Customer classification: A Mamdani fuzzy inference system standpoint for modifying the failure mode and effect analysis based three dimensional approach

https://doi.org/10.1016/j.eswa.2021.115753Get rights and content

Highlights

  • A fuzzy logic based approach is proposed for customer classification.

  • Mamdani MISO FIS is utilized to modify a FMEA based customer classification methodology.

  • Marketing practitioners should not classify customers regardless of the context – market type – they are planning for.

  • They may encounter four potential market scenarios, including monopoly-business, competition-business, monopoly-consumer and competition-consumer markets.

Abstract

Customer categorization using a three dimensional loyalty matrix, based on failure mode and effect analysis (FMEA), is an innovative approach for customer classification but is vulnerable to FMEA limitations. The main purpose of this research is to utilize a multi input single output Mamdani fuzzy inference system (FIS) to cope with the traditional FMEA inherited shortcomings. Besides, the classification logic and classes of the Loyalty Matrix methodology have been adopted for the purpose. We have also identified four potential market scenarios and evaluated the performance of the proposed methodology within these contexts. Correspondingly, four tailored FIS’s consisting of a total of 108 fuzzy rules have been developed. Empirical results indicate that the new approach successfully resolved serious issues such as data uncertainty, weight ignorance, the same output value computation from different input values and the discontinued output.

Introduction

Of the paramount characteristics ubiquitous amongst almost all thriving enterprises is offering class specific value prepositions to each category of customers. Such a strategy is usually initiated with customer classification, a concept evolving to be integral to marketing and being applied by numerous firms to meet customer needs more appropriately (Floh, Zauner, Koller, & Rusch, 2014). Customer classification is also a prerequisite for targeting the best customer segment (Madzík & Shahin, 2020) and predicting the customer churn (Abbasimehr, Setak, & Soroor, 2013). Literature review indicates that a considerable number of such classification efforts have been made through the lens of the customer loyalty level.

Loyalty is a strong commitment to consistently re-purchase or re-patronize a preferred product or service in the future, which, as a result, leads to repetition of buying the same set of the brand, in spite of the fact that situational impacts and marketing attempts may potentially result in switching behavior (Oliver, 1999, p. 34). This definition elucidates the fact that loyalty serves as a shield protecting sellers against their rivals (Mascarenhas, Kesavan, & Bernacchi, 2006), and (truly) loyal customers are committed to keep re-purchasing regardless of circumstance (Narayandas, 2005). Moreover, such customers are greatly cooperative and willing to provide positive word of mouth and recommendation (Tanford & Baloglu, 2013).

While there is not a universal approach for measuring loyalty, literature shows several methods employed for the purpose, such as Net Promoter Score (Reichheld, 2003), Loyalty Ladder (Narayandas, 2005, Mascarenhas et al., 2006), Bandyopadhyay and Martell’s (2007) framework for consumer brand loyalty classification, and Loyalty Matrix used by Tanford and Baloglu (2013). In a recent study, Madzík & Shahin (2020) places a more explicit emphasis on including behavioral factors of loyalty as necessary supplements to the attitudinal one. In other words, inspired by the Risk Priority Number (RPN) calculation from multiplication of three risk factors in Failure Mode and Effect Analysis (FMEA) – a well-known technique in risk assessment and quality management – they proposed computing the Loyalty Priority Number (LPN) from the product of three loyalty dimensions: customer purchase value (V), purchase frequency (F) and loyalty (L). The LPN was, in turn, utilized as an overarching index for customer classification. However, the multiplication formula of basic FMEA, adopted in the Madzík & Shahin’s (2020) Three Dimensional Customer Classification (TDCC) methodology, has widely been criticized in literature. The TDCC limitations, together with their corresponding FMEA roots (Liu et al., 2012, Liu et al., 2013, Geramian et al., 2017, Geramian et al., 2019, Geramian et al., 2020), are summarized in Table 1.

Since these issues are of a FMEA nature, the present research utilizes the most popular approach for modifying the traditional FMEA shortcomings, this time, to resolve the TDCC limitations. This popular solution is the expert system and artificial intelligence approach of Fuzzy Inference System or FIS (Liu et al., 2013, Geramian et al., 2017, Geramian et al., 2019, Geramian et al., 2020). FIS is capable of resolving almost all of the abovementioned issues (Geramian et al., 2017, Geramian et al., 2019, Geramian et al., 2020). Moreover, as ratings of TDCC loyalty dimensions are determined through respondents’ judgment and prone to uncertainties hidden in human judgments, application of the fuzzy sets theory (Zadeh, 1965), in general, and FIS fuzzification (Geramian et al., 2017, Geramian et al., 2019, Geramian et al., 2020) and fuzzy numbers (Shahin, Barati, & Geramian, 2017), in particular, are of tremendous advantages in this regard. With respect to the fact that TDCC is of three inputs (loyalty dimensions) and one output (LPN), along with a questionable or unknown inputs to output relationship, we draw on a Multi Input Single Output (MISO) FIS of the Mamdani type (Mamdani and Assilian, 1975, Geramian et al., 2019, Geramian et al., 2020). Using a type 1 fuzzy system – versus more enhanced versions such as the type 2 fuzzy logic – the present study will act as an initial step in application of fuzzy inference in the TDCC problem. Therefore, it is mainly application oriented, by nature.

As a result, the objective of the present research is to utilize the Mamdani MISO FIS methodology to modify TDCC. For this purpose, first we identify four potential market types, as contextual scenarios, which customer classification analysts may encounter. Then, corresponding to them, four tailored FIS’s based on TDCC are developed. It is noteworthy that classes/categories of the proposed approach are designed on the basis of the Loyalty Matrix methodology used by Tanford and Baloglu (2013). Implementation of the developed approach in the identified scenarios facilitated highlighting outperformance of FIS based TDCC over basic TDCC from several aspects.

Academically, this study will equip the customer classification function of customer relationship management with the expert and intelligent system of FIS. To the best of the authors’ knowledge, the present study is the first research applying the fuzzy logic and, in particular, the Mamdani MISO FIS, to modify the FMEA based customer classification. In other words, such application will bring a higher level of accuracy to this specific manner of the classification problem through modifying the drawbacks of its underlying method – the traditional FMEA. As a result, integration of both the behavioral and attitudinal dimensions into a single classification oriented index will be carried out much more reliably.

Practically, it will shed light on the fact that the context matters when classifying customers. Identifying the potential contextual factors, this paper will draw a comprehensive picture of market scenarios that practitioners may face. Besides, it will provide four FIS’s, with each being tailor made to classify customers in a specific scenario. Accordingly, this research proposes a viable methodology that helps companies tailor their offerings and marketing strategies towards each specific customer class.

Section snippets

Loyalty based customer classification approaches

By measuring the extent to which a customer is willing to recommend a given company to others, Reichheld (2003) developed Net Promoter Score as a basis for customer categorization. The measure is computed via subtracting the percentage of detractors from that of promoters and used to classify customers into detractors, passively satisfied and promoters. Narayandas (2005) categorizes the customers’ loyalty behaviors in the business market setting into a hierarchy of levels called Loyalty Ladder.

Three dimensional customer classification

Madzík & Shahin (2020) developed TDCC, the general mechanism of which is based on the traditional FMEA. FMEA is an analytic approach for definition, identification and elimination of potential and/or known failures, errors and so forth from design, service, process and system, prior to the items reaching the customer (Stamatis, 1995, Liu et al., 2012). In traditional FMEA, risk of failures is prioritized through the RPN index, calculable via Eq. (1) (Qin, Xi, & Pedrycz, 2020).RPN=S×O×D

Where S,O

Proposed approach: FIS based three dimensional customer classification

Our proposed methodology has a framework (Fig. 1) inspired by the fuzzy FMEA methodology (Geramian et al., 2019, Geramian et al., 2020). In Fig. 1., as the direct relationship between the loyalty dimensions (V, F and L) and LPN is questionable in TDCC, it is depicted using a dotted arrow between crisp values of these inputs and the output value, which is indicated using the Defuzzified form of a Fuzzy LPN (Defuzzified FLPN).

Instead, a mapping between them can be established through FIS,

Application of the approach

In order to investigate applicability and effectiveness of the proposed methodology, it is applied in the context of some scenarios by considering several assumptive customers, each being characterized by three assumptive values for the loyalty dimensions. This provided us with the chance of implementing our proposed approach in comparison to TDCC, followed by findings and discussion.

Practical implications

This study sheds light on the fact that marketing managers should carry out the customer classification task with respect to the context – market type – they are planning for; otherwise, they may end up with misleading results. In competitive markets, managers should pay much more attention to attitudinal loyalty factors. In such contexts, a given company’s customers with only high behavioral loyalty – purchase value or frequency – indication may easily stop buying from the company because they

Conclusions

In the present research, we modified TDCC, an approach recently developed to classify customers, by means of the MISO Mamdani FIS and, in part, Loyalty Matrix. Benefits of such modification are extensive: tackling human judgment uncertainties extant within input data; providing both within and between scenario discriminability; considering relative importance weights of the loyalty dimensions; facilitating a relaxation of the questionable multiplication formula; introducing a gap free output

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. The authors are really thankful to the anonymous reviewers whose invaluable comments facilitated substantial improvement in the quality of this research.

References (41)

Cited by (8)

  • A novel similarity measure in intuitionistic fuzzy sets and its applications

    2022, Engineering Applications of Artificial Intelligence
    Citation Excerpt :

    These theories have their own advantages in dealing with uncertain information. Therefore, researchers have applied those theories to various fields, such as decision making (Aydın et al., 2020; Gao et al., 2020a; Liao et al., 2021), target classification (Luo and Zhang, 2020; Geramian and Abraham, 2021; Fu et al., 2020), clustering problem (Egrioglu et al., 2020; Budwong et al., 2021; Ding et al., 2021) and other mixed areas (Lopes et al., 2020; Ma et al., 2021; Solanki et al., 2021; Gao et al., 2020b). Since IFS were proposed, similarity measure in the IFSs has been studied, such as Chen firstly defined the similarity measure between IFSs (Chen, 1995).

View all citing articles on Scopus
View full text