, such as Importance-Performance Analysis (IPA), Quality Function Deployment (QFD), and so forth, view CS as a sum of product or service utilities and imply linear relationships between attribute performance and CS (Martilla & James, 1977, Oliver, 1997). As a result, summative methods aim at performance maximization, which has been criticized by many academics (L. F. Chen, 2012; Mittal et al., 1998; Oliva et al.,1992; Slevitch & Oh, 2010; Tontini & Silveira, 2007) as it can be wasteful because abundant empirical evidence shows that the relationship between CS and attribute performance is generally non-linear (L. F. Chen, 2012; Finn, 2011; S. Lin et al., 2010; F.-H. Lin et al., 2017; Mittal et al., 1998; Reichenbach et al., 2022; Ting & Chen, 2002). Thus, continually focusing on the greater performance of product or service attributes in an attempt to increase CS does not justify the additional investments (Brandt, 1988; W. Deng, 2007; Fuller & Matzler, 2007; Högström et al., 2010; Hui et al., 2004; Johnston, 1995; F.-H. Lin et al., 2017; Slevitch & Oh, 2010; Ting & Chen, 2002).
Multiplicative approaches account for non-linear relationships between CS and attribute performance aiming at performance optimization, that is, finding the combination of product/service attributes producing the highest level of CS. The method most associated with optimization and accounting for non-linearity was developed by Kano et al. (1984) and is frequently called the Kano model. The Kano model embraces the multiplicative nature of CS function and classifies product attributes based on their non-linear impact on CS. It has been widely applied in both product development and operational performance optimization (Fuller et al., 2006; Pandey et al., 2022; Pugna et al., 2021).
The original Kano method was praised for its simplicity, but criticized for many other shortcomings (L. F. Chen, 2012; Y. F. Kuo, 2004; Mikulić and Prebežac, 2011; Velikova et al., 2017; Yang, 2003). One of the major criticisms is the traditional Kano method takes attribute performance only in conditions of fulfillment/presence or non-fulfillment/absence. It fails to account for two levels of impact analysis: (1) product/service attribute level and (2) overall satisfaction (Violante & Vezzetti, 2017). Thus, the original method lacks operational precision and does not evaluate the range of impact that attribute performance has on the overall CS. As Violante and Vezzetti (2017) pointed out, the original Kano model/method only provides qualitative descriptions of various relationship patterns with minimal quantitative analysis or measurement of the relationships between attributes and CS.
Justification and Objectives of the Study
Since the traditional Kano categorization method has some major limitations, several modifications of the original method as well as completely new approaches were developed to address that issue, that is, regression-based methods, the Association Rule method, the Neural Network method, and so forth. (F.-H. Lin et al., 2017; Madzík, 2018; Mikulić & Prebežac, 2008, 2011; Pugna et al., 2021; Reichenbach et al., 2022; Witell et al., 2013). Those methods have not been comprehensively reviewed and compared, and, more importantly, there is no consensus on the suitability or fit of each method for a particular research situation based on strengths and weaknesses (Violante & Vezzetti, 2017).
An analysis of the Kano model literature review publications conducted since 2008 (see Table 1) shows that none of them fully synthesized the developments in the Kano categorization methods (Gregory & Parsa, 2013; Mikulić & Prebežac, 2011; Pandey et al., 2022; Reichenbach et al., 2022; Shahin et al. 2013; Song, 2016; Witell et al., 2013).
Summary of Kano model/ method review publications.
Designates reviews in the field of hospitality and tourism.
To illustrate, the highly regarded and cited work by Mikulić and Prebežac (2011) only examined methods developed by 2010—that is, the original Kano method, Penalty-Reward Contrast Analysis (PRCA), the “Importance Grid” (IG), Critical Incident Technique (CIT), and Analysis of Complaints and complements (ACC)—which are not the most recent methodological advancements in the area. In another notable attempt, Witell et al. (2013) mentioned several categorization methods, but primarily focused on the themes emerging in research related to the Kano model. Similarly, Gregory and Parsa (2013) discussed the original Kano model and method, its later combinations with SERVQUAL and QFD, and one of the regression methods, but concentrated on the evolution of contextual applications and findings of the Kano model publications. In another review attempt, Shahin et al. (2013) as well as Pandey et al. (2022) focused on the typology of the Kano model and contextual applications, rather than categorization methods.
In the most recent methodological analysis attempts that were outside of the scope of this systematic literature review, Song (2016) conducted a critical review of the original Kano method’s wording and its impact on attribute classification, but the focus of the analysis was on the traditional approach (functional and dysfunctional questions) and its modifications. F.-H. Lin et al. (2017) and Reichenbach et al. (2022) only reviewed and evaluated regression-based categorization methods, thus lacking comprehensiveness. The work of Violante and Vezzetti (2017) provided a good overview of several qualitative and quantitative Kano approaches (such as Fuzzy, Continuous Fuzzy, Analytical Kano models, PRCA, etc.) and described strengths and weaknesses of those methods, but their work was still fragmented as it did not include recent, more advanced regression methods as well as the Association Rule method and the neural network method. Thus, there is no publication that provides a complete synopsis and roadmap of the existing methods. There is no typology or grouping of the methods that would allow us to look at them in a systematic manner. More importantly, there is a lack of a comprehensive, systematic analysis of the methods in terms of their strengths, weaknesses, and fit for different types of research projects.
Having a comprehensive typology of available methods, which includes limitations, strengths, and fit, would be particularly beneficial for hospitality and tourism (HT) academics. A review of 39 empirical studies involving Kano model categorizations conducted since 2011 in the HT field revealed that 43% employed the original Kano categorization method (i.e., Chiang et al., 2019; Go & Kim, 2018; Gregory et al., 2015; Shen et al., 2021, etc.), 20% used PRCA and its derivatives (Alegre & Garau, 2011; Pratt et al., 2020; Velikova et al. 2017, etc.), 13% used moderated regression approach (Chen, 2014; Harrington et al., 2017; Pyo, 2012, etc.), and 8% used Fuzzy Kano Methods (Lin et al., 2015; Yeh & Lin, 2017; Yeh et al., 2019), which are not the most advanced Kano categorization approaches. With regard to the most progressive tools, none used the Ridge Regression method, only one study utilized the Association Rule Method (Chen, 2015), and two used the Neural Network approach (Mikulić et al., 2012; Zhou & Yao, 2023). Such numbers may point to a lack of awareness about the most advanced categorization methods or poor understanding of how to use them among HT academics.
Thus, systematically analyzing research methods holds utility for multiple reasons. First, it empowers HT academics and practitioners to align chosen methods with their specific research inquiries, resulting in more robust and valid findings. By understanding the intricacies of the available methods, HT academics can tailor their approach to the nuances of their study, enhancing the accuracy of outcomes. Second, such analysis is pivotal in addressing the limitations of existing methods. By scrutinizing the strengths and weaknesses of current approaches, HT academics can identify gaps and biases, fostering methodological innovation and refinement. Ultimately, this process bolsters the overall quality and credibility of research and knowledge advancement.
Consequently, the present manuscript aims to address the limitations of the existing literature review studies on the subject and provide a comprehensive critical overview of Kano categorization methods, including the traditional Kano method and its derivatives, as well as newer approaches, such as regression-based methods, the Association Rule method, and the Neural Network method. In pursuing this objective, the present study:
Classifies existing methods into main categories based on the underlying principles and categorization approaches.
Describes the categorization procedures of each method (what and how) to give a one-stop directory of the available methods.
Provides a comparative critique by pointing out the strengths and weaknesses of each method.
Points on the current issues on the subject and directions for future research.
This study makes significant contributions to the existing body of knowledge in both theoretical and practical realms, providing valuable insights and benefits for academics and practitioners alike. From a theoretical perspective, the unique contributions of the current review stem from the fact that it provides a systematic roadmap to HT academics who want to use the Kano model, and gives them a directory by organizing the existing Kano categorization methods into clusters. This is a major step forward in better understanding the widely used Kano model as its application can sometimes be intricate. By offering a clear and structured guide, HT academics should be able to navigate the complexities of the existing methods with greater ease and confidence.
Moreover, a comprehensive directory of existing methods can improve awareness and understanding of the tools that are available, and streamline the process of selecting the most appropriate approach for specific objectives, thus ensuring more focused and targeted studies. Another key theoretical contribution of this study is the inclusion of a detailed procedural outline. By presenting a step-by-step explanation of each method’s application, this study enables academics to implement them effectively and with confidence. This comprehensive procedural understanding ensures that academics can avoid potential pitfalls and inaccuracies, leading to more robust and reliable research outcomes.
Furthermore, the study not only systemizes and describes the different categorization methods but also critically assesses their strengths and weaknesses. By shedding light on the pros and cons of each approach, this paper allows HT academics and practitioners to gain a nuanced understanding of the limitations and potential biases of the various methods related to the Kano model. This critical analysis lays the groundwork for future HT scholars to improve and expand the methods’ applicability and accuracy. Additionally, the study identifies several theoretical problems, terminological ambiguities, and conceptual issues within the Kano model, highlighting the areas that require further exploration and refinement in future research.
From a practical standpoint, this study serves as a comprehensive information source for academics and practitioners interested in employing the Kano model categorization. Rather than having to search for scattered resources, they can find all the necessary information and guidelines in one place. This ease of access saves time and effort, encouraging more academics and professionals to adopt the Kano model in their projects. More importantly, the study facilitates the usage of the most advanced and progressive methods, potentially improving the quality of research. Additionally, the study empowers HT academics and practitioners to select the most suitable approach based on their specific needs and objectives. By understanding the nuances and capabilities of each method, they can make informed decisions about which technique aligns best with their study’s goals and conditions, thus enhancing the relevance and validity of their findings.
Finally, the study offers a holistic and integrated perspective of what drives customer satisfaction. By understanding the Kano model and its various categorization methods, academics and practitioners gain deeper insights into the multifaceted nature of customer preferences and expectations. This understanding is invaluable for designing products, services, and experiences that align with customer needs, ultimately leading to improved customer satisfaction and loyalty.
Research Method
The current study employed a systematic literature review approach as recommended by Palmatier et al. (2018) and Tranfield et al. (2003). The procedures and analysis included the following stages: planning (formulation of questions and criteria, identification of key search criteria, relevant databases, and timeframe), searching and reviewing (screening, extraction, selection), and synthesis and reporting of the results (see Figure 1).
Research Framework.
Research Questions and Search Criteria
The main research questions for the study were: (1) What are the existing methods developed for Kano model categorization? (2) Can those methods be clustered based on common characteristics? (3) What are the strengths, weaknesses, and fit of those methods?
The Inclusion Criteria and Sources
Only peer-reviewed journal articles were included in the study. Literature from the fields of business, social sciences, hospitality, tourism, economics, computer science, and engineering were included. The primary selection criteria were using the Kano model as a theoretical framework and development or focus on categorization methods. ABI/INFORM, EBSCO, Emerald, Elsevier Scopus, John Wiley publication, JSTOR, Sage, Springer, and Taylor and Francis were used for searching publications. The following search strings “Kano,” “categorization,” “attributes,” “classification,” “method,” “approach,” and “framework” were applied. Titles and abstracts of the identified publications were used for primary selection. Additional publications were identified based on the reference lists using the “Cited by” feature of Google Scholar®.
A list of 1,123 publications was obtained from the first step of the primary search. The initial screening produced 255 remaining articles that were related to the Kano model categorization/classification methods. In the next step, the articles were thoroughly examined, and 188 were excluded from the list due to duplication (already established Kano categorization methods). In all, 67 remaining articles were studied in detail, which led to the removal of another 10 based on inclusion criteria. Thus, the remaining 57 articles were selected for the final analysis.
Results: Critical Overview of Kano Model Categorization Methods
The Original Kano Model and Method
The original Kano model was proposed by Kano et al. (1984). The model captures different patterns of relationships between attribute performance and CS under the assumption that not all attributes have linear relationships with CS (Kano et al., 1984). The Kano model plots satisfaction on the Y-axis and attribute fulfillment/performance on the X-axis and reveals five categories of attributes based on their predicted effect on CS: the must-be quality, the attractive quality, the one-dimensional quality, the indifferent quality, and the reverse quality (Figure 2).
Must-be quality attributes represent basic functions of the product/service that serve as necessary prerequisites or must-haves. These attributes exhibit a curvilinear impact on CS in that customers are dissatisfied when these attributes are not present or performed inadequately. On the other hand, these attributes do not produce satisfaction when present or performed well. To illustrate, non-functioning plumbing in a hotel room will produce dissatisfaction, but when plumbing functions well it will not necessarily lead to customer delight because it is just expected.
One-dimensional quality attributes are described as performance needs because they have a symmetric impact on CS: the higher the attribute performance, the greater the customer satisfaction, and vice versa. For instance, the better the room service in terms of speed, quality of food, and service, the higher the satisfaction. Conversely, poor room service will lead to a similar level of dissatisfaction.
Attractive quality attributes are often called “delights” or “excitement” factors. They also have a curvilinear impact on CS and result in an increase in satisfaction when present or when performance is high. However, they do not bring dissatisfaction when not provided or when performance is low. Unexpected perks like complimentary spa treatments, a personalized note with a gift in one’s room, or a free minibar, are good examples. These features are not typically expected, so they do not produce negative feelings when absent, but when present, can significantly boost satisfaction.
Reverse-quality attributes are the opposite of one-dimensional attributes. These attributes have a linear negative impact on CS. For example, some guests might find high-tech rooms with overly complex controls for lights, blinds, or temperature to be frustrating rather than pleasing, or while attentive service is appreciated, service that is overly interfering or ingenuine might irritate some guests.
Indifferent quality attributes produce no substantial impact on CS based on their presence/absence or adequate/inadequate performance. These attributes have no impact on CS and stay independent of change in CS, for example, whether the TV is brand A or brand B might not impact the satisfaction levels for a majority of hotel guests.
The Kano Model.
Based on these categories, Kano et al. (1984) proposed the following recommendations for product development or product positioning: must-be quality and one-dimensional attributes should be given priority to maintain competitiveness; however, to achieve maximum satisfaction and differentiation from competition, attractive quality features should also be provided.
To categorize attributes into the five categories, Kano et al. (1984) developed a special questionnaire consisting of pairs of questions in functional and dysfunctional forms for each attribute of a product/service. Functional questions in each pair ask about respondents’ feelings in the case of fulfillment. For example, “How do you feel if mobile check-in is offered by a hotel?” Dysfunctional questions ask about feelings in conditions of non-fulfillment. For instance, “How do you feel if mobile check-in is not offered by a hotel?” Based on the frequencies of responses, the attributes are categorized by means of a special evaluation table (see Figure 3).
The Kano Categorization Method.
The original Kano method has several strengths and weaknesses. Some of the strengths of the original Kano method include ease of use and interpretation, simplicity of analysis, ability to identify the desires of customers, and consideration of non-linear patterns (Madzík, 2018). The Kano method works well when the goal is to identify what happens when a certain product/service attribute is provided (presence/fulfillment condition) or not provided (absence/unfulfilled condition). That is why the original method works well for new product development (Mikulić & Prebežac, 2011).
However, a significant problem can emerge when conditions of “fulfillment” and “non-fulfillment” go beyond mere absence and presence and include a range of various degrees of attribute performance. For example, the speed of hotel check-in or room service delivery can vary significantly and range from several minutes to hours in some instances. Most real-world conditions imply that attributes of a product or service have a range of performance. Unfortunately, the original Kano method does not account for that. Interestingly, in contradiction to the categorization method, the Kano model depiction does reflect a range of attribute performance/fulfillment and CS, not just binary states of presence or absence (Figure 1). The X-axis represents a continuum of attribute performance from completely fulfilled to completely unfulfilled and the Y-axis represents a range from complete satisfaction to complete dissatisfaction.
One of the major weaknesses of the original method is the inability of the Kano questionnaire and categorization table to account for an array of attribute performance conditions. Moreover, the common argument that the Kano questionnaire remains the most appropriate approach to identify original Kano categories (Mikulić & Prebežac, 2011) can be challenged since the original method only works for situations when a product/service attribute is present or absent, without other possibilities. Additionally, an indifferent category implies no impact on CS, which is hardly possible from a statistical perspective, since attribute performance and satisfaction would need to correlate at precisely r = 0 across all variations in performance.
Other criticisms of the original method include its lengthy, cumbersome, and often perplexing questionnaire. Two similar questions are asked for each attribute in functional and dysfunctional form, requiring respondents to envision what happens when an attribute is present or absent (F.-H. Lin et al., 2017; Matzler, Bailom et al., 2004). This often confuses respondents and decreases their willingness to complete the questionnaire (L. F. Chen, 2012).
Additionally, the original method is qualitative in the sense that it does not estimate the extent of CS produced by the attributes (S. Lin et al., 2010), and it does not fully reflect real-world experiences as it asks questions about a hypothetical situation and does not require respondents to actually experience a certain product/service attribute (Mikulić & Prebežac, 2011). As Violante and Vezzetti (2017) state, the original Kano method captures qualitative descriptions of various relationship curves without showing quantitatively the relationships between attribute performance and CS. As a result, the original method only identifies the possible impact of an attribute’s provision or non-provision on CS. It does not show the strength of the attribute performance effect, as it only places attributes in the Kano model categories. Also, it does not allow the academics to identify which attributes should be improved due to their current performance. Another common criticism is that due to its qualitative nature, the Kano categories cannot accurately reflect the extent to which the customers are satisfied, thus lacking a quantitative measurement of CS (Violante and Vezzetti, 2017). Finally, the original Kano method is constructed according to customers’ perceptions at a certain time point, which may change over time and with a new situation (F.-H. Lin et al., 2017).
Modifications of the Original Kano Model
Qualitative methods
Matzler and Sauerwein (2002) and Mikulić and Prebežac (2011) refer to the Critical Incident Technique (CIT) and Analysis of Complaints and Complements (ACC) as qualitative approaches for the Kano model. CIT is often associated with the two-factor theory of work satisfaction by Herzberg (1967) and ACC has been discussed by Cadotte and Turgeon (1988), Lewis (1987), and Oliver (1997). Both methods are based on the notion that certain attributes are associated with satisfaction and some with dissatisfaction. Consequently, the attributes can be categorized by counting how many times customers mention them in a positive context or a negative context (Mikulić & Prebežac, 2011).
When the number of negative incidents/complaints is considerably larger than the number of positive incidents/compliments, the attribute is viewed as a “must-be.” When the number of positive incidents/compliments is significantly larger than the number of negative incidents/complaints, the attribute is considered “attractive.” When the number of positive incidents/compliments and the number of negative incidents/complaints are the same, the attribute is categorized as “one-dimensional.”
The strengths and weaknesses of the CIT and ACC methods are similar to the traditional Kano method. However, the biggest problem with those methods is that they do not allow researchers to determine comprehensively which attributes should be included because only the attributes derived from the customers’ comments are considered (Mikulić & Prebežac, 2011), and therefore they tend to reflect extreme cases of satisfaction or dissatisfaction (Oliver, 1997).
Semantic Measurement Modifications
There have been several attempts to modify the original Kano method in terms of measurement wording. Some attempts were centered around the fact that the attribute categories could not precisely reflect to what extent customers are satisfied. As Song (2016) points out, the intrinsic problem of Kano’s wording comes from the ambiguity of the five original measurement items and paired questions.
To address the issue with measurement semantics, Berger et al. (1993), Tan et al. (1999), Tan and Pawitra (2001), Nilsson-Witell and Fundin (2005), Busacca and Padula (2005), Witell and Lofgren (2007), Sireli et al. (2007), Rejeb et al.(2008), and Crostack et al. (2010) have changed wordings of the 5-point measurement scale to make it less ambiguous for respondents.
Placement in addition to the wording of questions was altered in two studies. Lee and Newcomb (1997) changed the sequence of the paired functional and dysfunctional questions to random placement but did not test the effectiveness of that step. Mikulić and Prebežac (2011) changed the paired questions from provision-based (provided/not provided) to performance-based (feature/attribute is provided/not provided and works well/poorly). The authors detected a difference in responses and classification between two types of paired questions but did not demonstrate the superiority of one type of question over the other. A good summary and analysis of publications on the measurement semantics of the Kano model was done by Song (2016).
Original Kano method with modified categories and quantitative indices
Some modification attempts introduced subsidizations of attribute categories and more quantified measures, that is, indices, into the original Kano model (Anne & Grønholdt, 2001; J. K. Chen & Lee, 2009; Ek & Cıkıs, 2015; Y. F. Kuo et al., 2012; Shahin et al., 2013; Shen et al., 2000; Song, 2016, Tan et al., 1999; Tontini, 2000, 2003; Yang, 2005). Good examples of those studies are Tontini (2000, 2003) where attribute-level scale included the following items:
Very satisfied
Satisfied
It should be this way
It is indifferent
I can live with it
Dissatisfied
Very dissatisfied
Other.
Based on the obtained responses Tontini (2000) developed an extended categorization framework summarized in Table 2.
Tontini’s (2000) Categorization Framework.
Note. VA = very attractive; A = attractive; O = one-dimensional; N = neutral; M = must-be; VM = very must-be; R = reversal; satisf. = satisfied; dissat. = dissatisfied, indiff. = indifferent.
After categorizing responses based on the table above, Tontini (2000) proposed calculating the average satisfaction impact (SI), similarly to Berger et al. (1993), using the formula:
Correspondingly, the average dissatisfaction impact (DI) can be calculated using the following formula:
Here VM, M, O, A, VA, and N are the percentages of very must-be, one dimensional, attractive, very attractive, and neutral answers for a specific attribute. Then, Tontini (2000) suggested plotting the extent of satisfaction or dissatisfaction along two axes, shown in Figure 4. See the original publication for more details.
Coefficients Plot (Tontini, 2000).
Similarly, Yang (2005) incorporated the notion of importance and increased the number of categories to eight: highly attractive, less attractive, high value-added, low value-added, critical, necessary, potential, and carefree (see Figure 5). In the proposed eight-category model, if an attribute in the must-be category is considered important by the customer, then it is called a critical requirement factor, and if it is considered less important, then it is a necessary factor. One-dimensional attributes increase CS; therefore, those are labeled as value-added, high, and low. Attractive attributes of importance are called highly attractive and those of less importance are called less attractive. Indifferent attributes of less importance are called carefree and those of more importance are potential.
Kano Model with Eight Categories.
An interesting feature of this type of Kano model is the locations of the starting and ending points of categories: more important categories have a steeper incline relative to the Y-axis, reflecting a more noticeable pattern of the category’s relationship to customer satisfaction (Madzík, 2018).
A similar modification of the original Kano model is based on the assumption that every customer assesses the quality of a product according to their perception of the extent to which their requirements have been met, along with the attributes’ importance (Pouliot, 1993; Yang, 2005). In comparison with the previous modification by Yang (2005), the starting and ending points of the respective curves are identical (see Figure 6), whereby A1 represents a less attractive category, A2 represents an attractive one, and A3 represents a highly attractive one. Similarly, M1 represents a less important must-be category, M2 represents a must-be category, and M3 represents a highly important must-be category. This model also adds negative equivalents of the must-be category and attractive-to-reverse category (Shahin et al., 2013).
Kano Model – Type III (Shahin et al., 2013).
Building upon the modifications mentioned above, Song (2016) introduced the Better–Worse (B-W) method with a new paired question format, a 5-point ordinal scale for attribute level response, and the Potential Satisfaction/Dissatisfaction Index. Based on Prospect Theory by Kahneman and Tversky (1979), Song (2016) divided the one-dimensional attribute category into four subcategories: “Highly one-dimensional” (OH), “Less one-dimensional” (OL), “One-dimensional with an attractive tendency” (OA), and “One-dimensional with a must-be tendency” (OM). The attractive category consisted of “Highly attractive” (AH) and “Less attractive” (AL). The must-be category included “Highly must-be” (MH) and “”Less must-be (ML). For the indifferent category, Song classified as “I” attributes that received “neutral” for both functional and dysfunctional questions. Like Berger et al. (1993), Song included the “Skeptical” category (S), the attributes that received the same or similar answers for both functional and dysfunctional questions. The reverse category (R) consisted of the attributes that received negative answers for functional questions and positive answers for dysfunctional questions. Figure 7 illustrates the classification procedure in the B–W method.
B–W Method Categorization Framework.
To quantify the impact on CS, Song (2016) introduced the potential satisfaction index (PSI) and potential dissatisfaction index (PDI), which are similar to SI and DI by Berger et al. (1993).
PSI indicates how much CS will increase when an attribute is evaluated more positively.
where
i = respondent (1, . . ., n);
j = quality attribute (1, . . ., m);
Sij = satisfaction level of respondent i when attribute j gets better; and (1 = very satisfied, 2 = satisfied, and 3 = neutral).
Song (2016) states that “dissatisfied” and “very dissatisfied” answers indicating negative assessments are not considered in the PSI calculation because PSI is only relevant for conditions of satisfaction.
PDI indicates how much customer dissatisfaction will increase when an attribute is evaluated negatively.
where
i = respondent (1, . . ., n);
j = quality attribute (1, . . ., m);
Dij = dissatisfaction level of respondent i when attribute j gets worse; and (3 = neutral, 4 = dissatisfied, and 5 = very dissatisfied).
Likewise, “satisfied” and “very satisfied” answers for worse-questions are not used in the PDI calculation.
When compared to the original Kano approach, the B-W method showed different and improved results (Song, 2016). Nevertheless, those results need additional empirical validation. The categorization results were in high agreement (91%) with the original method, but there were inconsistencies between the two approaches. Additionally, the B–W method showed better attribute discrimination capabilities than other classification methods using the 5-point ordinal scale (Bhattacharyya & Rahman, 2004; Busacca & Padula, 2005; J. K. Chen & Lee, 2009). Song (2016) also reported that the ordinal scale method generated more one-dimensional attributes than the original Kano method.
Another group of modifications of the original model comes from the fact that reverse and indifferent quality attributes are very rare (Matzler & Sauerwein, 2002; Witell et al., 2013). As a result, the original Kano model was reduced to a three-factor model (see Figure 8) including only must-be, attractive, and one-dimensional quality categories (Brandt, 1988; L. F. Chen, 2012, W. J. Deng et al., 2008; S. Lin et al., 2010; Little et al., 2015; Matzler & Sauerwein, 2002; Matzler et al., 2004; Slevitch & Oh, 2010; Staus & Becker, 2012; Velikova et al., 2017). In the three-factor literature (Bartikowski & Llosa, 2004; Berger et al., 1993; Busacca & Padula, 2005; Cadotte & Turgeon, 1988; Erto & Vanacore, 2002; Friman & Edvardsson, 2003; Fuchs & Weiermair, 2003; Johnston, 1995; Kondo, 2000; Y. F. Kuo, 2004; Matzler & Hinterhuber, 1998; Matzler & Sauerwein, 2002; Matzler et al., 1996; Matzler et al., 2003, 2004; Mikulić & Prebežac, 2008, 2011; Slevitch et al., 2013; Witell et al., 2013), attribute categories have been variously labeled:
must-be as basic, core factors, or dissatisfiers,
one-dimensional as performance or hybrid factors,
attractive as excitement, facilitating factors, or satisfiers.
Three-Factor Model.
Fuzzy Kano model and Continuous Fuzzy Kano model
The Fuzzy Kano model (FKM) is grounded in the Fuzzy Set Theory (Zadeh, 1965) that addresses the rationality of uncertainty and vagueness of human thought. The main proposition of the theory lies in the use of the membership function to express the degree of respondents’ feelings based on their own choices making it closer to real human thinking (Lee & Huang, 2009; Violante & Vezzetti, 2017).
Instead of offering a single answer or a certain range of the answer in a survey (true or false logic), the FKM approach offers respondents an opportunity to have various answers capturing degrees of response, thus allowing more realistic expression of ideas to the issue (Wang & Fong, 2016). Similarly to the original Kano method, the FKM method uses functional and dysfunctional questions. However, in contrast to the original method, the TKM approach allows the selection of various answers reflecting degrees, as shown in Table 3.
Fuzzy Kano model questionnaire applied.
The normalization process is used to ensure that the sum of possibility degrees among multiple answers equals unity. A fuzzy scale is depicted by fun = (0.75, 0, 0.25, 0, 0) and dys = (0, 0, 0.1, 0.8, 0.1). Using the matrix algebra, a 5 × 5 fuzzy relation matrix R is obtained via (fun)t × (dys), where a superscript t denotes the transpose operation. Then a five-element row vector is used to display responses:
After a relation matrix R is obtained, the identification of Kano categories for each attribute is done as shown below (Wang & Fong, 2016):
The possibility degree among Kano categories is obtained as follows:
Once the possibility degree of each attribute is gathered, its mean value is aggregated and extracted to derive their importance weights (Wang & Fong, 2016).
The strengths of the Fuzzy Kano approach include relative ease of use, simplicity of analysis, and better fit to human thinking. The weaknesses are similar to the original Kano method. Plus, the FKM approach requires more time and effort from the interviewers and interviewees due to the complexity of the assessment. The FKM cannot rank attributes within the same category and can have difficulty in categorizing accurately when there is only a small difference between two or more categories.
The method also suffers from the discontinuity problem (Violante & Vezzetti, 2017).
The latter issue was addressed by Wu and Wang (2011), who developed the Continuous Fuzzy Kano Model (CFKM). The CFKM uses a continuous approach for fuzzy Kano evaluations, which involves choosing only one point as an answer to each functional or dysfunctional question (see Figure 9). The method also includes an assessment of attribute importance on a 9-point Likert-type scale.
CFKM Question and Measurement Sample.
The CFKM uses the traditional Kano categorization table but introduces an influence value to represent the contribution of each of the 25 combinations in the original Kano categorization table. The CFKM also incorporates an evaluation index (EI) to prioritize attributes. Wu and Wang (2011) provide two alternative mechanisms to do that: the first one is based on the ranking of the EI and the second is based on a threshold of the EI values. Wu and Wang (2011) also introduced a modified fuzzy Kano questionnaire with closed-ended questions, which fits for being self-administered because it is easier and quicker to complete in comparison with the FKM questionnaire.
Original Kano method integrated with QFD, TRIZ, and SERVEQUAL
The original Kano method has often been used in combination with other methods for measuring quality, particularly of new products. Examples of such methods include quality function deployment (QFD), TRIZ creative problem strategy, and Parasuraman and colleagues’ (1985) SERVQUAL model of service quality Lin et al., 2011; Su and Lin, 2008; Tan & Pawitra, 2001; Tan & Shen, 2000; Tan et al., 1999). The main purpose of incorporating those methods with the Kano categorization was to approach the process of new product development or product improvement comprehensively while incorporating consumer feedback, competition, and the company’s resources into the decision-making process. This stream of research originated in the field of Total Quality Management, which is commonly used in engineering and product design (Tan & Pawitra, 2001).
Tan et al. (1999), Tan and Shen (2000), and Tan and Pawitra (2001) integrated Kano categorization into QFD, a systematic method for making strategic decisions that allows researchers to evaluate a combination of product features in terms of how those meet the needs of the customers, and also accounts for competition and availability of internal resources (Tan et al., 1999). QFD is a graphic tool for defining the relationship between customer preferences and product abilities. It is based on the idea that products should be designed to reflect customers’ needs and desires. QFD has a house-like shape with a correlation matrix as its roof, the customer wants versus product features as the main part, competitor evaluation as the porch, and so forth. A typical QFD matrix includes importance scores, competition scores, target scores, and improvement ratios (the difference between actual and desired evaluations). Those scores are used to benchmark the performance of the product or service in meeting customers’ needs with the competition’s performance as well as the comparison of customer satisfaction scores. QFD analysis also includes computing several ratios, such as the improvement ratio and the raw importance, which point to what needs to be adjusted. Then, the Kano model is used to adjust the priority of each attribute based on its impact on satisfaction, that is, the percentage of customer satisfaction improvement that can be achieved by increasing the same percentage on the product or service performance (Tan & Shen, 2000). The Kano method is used in this case because it accounts for the non-linear relationships between satisfaction and performance.
Later, Tan and Pawitra (2001) added a SERVEQUAL component to Kano with the QFD method. SERVQUAL is a model by Parasuraman et al. (1985,1988) consisting of five dimensions of service quality: tangibles, reliability, responsiveness, assurance, and empathy. It is also used as a diagnostic technique for uncovering strengths and weaknesses in service quality by evaluating quality as a function (difference) between expected service and perceived service. Figure 10 shows how the Kano model is combined with SERVEQUAL.
Kano Method Integrated With SERVEQUAL.
Integrating the Kano method with QFD and SERVQUAL has several advantages. It helps to prioritize an organization’s weaknesses based on its position in the Kano model, thus improving chances of higher customer satisfaction. Additionally, it allows SERVQUAL to concentrate on the attractive attributes that are valued during the process of product/service innovation (Tan & Pawitra, 2001). At the same time, the two approaches described above carry the same weaknesses as the original Kano method.
Stemming from the field of industrial engineering, Lin et al. (2008, 2011) proposed to incorporate the Theory of Inventive Problem Solving (TRIZ) by Altshuller (1999) as well as QFD along with the modified traditional Kano method as suggested by Yang (2005). Similarly, Shahin et al. (2017) combined TRIZ with the Kano approach. The TRIZ approach aims to solve contradictive problems and generate new ideas. Contradictions stand for outcomes that can lead to useful or harmful results and make one or more sub-systems fail. Altshuller (1999) outlined 39 parameters as improved attributes or deteriorated attributes that can be arranged in a 39 × 39 contradictive matrix in which the parameters in the vertical rows include improved attributes and in the horizontal columns deteriorated attributes are placed (Lin et al., 2008). For example, when purchasing a personal computer, the big size of the computer monitor can be viewed as an improved attribute for consumers, but it increases the weight of the computer at the same time; thus, it is a deteriorated attribute as well.
Lin et al. (2008, 2011) proposed a sequential categorization procedure that included several steps. Step 1 involves QFD Analysis and starts with a survey that identifies the features customers need and desire. Based on the results, the QFD matrix translates the needs of the customers into the design targets of a proposed product. Then, a correlation matrix of quality elements is constructed and the correspondence relationships between the individual quality elements are shown in the roof part of the matrix. After that, the contradiction relationships between the quality elements are examined. In Step 2, the contradiction relations between different quality attributes are determined and recorded in the dome of the QFD table. If the contradiction relationships exist, the TRIZ technique is implemented in Step 3. According to the TRIZ contradiction matrix, the intersections of the improving and worsening parameters are denoted and TRIZ principles are applied to the specific conflict points to generate all possible new functions of quality elements. Consequently, possible new functions of quality elements are generated. In Step 4, Lin et al. (2011) use the original Kano questionnaire with functional and dysfunctional questions and Yang’s (2005) extended attribute categories. Additionally, Lin et al. (2011) report importance and satisfaction scores for each evaluated attribute and, based on that, make recommendations if new product features should be implemented.
The advantages of incorporating the Kano method with TRIZ and QFD include a better ability to evaluate new product ideas, not only relative to customers’ potential interests but also regarding the existing product features and the company’s resources and abilities. Additionally, Lin et al. (2011) report a better ability of their method to create attractive quality for novel product features, which can lead to a competitive advantage. At the same time, the method only measures preferences based on a hypothetical situation, not actual experience and performance, and carries the same disadvantages as the original Kano method.
Table 4 presents a summary of the methods based on the original Kano categorization method.
Summary of the methods based on the original Kano categorization approach.
Regression Methods
To overcome the original Kano method’s disconnect from real experiences and actual performance, several regression methods were developed. Those methods apply measures of attribute-level performance and overall CS, which are easier to collect and provide a more comprehensive assessment (L. F. Chen, 2012). Most of the regression methods use dummy variable coding based on high and low attribute performance (Anderson & Mittal, 2000; Brandt, 1987; L. F. Chen, 2012; Mittal et al., 1998; Vanhoof & Swinnen, 1998). Regression analysis is then performed with the dummy variables as predictors and overall CS as the dependent variable. Then, the impacts of low or high attribute performances on overall CS are used for Kano categorization.
Importance grid (IG)
The Importance Grid method is one of the earliest regression methods that incorporate elements of Importance-Performance Analysis (Martilla & James, 1997). The IG was first cited by Vavra (1997) and later used in numerous studies (Bartikowski & Llosa, 2004; Busacca & Padula, 2005; Fuchs, 2002; Fuchs & Weiermair, 2003, 2004; Martensen & Gronholdt, 2001; Matzler & Hinterhuber, 1998; Matzler et al., 2003; Riviere et al., 2006). In this method measurements of explicit and implicit attribute importance are used to classify attributes.
Customers’ importance ratings are used to assess explicit importance. However implicit importance is derived by regressing measures of attribute performance against overall satisfaction. Standardized beta coefficients or partial correlation coefficients are used as measures of implicit importance (Bartikowski & Llosa, 2004; Busacca & Padula, 2005; Fuchs, 2002; Matzler & Sauerwein, 2002; Matzler et al., 2003, 2004). Then, the measures of explicit and implicit importance are used to classify attributes in a two-dimensional grid with four quadrants using grand means of implicit and explicit importance as thresholds. “Must-be” attributes have high (above average) explicit importance, but low (below-average) implicit importance. “One-dimensional” attributes have high explicit and implicit importance or low explicit and implicit importance. “Attractive” attributes have low explicit importance, but high implicit importance.
One of the advantages of the IG method is its simplicity and ease of use. Additionally, explicit importance can reflect attribute expectations from customers’ perspectives. Implicit importance can be a proxy of an attribute’s impact on CS. Because the IG method utilizes both direct and statistically derived importance measures, using a combination of these two types of importance may be useful for improvement strategies based on attribute-prioritizations (Mikulić & Prebežac, 2009). The IG method also has several serious weaknesses. Matzler and Sauerwein (2002), Fuchs and Weiermair (2003), Busacca and Padula (2005), and Mikulić and Prebežac (2011) criticize the validity of the IG method due to there being no theoretical justification for using explicit and implicit importance for Kano attribute categorization. Mikulić and Prebežac (2011) point to another weakness: Since the IG method implies that the different Kano model categories can be identified by comparing the relative grid positions, that means that the categorization of any given attribute is contingent on all the other attributes as reference points. Therefore, the IG method will always result in an attribute categorization into all of the Kano attribute groups whenever there are any differences in explicit and implicit importance. It also means that a given attribute categorization can change if the set of analyzed attributes is modified. The latter contradicts the assumption of the Kano model that the categorization of a given attribute should be objectively defined and consistent across any set of attributes.
Additionally, the IG method does not reveal the degree of the attribute’s impact asymmetry toward satisfaction or dissatisfaction. Although such a measure was proposed by Mikulić and Prebežac (2008), it is not part of the original IG method. Consequently, although the IG method cannot be regarded as a reliable approach for assessing attributes into Kano categories, it can be described as an advanced version of the importance-performance analysis because it overcomes some of the shortcomings of the original IPA method (Bartikowski & Llosa, 2004; Mikulić & Prebežac, 2011; Oh, 2001; Slevitch & Oh, 2010).
Penalty-Reward Contrast Analysis (PRCA)
The PRCA method was the first to utilize dummy coding for identifying the Kano categories. It was introduced by Brandt (1987) to identify value-enhancing attributes in transportation services and was later modified by other academics (Mittal et al., 1998). However, since then, the PRCA approach has become commonly used for Kano categorizations in many fields (Bartikowski & Llosa, 2004; Matzler et al., 2004; Ting & Chen, 2002; Velikova et al., 2017).
The PRCA technique begins by transforming the attribute performance scores into two dummy variables: a low attribute performance (penalty) variable and a high attribute performance (reward) variable. To transform the initial performance scores into the penalty variable, each score is recorded such that if the attribute score is below the mean, it is coded as a “1” (low performance). It is coded as a “0” (low performance absent) otherwise. Conversely, for the reward variable, the performance score above the mean is coded as “1” and a “0” is coded otherwise. Then, the penalty and reward dummy variables are regressed against the overall satisfaction score based on the following equation:
where CS i is a measure of customer satisfaction for the ith customer, and D1ij and D2ij are the penalty and reward performance dummy variables, respectively, for the ith customer and the jth attribute. When the two separate dummy coded variables are regressed against the measure of overall CS, two regression coefficients are obtained. Mikulić and Prebežac (2011) strongly recommend that only unstandardized dummy regression coefficients should be used with this method since standardized regression coefficients can skew the information provided with the unstandardized coefficients when the dummy variables have an equal distribution of ones and zeros or the same frequencies of lowest and highest performance scores.
The resulting low performance attribute coefficient β1j (the penalty) reflects the attribute’s negative impact on CS. Alternatively, the high-performance attribute coefficient β2j (the reward) shows a positive impact on CS. If there is a difference between the penalty and reward coefficients, this points to an asymmetric impact on CS. If the penalty coefficient fails to achieve significance, while the reward coefficient is both significant and positive, then the attribute is classified as an attractive factor. If the penalty coefficient is negative and significant, while the reward coefficient is not significant, then the attribute is classified as a must-be factor. If the penalty coefficient is negative and significant while the reward coefficient is positive and significant, the attribute is considered a one-dimensional factor. Additionally, if both coefficients are significant and equal in absolute terms, then it is again classified as a one-dimensional factor.
In 2008, Mikuliċ and Prebežac proposed an extension to the PRCA approach, which combined PRCA with some elements of Importance-Performance Analysis (Martilla and James, 1977), similar to the IG method. They recommended a three-step framework that included two additional steps: Impact Range-Performance Analysis (IRPA) and Impact Asymmetry Analysis (IAA). The revised approach included:
PRCA based on the methodology by Brandt (1987),
IRPA, similar to IPA as it uses scores of an attribute’s range of impact on overall CS, instead of attribute-importance,
IAA, an analysis of the asymmetry of attribute impact on overall CS.
In the proposed framework, absolute values of penalty and reward indices for each attribute are summed to provide indicators of an attribute’s Range of Impact on overall CS (RIOCS). Then, Mikuliċ and Prebežac propose calculating an Impact Asymmetry (IA) index which quantifies the asymmetry of an attribute’s impact on overall customer satisfaction. The IA index shows the extent to which an attribute has a satisfaction-generating potential (SGP) compared with its dissatisfaction-generating potential (DGP). The following notation is used:
ri : reward index for attribute i or alternatively the regression coefficient for the reward dummy variable.
pi : penalty index for attribute i or alternatively the regression coefficient for the penalty dummy variable.
After indices calculations, the arithmetic means of attribute-performance scores are depicted along the vertical axis of a two-dimensional grid, and RIOCS scores are placed along the horizontal axis. The grand mean of the performance scores and the grand mean of the RIOCS scores are used to divide the grid into four quadrants, similar to IPA. Attributes that perform below average but have RIOCSs close to or above average should be given high priority. Medium priority should be assigned to attributes with RIOCS and performance above average. Low priority should be assigned to attributes with RIOCS below average and performance above average.
For the IAA step, another two-dimensional grid is constructed, with RIOCS scores depicted along the horizontal axis and IA scores along the vertical axis. In addition, a line should be drawn at IA = 0. Mikulić and Prebežac (2008) suggest interpreting the IAA grid as follows: Attributes with IA less than 0, in the lower part of the grid, have a greater potential to create dissatisfaction than satisfaction; the authors label these attributes as “dissatisfiers,” which correspond with must-be/ basic factors. Conversely, attributes with IA greater than 0, in the upper part of the grid, are labeled as “satisfiers,” which are similar to attractive or excitement factors. As attributes move towards the IA = 0 line, the potential to create satisfaction and dissatisfaction becomes equal, and thus, these attributes can be referred to as “hybrids,” or one-dimensional attributes. The authors further subdivide attributes into five categories according to the degree of asymmetry of their impact on overall customer satisfaction:
(1) “delighters” (IA > 0.4);
(2) “satisfiers” (0.4 ≥ IA. > 0.1);
(3) “hybrids” (0.1 ≥ IA ≥ −0.1);
(4) “dissatisfiers” (−0.1 > IA ≥ −0.4); and
(5) “frustrators” (IA < −0.4).
In addition, the attributes are also subdivided into three categories according to their RIOCS: “high-impact attributes” (RIOCS > 0.225); “medium-impact” attributes (0.125 ≤ RIOCS ≤ 0.225); and “low-impact attributes” (RIOCS < 0.125).
The advantages of PRCA and PRCA-IRPA-IAA include application simplicity, connection to actual performance assessment, and ability to discriminate among attributes in terms of linear and non-linear impacts on CS (Hu et al., 2020; Mikulić & Prebežac, 2011). An additional strength of this model is that it considers attribute importance as a function of attribute performance in a way that does not necessarily correspond with the attribute’s importance as perceived by customers, but rather with the attribute’s actual impact on CS. Additionally, the model provides the ability to prioritize attributes as opposed to simply providing a yes/no type response for each attribute.
The method’s shortcomings include utilizing a model designed for continuous dependent variables that are discrete, and assuming the effect of a unit increase in high performance or low performance is the same throughout the spectrum of high and low performance settings. That is, the model assumes that both relationships are linear, which contradicts the hypothesized Kano model, which implies a quadratic relationship. Finally, this approach does not allow for interaction terms. While this is not a requirement for establishing the categories of the Kano model, it may be helpful in implementation as products and services have multiple attributes/features.
Logistic regression
F.-H. Lin et al. (2017) proposed a method based on PRCA but utilizing logistic regression to estimate the odds ratio of satisfaction to dissatisfaction due to attribute-level performance. In the proposed approach, the research design, data collection, and coding were based on both the PRCA method and the Fama and French’s (1993) regression model. F.-H. Lin et al. (2017) divided customers into two groups based on whether they were dissatisfied or satisfied. Quality factors were divided into three groups according to performance levels, and collected data was coded as dummy variables for the logistic regression analysis.
The classification of attributes involved fitting a logistic regression model for each of the k independent performance variables. For each of these variables, a logistic regression model was employed to determine the odds ratio of customer satisfaction at the high-performance level of the independent variable to customer satisfaction at the low performance level of the independent variable. This was denoted as ORSi, i = 1, . . ., k. Similarly, logistic regression was also utilized to obtain the estimated odds ratio of customer dissatisfaction at the low level of the independent performance variable to the high level of the independent performance variable. This is denoted ORDi, i = 1, . . ., k.
Now let
F.-H. Lin and colleagues’ (2017) odds decision diagram is shown in Figure 11. As denoted above, the angle θi is obtained from the inverse tangent function of ORSi and denoted as the customer satisfaction index. The angle ωi is obtained from the inverse cotangent function of ORDi and designated as the customer dissatisfaction index. Customer satisfaction and dissatisfaction are represented on the right and left sides of the horizontal axis. The horizontal component of the quality attribute vectors is labeled as the satisfaction odds ratio or the dissatisfaction risk ratio, depending on whether the vector points right or left, respectively. The difference in the horizontal components of each vector assesses the strength of the asymmetric relationship between quality attributes and customer satisfaction (see Figure 12).
Odds Decision Diagram for Customer Satisfaction.
ANN Structure.
F.-H. Lin et al. (2017) state that when the attribute vector rotates clockwise from the vertical axis and θi approximates 0, a low probability of CS with the attribute in question is indicated. When the angle is close to π/2, the probability of customer satisfaction with the attribute is high. By contrast, when the vector rotates counterclockwise, the probability of customer dissatisfaction is high. The authors suggest that the left and right sides of the decision-making diagram represent the quality attributes that influence customer dissatisfaction and satisfaction, respectively. Considering the decision sequence, the quality attributes on the right side must be maintained before those on the left side can be improved.
The biggest strength of the logistic regression method is that it accounts for asymmetric relationships between attribute performance and CS in a nonlinear model. Additionally, it also provides a mechanism for prioritizing or ranking performance attributes. However, it still shares some of the weaknesses of the other regression methods, such as a disregard for interaction terms and the application of a continuous model to ordinal data.
Moderated regression methods
While PRCA remains most frequently used for Kano model classification, it has been criticized for possibly being misleading when customer responses are skewed (S. Lin et al., 2010). To address this issue, S. Lin et al. (2010) proposed a moderated regression method, as shown in the following equation:
Here CS i represents the overall satisfaction of the ith customer, Xij is the performance level of the jth attribute rated by the ith customer, Zij is the moderator variable and m is the average or common performance used to classify Zij into three performance levels. Thus, the coefficient β* 1j represents the influence of the performance of the jth attribute on customer satisfaction, while β* 2j is the coefficient representing the interaction effect. S. Lin et al. (2010) suggest obtaining the change in the coefficient of determination (ΔR2 i ) between the two equations CS i = α j + β* 1j Xij and CS i = α j + β* 1j Xij + β*2jXij * Zij. When ΔR2i is not significant, then the jth attribute is classified as a one-dimensional factor. In other cases, the moderated regression coefficient β* 2j should be examined using the following rules: (1) if β* 2j > 0, then the jth attribute performance has a greater impact on overall satisfaction than at a low performance level and should be classified as an attractive factor; (2) if β* 2j < 0, then the jth attribute would be considered a must-be factor.
Though there is an improvement relative to the PRCA method, Lin and colleagues’ approach was criticized by L. F. Chen (2012) who pointed out that considering only one interaction item in the moderated regression approach could lead to a misclassification since the effects could be confounded regarding low and high performance. Additionally, there is still the issue of utilizing a continuous model with ordinal data.
Ridge Regression
To eliminate the flaws of the method proposed by S. Lin et al. (2010), L. F. Chen (2012) proposed a method utilizing Ridge Regression to avoid problems with multicollinearity. The new approach accounts for the interaction between attribute performance and CS at different performance levels, as well as for the issue of multicollinearity stemming from the use of multiplicative and cross-product terms in regression analysis. Chen’s approach is based on the following moderated dummy variable regression equation:
Here CS i represents the overall customer satisfaction of the ith customer, Xij is the performance level of the jth attribute rated by the ith customer, D1j is a penalty performance dummy variable, and D2j is a reward dummy variable for the jth attribute. The coefficients δ 1j and δ 2j represent the moderators in the low performance (penalty) and high performance (reward) conditions, respectively. In addition, two cross-product terms, Xij × D1 ij and Xij × D2ij, are included to identify the interaction effect of penalty and reward conditions on the relationship between attribute performance and customer satisfaction, and the coefficients, δ 3j and δ 4j , represent the effect of interaction on these relationships.
Based on the guidelines of Cohen et al. (2003) for synergistic/enhancing, buffering, and interference/antagonistic interactions, Chen recommends the following rules. If the coefficients of both the predictor and moderator are of the same sign, an increase in performance raises CS because of the fulfillment (reward) condition (β j δ 2j > 0), and a decrease in performance reduces customer satisfaction (β j δ 1j > 0) due to the non-fulfillment (penalty) condition. However, if one of the following situations exists, a strict and positive moderation effect is not supported, implying that an improved performance will not result in higher customer satisfaction, or that a decrease in performance will not result in lower customer satisfaction. L. F. Chen (2012) recommends examining the impact on CS in the following two conditions:
R-1. The predictor and moderator are of opposite signs: β j δ 1j < 0 or β j δ 2j < 0 (buffering interaction).
R-2. The coefficients of the predictor and cross-product term are of opposite signs: β j δ 3j < 0 or β j δ 4j < 0 (interference or antagonistic interaction).
Combining the observation of the impact on CS in these two conditions is utilized to classify attributes according to the Kano quality categories. The classification scheme is developed and constructed as an evaluation table, as shown in Table 5. To illustrate, if an increase in performance leads to higher customer satisfaction and a decrease in performance results in lower customer satisfaction, then the attribute is classified as one-dimensional. If an increase in performance leads to higher customer satisfaction and a decrease in performance does not result in lower satisfaction, then the attribute is classified as attractive. When neither the main or interaction effect exists, then no conclusion can be drawn, and a mixed-class distribution is present. However, the two most likely categories could be suggested, based on observation of the opposite fulfillment condition.
Rules for classifying quality attributes.
Notes: M: must-be, O: one-dimensional, A: attractive, I: indifferent, X: mixed class.
The categories inside the parentheses represent the two most probable categorie.
The reported overall accuracy of Chen’s Ridge Regression approach was 64.29% (18 out of 28), outperforming the PRCA and S. Lin and colleagues’ (2010) moderated regression method. Chen’s approach exhibited good classification power in predicting must-be and mixed-class quality attributes.
One of the weaknesses of Ridge Regression is the tradeoff between variance and bias. It may produce biased estimators in pursuit of lowering the variance of the obtained estimators. However, if the variance decrease is significant, this possible bias increase might be acceptable. Another obstacle in using Ridge Regression is the need to choose an appropriate value of k through the ridge trace, which is often a judgmental choice (Hoerl & Kennard, 1970). Additionally, the model assumes a particular form for the relationship between customer satisfaction and attribute performance, which may not be reasonable in some situations. The model also assumes the data is continuous when ordinal data is more often observed. Finally, while the estimated values of the coefficients could be utilized to prioritize the attributes, this would need to include the standard deviations in some manner, and it is not explicit in this formulation. The summary of regression methods is presented in Table 6.
Summary of Regression Methods.
Association Rule Method
Ahmad et al. (2012) proposed using conditional probabilities in a rule-based method to categorize attributes based on their effects on CS. Ahmad and colleagues’ (2012) method involves computing the significance of feature values (attribute performance values) based on the rationale that each feature value has different associations with different class subsets, and that the discriminating powers of feature (attribute performance) values:
(a) Basic/must-have features
If feature values have only two types of support sets (one with customer dissatisfaction values and another with customer satisfaction values), then the feature is a basic feature.
(b) Performance features
If feature values have different support sets that change gradually from strong dissatisfaction values to strong satisfaction values, then the feature is a performance feature.
(c) Excitement features
If some of the feature values have similar and mixed support sets with low discriminating powers and the remaining feature values have support sets with strong customer satisfaction values, then the feature is an excitement feature.
(d) Random features
If all the feature values have similar support sets consisting of customer satisfaction and dissatisfaction values and all the feature values have low discriminating powers, then the feature is a random feature.
Additionally, Ahmad et al. (2012) proposed the following algorithm to compute the support set and the discriminating power of a feature value:
Algorithm_discriminating_power
/* It computes the support set and the discriminating power of a filature value
Begin
{ If p(t/
{ with
Add t to
Else
end for
End
The central idea of the proposed rule-based method for categorizing features/attributes based on the rules above, which are grounded in support sets, and the subset for which the association reaches a maximum, is defined as the support set of the feature value. The method also implies that a significant feature is likely to have different support sets for different feature values, while this may not be so for an insignificant feature. Given a feature value, its relative frequency in different classes is used to measure the feature’s value-to-class association. Ahmad et al. (2012) present a detailed set of assumptions and guidelines for computing the significance of feature values, support sets, and discriminating powers, which can be found in their manuscript. Additionally, Chen (2015) provides guidelines and a good example of the method usage.
To sum up the strengths of the Association Rule method, it is designed for use with both ordinal and categorical data. Additionally, it does not assume any specific relationship between the categories and the overall satisfaction score; that is, it does not model the effects as being linear or assume any other specified model to describe the effect of the variables. For situations where the relationships are highly irregular or differ sharply from commonly considered models, this method could easily outperform its competitors.
Alternatively, on the weakness side, the Association Rule Method only considers increases or decreases and, as such, can miss out on information that may be explained more completely via a functional model (Ahmad et al., 2012). Additionally, it is often the case that the relationships between variables, while not exactly linear, behave linearly in subsets of interest. In such cases, the Association Rule would likely suffer a loss in power over methods that incorporate linearity or are model-based. Additionally, due to its complexity and requiring special programming skills, this method can be difficult to implement for many practitioners and academics. Finally, the Association Rule Method also does not provide a mechanism for prioritizing the attributes.
Artificial Neural Network Method
The Artificial Neural Network (ANN), also known as the back-propagation artificial neural network, is a computational model resembling human biological nerve properties, and operated based on a learning approach with examples (Garver, 2002; Larasati et al., 2011). The ANN family is based on utilizing Adaptive Resonance Theory (ART) to cluster arbitrary data into groups with alike characteristics (C. C. Chang et al., 2009).
ANN consists of three main layers— the input, hidden, and output layers—where the number of hidden layers is determined by the complexity of the data structure (see Figure 7). The core element in an ANN model is the “neural processing unit” or “neuron” located in the hidden layer (see Figure 5). Each neuron determines the optimal connection weight w = (w1,. .,wn) of each individual input through the learning algorithm set in the network. The neuron aggregates the weighted value from each input into a single value using summations. Then, the output is calculated by applying an activation function to the aggregate weighted value (West et al., 1997).
Within the ANN model, non-linear activation functions are used during the processing of the output neurons to link the weighted sums of neurons in one layer to the neurons in the succeeding layer. The type of activation function used in the model depends on the outcome range in the output layer. The most common activation function used in the ANN model is the Sigmoid function, which is a bounded non-decreasing and non-linear function similar to the logit function used in the logistic regression model (Larasati et al., 2011). The Sigmoid activation function can be expressed as:
where
Mikulić and Prebežac (2012) compare the ANN structure to regression such that the input neurons act like independent variables or predictors and the output neurons serve as dependent variables. However, unlike regression, predictors and dependent variables are connected by hidden layers determined by the path weights in the ANN model (Behara et al., 2002; Garver, 2002).
The performance of an ANN model depends on the parameters that are set during the construction process: the number of hidden layers, the number of neurons (nodes) in each layer, the activation function, and the learning algorithm (Michael et al., 2017). ANN models can be applied to solve clustering, classification, and association problems (Witten et al., 2016). The ANN model is classified as a supervised learning model when it is applied to solve classification and association problems, and defined as an unsupervised learning model when the model is used to solve a clustering problem. The most preferred supervised learning rule for training ANN models is the back-propagation (BP) algorithm (Philips et al., 2015).
ANN models have been applied to several Kano model studies (Bi et al., 2019; C. C. Chang et al., 2009; Mikulić et al., 2012). For example, C. C. Chang et al. (2009) applied an Adaptive Resonance Theory (ART) ANN model with an unsupervised algorithm and two layers for clustering customers with similar needs, then used the traditional Kano method for each identified cluster. The first layer contained the product attributes collected from customer reviews. The second layer represented the number of customer clusters.
Bi et al. (2019) applied ANN for the actual attribute categorization and used a back propagation neural network (BPNN) to identify the effect of positive or negative customer sentiments about product/service features on customer satisfaction. In this case, the weights were trained by utilizing the errors of the predicted results of the trained BPNN. The resulting Ensemble Neural Network Model, which measured the effects of customers’ positive and negative assessments on customer satisfaction, was obtained by aggregating the effects of the trained BPNNs and the weights of the trained BPNNs. The summary of the proposed ANN training framework is presented in Figure 13.
Bi et al.’s (2019) ANN Training Framework.
Bi et al. (2019) then suggested a categorization approach based on the obtained positive and negative weights,
where
In the next phase, Bi et al. (2019) proposed a categorization approach using the model and rules described below. It should be noted that Bi et al. (2019) referred to attributes as customer satisfaction dimensions (CSDs).
In Figure 14, part (a) represents the traditional Kano model. In part (b), an attribute performance (CSD) is represented by a curve, which in turn is represented by a point in (c) where the horizontal axis is defined as
If both ∣
(if
if
if
if
Bi et al.’s (2019) Categorization Model.
One of the critical issues when using ANN is the possibility of over-training, which means that a network has too many iteration processes, resulting in essentially memorizing a certain data set (W. J. Deng et al., 2008; Garver, 2002). Over-training often results in an over-fitted model, producing results that cannot be generalized to the entire population due to the data memorization. One potential procedure to prevent an overfitted model is to apply a cross-validation procedure (West et al., 1997). This procedure splits the data into a certain number of subsamples. Some subsamples are used as a training data set to build a neural network model, while the other subsamples are used to validate and test the model’s performance. Another factor that plays an important role in preventing over-training is the setting condition for terminating network learning, which depends on the architecture and learning algorithm of the network (Larasati et al., 2011).
Compared to other techniques, the ANN model has a higher ability to extract desired information from complex data patterns and to detect complex data trends. The adaptive system in the ANN model enables the model to change its structure based on external or internal information that flows through the network during the learning stage. ANN is also the best example of adaptive learning, which is self-organized and designed for real-time operations with a high error tolerance level (Gill & Mittal, 2016; Witten et al., 2016). Some of the ANN weaknesses include the need for a large data set for training, validating, and testing the model; the high sensitivity to both the parameter settings and the learning algorithm; and the requirement of very careful data preprocessing (Japkowicz & Stephen, 2002; Kotsiantis et al., 2007; S. Zhang et al., 2003).
Summary of Methods’ Fit
To facilitate the selection of methods aligned with hospitality and tourism academics’ specific investigative aims and constraints, Table 7 provides a summary of the reviewed methods in terms of fit for various data, objectives, and research types.
Summary of Kano categorization methods’ fit.
Discussion, Implications, and Directions for Future Research
The Kano model and its theory of attractive quality remain of much interest to academics and practitioners due to the continuous nature of product or service quality improvement to achieve better customer satisfaction. Those who are pursuing the goal of identifying an optimal set of product/service attributes to yield maximum CS adopt the Kano model because it provides valuable information for achieving this goal (Mikulić & Prebežac, 2011; Velikova et al., 2017).
Nevertheless, the challenges of identifying the relations between attribute performance and CS and producing the optimal set of CS drivers remain present (Bi et al., 2019; Pandey et al., 2022; Tontini et al., 2017). Although the number of studies using the Kano model has been on the rise in recent years, there has also been a growing need for better quantitative evaluation and decision support for the model application (Hu et al., 2020; F.-H. Lin et al., 2017; Madzík, 2018). To answer this call, academics who have adopted the Kano model have been trying to improve its classification ability, often by integrating it with other tools (Ahmad et al., 2012; Bi et al., 2019; L. F. Chen, 2012; Matzler & Hinterhuber, 1998; Matzler et al., 2004; Mikulić et al., 2012).
As shown in this manuscript, no classification approach is free of limitations, and academics and practitioners should be aware that these classifying approaches can produce different categorizations. Mikulić and Prebežac (2011) warn that in many instances these multiple approaches should be regarded depending on the context and goals in question. One of the key findings of this manuscript is that the method selection should be circumstantial and account for such factors as:
• the life stage of the product and the purpose of classification.
• the type of available data.
• technical and analytical resources.
• the quality of data.
Several important implications stem from this notion. As hospitality and tourism professionals continue to display a strong interest in identifying product/service attributes that drive satisfaction, and categorizing product/service attributes according to the Kano model, they should become familiarized with and choose the most progressive methods that fit their objectives. As was stated in the Introduction section, the majority of HT academics continue using the original approach, and only a few studies used such advanced methods as moderated regression, the Association Rule method, or the neural network approach. That has several negative consequences for hospitality and tourism research. Using outdated categorization methods can significantly erode the quality and accuracy of findings and can result in subpar research outcomes and misguided strategies, ultimately diminishing the ability to understand evolving customer preferences due to the compromised accuracy of attribute classification. Additionally, relying on flawed methods may tarnish the image of hospitality and tourism research as such obstinacy in the face of evolving tools may inadvertently diminish its reputation for methodological rigor and relevance. As contemporary research domains are marked by advanced techniques, such reluctance to adapt may impede the Hospitality and Tourism field’s capacity to produce robust findings, thus compromising the discipline’s esteem within the broader scholarly community.
At the same time, HT academics should be aware that the Kano methods can provide potentially different categorizations and should regard those methods with caution and choose the one that fits best based on their goals and conditions. For example, the original Kano method may perform well during the design stage of product/service development because it produces the categorization based on the potential feature’s ability to elicit satisfaction or dissatisfaction if present or absent. Thus, when introducing new features, the original method can illustrate how customers view an attribute at the individual level. However, when the level of performance and importance of attributes are sought, regression methods may perform much better, particularly considering the temporal nature of the Kano model. All regression methods can be applied to existing attributes to show the effects of actual performance on overall satisfaction with a product or service. Additionally, the produced indices can be summed to obtain a measure of an attribute’s impact on overall customer satisfaction. When ranking is one of the pursued goals, logistic regression may present a better choice, as it provides a mechanism for ranking attributes based on their impact on customer satisfaction. In cases where the interaction between attribute performance and customer satisfaction should be accounted for, or where there are concerns about skewness, then moderated regression or ridge regression methods should be applied to achieve better classification results.
For academics and practitioners dealing with ordinal and categorical data in situations when relationships among variables appear highly irregular, the Association Rule method may outperform other methods. However, this method is complex and novel and, thus, some academics and practitioners may not have the skills or tools to implement this method. Similarly, the ANN method might not be easily adopted due to its novelty and the skills necessary for its implementation. Nevertheless, when there is a need to use different types of data in conditions of indefinable non-linearity, ANN might be the best method to choose. When making decisions on existing products/services using actual attribute performance data, attribute-performance-based methods like ridge regression, and so forth, are recommended. A shortcoming of those methods, however, is that they can be complex and require advanced skills and software packages. Additionally, some of those methods work best for a particular type of data or are susceptible to such issues as multicollinearity or skewness.
From a practical perspective, in the competitive landscape of hospitality and tourism, practitioners can derive substantial benefits from applying Kano categorization methods as this allows them to get a comprehensive understanding of qualities driving customer satisfaction and tailor their strategies to achieve an optimum combination of product or service features. Such understanding proves invaluable in resource prioritization, where businesses strive to stand out and efficiently allocate resources for maximum impact on customer satisfaction. This strategic focus is pivotal for gaining a competitive edge and ensuring the continued relevance of services and products. Satisfied and delighted customers are the lifeblood of the industry, contributing significantly to repeat business and brand loyalty. The dynamic nature of the tourism and hospitality industry necessitates adaptability. Kano categorization methods provide a means to comprehend and respond to shifting dynamics in customer preferences. This adaptability allows businesses to stay ahead of emerging trends, incorporate necessary advancements, and address evolving customer needs effectively.
In addition to the benefits described above, practitioners can use this method review to choose the categorization approach best suited to their specific circumstances. The hospitality and tourism industry often faces operational constraints like budget limitations and feasibility of analysis. Knowing which Kano method works best under different constraints empowers practitioners to make informed decisions. For example, small independent restaurants with tight budgets might opt for the simpler PRCA approach, while large hotel chains can invest in more advanced ANN models. In essence, by selecting the method that fits their unique situation, practitioners can optimize resource allocation, respond to changing dynamics, tailor services, make decisions given limited resources, drive innovation, and increase customer satisfaction and loyalty. This knowledge of the most appropriate tool becomes critical for sustained success in a dynamic and competitive business.
As Kano model research continues to grow in the field of Hospitality and Tourism, a warning should be established based on the worrisome stream of studies that used the Kano model inappropriately, taking liberties in its interpretation. For example, Kaya (2022) substitutes attribute performance for customer needs and uses the importance scale as a measure of attribute performance. The Kano model and methods do not involve measures of implicit importance and view a product or service as a bundle of attributes/features, not customer needs. Tsang et al. (2022) measures attribute performance/fulfillment with a Likert-type “agree/disagree” scale and makes categorization decisions based on the values of beta-coefficients from a regression equation where CS is a dependent variable and attribute measures serve as dependent variables, which completely contradicts not just the original Kano method, but all regression categorization methods as well.
Additional troublesome misuse of the Kano model and methods comes from the studies involving Big Data and sentiment analysis, which use online reviews as proxies for attribute performance and ratings as measures of CS (Y. Chen et al., 2022; Lu et al., 2023; Zhao et al., 2024; J. Zhang et al., 2018).
Using customer reviews as a measure of attribute performance introduces significant challenges related to validity and reliability. While these reviews can offer insights into customer experiences, they often exhibit selection bias, as extreme opinions tend to dominate, potentially misrepresenting the general sentiment. Also, using sentiment strength as a measure corresponding with the level of attribute performance or level of fulfillment seems problematic as it might be tied to the semantic preferences of customers and how they tend to express themselves, not their actual product or service attribute performance rating. Some reviewers may be accustomed to stronger verbiage which may not reflect the actual experience.
Serious reliability issues may stem from reviewer bias occurring when emotions and personal biases distort the accuracy of reviews, leading to an unreliable depiction of the actual performance. The temporal aspect is also crucial: reviews might not capture the current performance status, potentially magnifying or downplaying achievements or shortcomings. Manipulation of reviews by competitors further compounds the reliability challenge, introducing potentially false information that skews evaluations. Moreover, online reviews tend to lack a comprehensive assessment of all relevant product or service features and include only those that reviewers were particularly pleased or displeased with. Such fragmented assessment goes against the Kano model principle, which implies that a single customer/reviewer evaluates the entirety of product/service features and makes an assessment of how those features make them feel, or impacts their overall satisfaction. Moreover, the demographic and psychographic characteristics of reviewers might not accurately reflect the wider customer base, raising concerns about the representativeness of the feedback.
Two recent Hospitality and Tourism studies can be used as a cautionary tale for the problems of misusing the Kano Theory while using online reviews for categorization. Y. Chen et al. (2022) and Zhao et al. (2024) use positive and negative reviews as a proxy for good and bad attribute performance, thus subjecting their studies to the issues described above. In addition to that, Zhao et al. (2024) treat product/service attributes as requirements, which seems unfitting and goes against the nature of the Kano model. For instance, attractive attributes in the Kano classification are good to have, but not necessary, as they do not cause dissatisfaction when absent or performed poorly, or the indifferent category attributes cannot by any means be called requirements as those do not have any impact on CS. Furthermore, Zhao et al. (2024) utilize the strength and frequency of customer requirements/sentiments as a cornerstone for their proposed categorization. The Kano model is a framework that categorizes features or attributes of a product or service into different groups based on their impact on satisfaction and dissatisfaction. The model does not rely on the quantity or intensity of customer sentiments. Instead, it focuses on the nature of customer preferences and how they are fulfilled.
In addition to the questionable proxies for attribute performance, the two studies in question have a problem with measures of satisfaction. For example, Y. Chen et al. (2022) use hotel star ratings as a proxy for satisfaction, which is conceptually incorrect and goes against the Kano model, which does not imply such substitutes. Alarmingly, Zhao et al. (2024) do not employ any measures or proxies of CS. Therefore, the method they propose cannot be called an improved Kano model as it does not involve the major outcome of the Kano theory—customer satisfaction. Given these concerns, Hospitality and Tourism academics should be cautious about deviating from the true essence of the Kano model. While customer reviews are easily available and offer valuable qualitative insights, more comprehensive and objective quantitative measures are necessary for proper Kano model assessment.
To sum up, the current study aims to advance the effective and proper utilization of Kano categorization methods in the HT field, which is essential for raising commitment to high-quality research and the prestige of HT studies. It marks a progression towards academic rigor and relevance, establishing the field as an authoritative and influential domain. The shift towards advanced methods would underscore HT’s dedication to innovation, adaptability, and continuous improvement. Moreover, learning from the methodological errors in the studies described above serves as a vital reminder of the risks associated with inappropriate methodological choices. A thorough review of Kano categorization methods is, therefore, more than an academic pursuit; it is a critical step toward reinforcing the stature and impact of HT research in shaping and understanding customer experiences in the service industry.
Current Gaps and Directions for Future Kano Model Research
While certain improvements and progress have been made in the field of Kano categorization methods, there is still a significant gap in the empirical evaluation of these methods in terms of their categorization abilities. As of now, no empirical study has conducted a comprehensive and rigorous comparison of all the Kano categorization methods in terms of their categorization precision and relevance. Therefore, it is highly recommended that future research endeavors address this issue to provide more robust insights into the effectiveness and applicability of these methods. For instance, a dataset that perfectly reflects Kano model categories can be used to test the accuracy of the available categorization methods. By conducting such empirical evaluation, academics can identify which methods outperform others under specific circumstances and confirm the relative strengths and weaknesses of each approach.
It is commonly agreed that products and services are distinct domains. Services are characterized by intangibility, heterogeneity, and the involvement of customers in the creation process, which may present unique impacts on customer satisfaction. In contrast, products, with their tangible and often standardized nature, offer different conditions for applying Kano’s model. In current academic research, the application of Kano’s categorization in services versus products remains an area for exploration. This gap is particularly significant given the fundamental differences between the two domains. The distinct characteristics of services and products suggest that customer satisfaction drivers identified by Kano’s Model might manifest differently, necessitating a separate examination in each context. Furthermore, services often involve a higher degree of customization and personal interaction, which could significantly affect how customers perceive and value different aspects categorized by the Kano model. This variance in perception is less pronounced in products, where quality and features are more objectively measured. The subjective nature of service quality, shaped by customer experiences and interactions, therefore demands a more nuanced application and interpretation of Kano’s model, and there is a need to understand how the dynamic and interactive nature of services influences the Kano categories.
Lastly, the academic literature would benefit from a deeper exploration of how feedback mechanisms and continuous improvement processes in services influence the Kano categorization differently compared to products. Given the direct and immediate nature of service feedback, the impact on customer satisfaction and the subsequent categorization of service attributes may be more fluid and subject to rapid change. This contrasts with products, where longer development cycles might delay the reflection of customer feedback in Kano’s categorization. Bridging these research gaps can provide valuable insights for both academic theories in consumer behavior and practical applications in service design and product development.
Additionally, future studies should explore the practical implications of employing different Kano categorization methods in real-world settings. Understanding how these methods perform in actual Hospitality and Tourism contexts can provide valuable guidance for organizations seeking to enhance customer satisfaction and drive product/service improvements. Field studies and case analyses that apply these methods in diverse contexts and cultural circumstances can yield valuable insights into the generalizability and versatility of the Kano model’s categorization approaches.
In addition to empirical evaluations, future research should also focus on addressing the theoretical problems, terminological inadequacies, and conceptual issues identified in this study. By refining and enhancing the underlying theoretical framework, academics can strengthen the foundation of the Kano categorization methods and ensure their continued relevance and accuracy in understanding customer preferences and satisfaction. Furthermore, as technology and customer expectations continue to evolve, it is crucial for future studies to explore innovative adaptations and extensions of the Kano model and categorization methods. For instance, academics could investigate the integration of emerging technologies—such as artificial intelligence and machine learning—to enhance the categorization process and predictive capabilities of the model.
In closing, while the current study lays a solid foundation for HT academics by providing a comprehensive overview and critical analysis of Kano categorization methods, its focus on categorization methods only can be viewed as narrow, thus leaving much room for advancement in this field. Future research should focus on empirical evaluations, practical applications, theoretical refinements, and technological adaptations to further enrich the understanding and application of the Kano model. By addressing these recommendations, HT academics can contribute to a more sophisticated and practical understanding of product or service development and its impact on CS, benefiting both HT academia and industry alike.
Footnotes
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
