An adaptive nonlinear approach for estimation of consumer satisfaction and loyalty in mobile phone sector of India

Journal of Retailing and Consumer Services 21 (2014) 570–580

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services journal homepage: www.elsevier.com/locate/jretconser

An adaptive nonlinear approach for estimation of consumer satisfaction and loyalty in mobile phone sector of India T. Rahul n, R. Majhi 1 School of Management, National Institute of Technology, Warangal 506004, India

art ic l e i nf o

a b s t r a c t

Article history: Received 26 November 2012 Received in revised form 10 February 2014 Accepted 21 March 2014

To facilitate business growth assessment of customer's, satisfaction and loyalty levels in mobile sector are two important issues which need in-depth investigation. These two levels of customers are nonlinearly related to their corresponding attributes. The past studies have mostly assumed linear relation and have mostly used regression based models for estimation of these levels and the results are not encouraging. To overcome this limitation, the present study has developed simple nonlinear models for accurate estimation of these two parameters using their related key factors and results obtained are shown to be much better. This paper has also observed the positive effect of satisfaction on the loyalty estimation of customers. Employing the proposed nonlinear adaptive models, the service provider can also predict the satisfaction and loyalty levels of each of its customers which help the organization to determine the number of possible future churners. & 2014 Elsevier Ltd. All rights reserved.

Keywords: Customer satisfaction Customer loyalty Telecom sector Nonlinear adaptive models

1. Introduction India is the second largest and fastest growing mobile market in the world. The tele-density at the end of March 2011 reached to 70.89 when compared to 52.74 in March 2010 (TRAI annual report 2010–11). The number of mobile phone users in the country is expected to reach around 1.26 billion by the end of 2014 with a penetration of 97% (report from iSuppli mobile industry forecasts). The significant rise in mobile phone use is partially attributed to the presence of one of the largest young population in the world. Hence, India is a profitable market place for mobile phone service providers. But this penetration in market is below the EU average which implies that still there is tremendous scope of expansion of business in this area. Currently more than fifteen service providers are operating in Indian mobile phone sector. These service providers are attracting customers by providing competitive offers. However retaining customers have become a great challenge for the companies after the government has introduced features like mobile number portability. The ability to provide high degree of satisfaction is crucial for differentiating themselves from their competitors (Deng et al. 2010) specifically in telecommunications market. Therefore satisfying a customer requirement and understanding the level of customer loyalty are two important aspects of a company and need

n

Corresponding author. Tel.: þ 91 8978264848. E-mail addresses: [email protected] (T. Rahul), [email protected] (R. Majhi). 1 Tel.: þ91 870 2462852. http://dx.doi.org/10.1016/j.jretconser.2014.03.009 0969-6989/& 2014 Elsevier Ltd. All rights reserved.

an in-depth study. Customer satisfaction has a direct influence on customer retention (Choi et al., 2008; Hansemark and Albinsson, 2004). Hence it is important to understand the level of customer satisfaction and loyalty to improve the competitiveness of the firm. The rest of the paper is organized as follows: Section 2 reviews the literature on customer satisfaction and loyalty studies. Sections 3, 4 and 5 provide the research gap, sampling and instrument. In Section 6 the development of adaptive models for satisfaction and loyalty are dealt in-depth. Finally in Section 7 the managerial implications and in section 8 the limitations of the study are presented.

2. Review of literature 2.1. Customer satisfaction Customer satisfaction refers to psychological state resulting when the emotions surrounding disconfirmed expectations are coupled with consumers' prior feelings about consumption experience (Oliver, 1981). It is often considered as an important determinant of repurchase intention (Liao et al. 2009) and customer loyalty (Eggert and Ulaga, 2002). It is a burning important research topic in the information system area (Au et al., 2008). The ability to provide a high degree of customer satisfaction services is crucial to service providers in differentiating themselves from their competitors. If the customer have good experiences in using mobile phone services, then he/she will have cumulative customer

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

satisfaction. Since the customer satisfaction reflects the degree of a customers' positive feeling for a service provider in a mobile phone context, it is important for them to understand the customer's opinion about their services. The literature review reveals that long-term success of a firm is based on its ability to respond to changing customer needs and preferences (Narver and Slater, 1990; Webster, 1992). Higher customer satisfaction leads to stronger competitive position resulting in profitability of a firm (Fornell, 1992) and also helps in lowering the business cost and the cost involved in attracting new customers (Chien et al.,2003). Satisfied users will have higher usage level of mobile services than those who are not satisfied and they are more likely to possess stronger continuous intention and recommend to their friends and relatives (Zeithaml et al., 1996). 2.2. Customer loyalty Customer loyalty is company's important asset. By maintaining customer loyalty, a firm develops long term mutually beneficial relationship with customers (Pan et al., 2012). It is found to be a key determinant of a long term viability of a brand (Krishnamurthi and Raj, 1991). In addition, compared to loyal customers, the others are much more influenced by negative information about the products or services (Donio et al., 2006). Therefore retaining existing customers and strengthening customers' loyalty appear to be very crucial for mobile service providers to gain competitive advantage. In increasingly competitive markets, building strong relationship with customers i.e., developing loyalty with consumers is considered as one of the key factors in winning market share and developing competitive advantage (Laurn and Lin, 2003; Nasir, 2005). Loyal customers are crucial for business survival (Semejin et al. 2005) because attracting new customers is considerably more expensive than retaining old ones (Reichheld and Schefter, 2000). In a recent study, authors (Deng et al., 2010) have measured the customer loyalty as the customer's behavioral intention to continuously use mobile instant message with their present service providers, as well as their inclinations to recommend the mobile instant message (MIM) tool to other people. If a service provider can satisfy the needs of the customer better than its competitors, it is easier to develop loyalty (Oliver, 1999). 2.3. Linkage between satisfaction and loyalty In the last decade, research on customer satisfaction and customer loyalty has gained increasing importance in both online and off-line business. Therefore enterprises attempt to increase their market share by maximizing customer retention (Tsoukatos and Rand, 2006). The potential and opportunity value of customers earned over a long period is another advantage to maintain existing customers (Seo et al., 2008). A high level of customer's satisfaction has positive impact on customer loyalty (Mittal et al., 1998). According to the findings (Sivadass and Baker-Prewitt, 2000), customer loyalty is the ultimate objective of customer satisfaction measurement. Further with the advancement of information technology customers are becoming more and more open to understand the brand, thus satisfaction alone may not be adequate to retain a long term relationship (Kassim and Abdullah, 2008). Accordingly, it is important for the service providers to identify appropriate factors that impact users' satisfaction and loyalty which would help to take proper measure to retain them. Perceived service quality and customer value serve as drivers of customer satisfaction (Lim et al., 2006). It is expected that this relationship may also be significant in the context of telecom service providers of India. The mobile phone users often choose the providers they trust to deal with. Trust can also be seen as a

571

critical factor for consumers to build and maintain relationships with providers (Semejin et al., 2005). Satisfaction has always been viewed as the main input for the customer loyalty. However, the satisfied users may switch to another brand because of the low switching costs (Lam et al., 2004). As a result customer's satisfaction is one of the important predictors of customer loyalty of service providers. Other moderate influences on loyalty are due to customer characteristics such as age, gender, usage and experience. Fornell (1992) has opined that high customer loyalty is mainly caused by high customer satisfaction whereas Clarke (2001) proposed that effective satisfaction must be able to create loyalty amongst customers. Previous studies have also shown that customer satisfaction positively affects customer loyalty (Choi et al., 2008) or negatively affects switching intention (Walsh et al., 2006). If a customer is dissatisfied with a service provider because of low service quality or other such factors, then he/she would more likely change to another. A few satisfied customers may complain of poor service experience but would not switch. Satisfaction and loyalty are not surrogates of each other (Oliver, 1999). It is possible for a customer to be loyal without being highly satisfied and vice versa. Firms need to gain a better understanding of the relationship between satisfaction and loyalty which would help to distribute their marketing efforts between satisfaction initiatives and loyalty programs. For instance if the firms find that loyalty is associated with increased satisfaction then they can directly focus on enhancing their loyalty programs (Gerpott et al., 2001; Kim et al., 2004; Lai, 2004; Lin and Wang,2006; Turel and Serenko, 2006; Wang and Liao, 2007). In the recent past several studies have been conducted to understand customer satisfaction and loyalty particularly on mobile services customers. Most of these studies emphasize that analysis of factors affecting customer satisfaction are important for the success of mobile service firms. Furthermore they have reported that customer satisfaction is the main important goal for mobile service providers to obtain economic success. Because of tremendous potentiality of business in this sector, the academic interests in mobile phone usage have also increased significantly. These includes contribution to social life, user preferences and organic features; examination of users' motivation (Dedeoglu, 2004; Ozcan and Kocak, 2003), mobile phone selection (Isiklar and Buyukozkan ,2007) and brand loyalty (Simsek and Noyan, 2009) etc. The users' perception of satisfaction and loyalty in India is different from that found in other countries. Therefore there is a need to explore factors influencing customers' satisfaction and loyalty of telecom service providers particularly in Indian scenario. Some authors have used structural equation modeling (SEM) (Kim et al., 2004) and neural networks (Goode et al., 2005) for similar studies. However the authors who have used neural networks have used multilayer artificial neural network based models for prediction purpose. Such models offer more computational complexity and consume more training time. Hence, there is a need to develop lower complex artificial neural network based models for prediction of satisfaction and loyalty which is more accurate and offer lesser training time.

3. Research gap Existing literature survey reveals that satisfaction and loyalty are highly correlated and satisfaction plays a positive influence on customer loyalty. Further the customer satisfaction primarily relies on connectivity, consumer service, customization and branding attributes of the customers. Similarly it has been demonstrated that the loyalty of customers is mainly influenced by reliability and trust factors. For achieving their results the authors have employed some conventional methods such as structural equation

572

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

model. Similar observations on customer satisfaction and loyalty can also be made by alternative approach of developing adaptive linear model as well as nonlinear prediction models based on artificial neural networks. To the best of our knowledge such an approach has not been suggested in previous publications on the topic. In this paper, an attempt has been made to fill up this research gap by proposing simple linear and nonlinear neural network based prediction models for satisfaction and loyalty. To achieve this, the attributes which influence the satisfaction and loyalty levels pertaining to the mobile phone service sector have first been identified. Then using the factor analysis, the attributes have been reduced to obtain the predominant factors that have direct bearing on satisfaction as well as loyalty. It is a fact that the customer satisfaction level is non-linearly related to the key attributes of the customers. Similar is the relation between the loyalty and its associated attributes. Therefore it is interesting to develop new models to demonstrate the nonlinear relationship between the key attributes with satisfaction as well as with loyalty levels. Keeping this in mind three Hypotheses have been proposed. Hypothesis 1: Key factors such as adaptability, customer service, connectivity, skills of the employee and advertisement are non-linearly related to satisfaction of a customer and hence its value can be accurately estimated by using a non-linear model taking these five factors as input to the model. Hypothesis 2: Key factors such as reliability, repurchase intention and continuity in usage are non-linearly related to customer loyalty and hence its value can be accurately predicted using a non-linear adaptive model taking these three factors as input. Hypothesis 3: The satisfaction value of a customer positively influences its loyalty level. This hypothesis can be tested if the estimated satisfaction of the customer obtained through the model is used as an additional input in the non-linear model for loyalty prediction, the revised estimation than the value should be closer to its actual value obtained from the customer's response. The main focus of this study is to develop efficient nonlinear models to estimate the customer satisfaction and loyalty for existing mobile service providers in India. The findings of the present investigation would help in providing some specific recommendations to practitioners and would offer valuable insights for carrying out future research in the area of customers' satisfaction and loyalty.

4. Sampling To estimate the satisfaction and loyalty levels using non-linear prediction models judicious choice of customers of mobile phone service providers is important for the purpose of analysis. Since the study was made in India, one of its largest states namely Andhra Pradesh was selected to collect relevant data for analysis. It was assumed that the results obtained from analyzing the data obtained from such a vast population would also be applicable to the customers of other parts of India. Most of the mobile phone service providers render their best possible services to the customers residing in the cities. Based on this reasoning our collection of data from the customers was limited to the main cities of Andhra Pradesh. Our targeted population was the city dwellers of all age groups so that the data collected from them would help to generate appropriate non-linear models for estimation of satisfaction and loyalty levels. The questionnaires were distributed to 300 people consisting different age groups out of which 240 responded. However 29 people did not respond to all our questions. Hence the

completed data collected from 211 customers were used in the investigation.

5. Instrument design The research objective was to estimate satisfaction and loyalty of mobile phone users. The questionnaire administrated for research was divided into four sections. The first section consisted of general questions on usage of a mobile service. The second section comprising of 25 questions was on satisfaction and 15 questions of section three were devoted to loyalty. The responses were taken on five point Likert scale, with one denoting strongly disagree and five denoting strongly agree. The last section consisted of questions on demographic factors of a consumer.

5.1. Data reduction Since the number of attributes in case of satisfaction is 23 and the corresponding attributes for loyalty is 15. There may be many redundancies in the selection of attributes. Hence factor analysis was used to reduce them and determine the key factors in each case. The KMO values of more than 0.8 in both the tables demonstrate sampling adequacy for both satisfaction and loyalty as shown in Table 1. The factor analysis using principal component matrix has led to the results provided in Table 2. It is observed that in case of satisfaction , the number of key factors have been reduced to five and three in case of loyalty. For each customer and for each factor, the responses in terms of Likert scale value have been averaged out. For each customers, five average values in case of satisfaction and three average values in case of loyalty are used as inputs in the prediction models for estimating the satisfaction and loyalty levels, respectively. Therefore to facilitate the training of satisfaction and loyalty estimation models; 80% of the total customer responses are used for training the model and remaining 20% are used for validation. In the questionnaire, provision was made to enter the satisfaction and loyalty levels of each customer on a scale of one to five. These values are normalized to lie between zero to one. The normalized satisfaction and loyalty values obtained from the customer's Table 1 KMO and sphericity values obtained for satisfaction and loyalty variables. Kaiser–Meyer–Olkin measure of sampling adequacy

0.869

(a) KMO and Bartlett's test for satisfaction Bartlett's test of sphericity Approx. chi-square df Sig.

1.752E3 253 0.000

Kaiser–Meyer–Olkin measure of sampling adequacy.

0.810

(b) KMO and Bartlett's test for loyalty Bartlett's test of sphericity

Approx. chi-square df Sig.

1.162E3 105 0.000

Table 2 Results of factor analysis using principal component analysis. Key factors Satisfaction Adaptability, consumer service, connectivity, skills of employees, advertisement Loyalty Reliability, repurchase intention, continuity in usage

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

response are used as target or desired values for training the adaptive models.

6. Development of nonlinear adaptive models for estimation of customer satisfaction and customer loyalty In this section both linear and nonlinear adaptive models have been proposed for estimation of satisfaction and their performance have been compared using real life data. 6.1. Development and validation of linear model for customer satisfaction prediction A linear adaptive model as shown in Fig. 1 is proposed for prediction of satisfaction level of customers. As depicted in this figure, for kth consumer AD(k), CS(k), CN(k), ES(K), and AV(k) represent the average normalized values of adaptability, consumer service, connectivity, employee skill and advertisement factors of ^ each consumer, respectively. The symbols S(k), QUOTE S(k) and e (k) signify actual, estimated satisfaction and error term of kth consumer, respectively. The weights, wi(k), 1rIr5 indicate the connecting weights of the model which needs to be updated so that for each consumer the square error term (e2(k)) becomes best possible minimum for each consumer. To achieve this, the normalized value of

573

each key factors of the consumer is applied sequentially and the corresponding known satisfaction value is compared to produce the error value, e(k). The weights of the model are updated according to well-known least mean square algorithm (Widrow and Stearns, 2002) given as: wi ðk þ1Þ ¼ wi ðkÞ þ 2m:xi ðkÞ:eðkÞ k ¼ 1; 2……211 i ¼ 1; 2; ……5

ð1Þ

xi(k)¼ the ith input of kth consumer which is in the present case is either of AD(k), CS(k), CN(k), ES(K) or AV(k). For obtaining the steady state weights of the model the average key factors of all the customers are applied sequentially until the square of the error attains to least possible value. The Convergence characteristics of adaptive linear model for satisfaction predicted is shown in Fig. 2. It indicates that when the training is complete the MSE attains almost a zero value. This characteristic shows that the model after completion of training phase is capable of estimating the satisfaction of customers. However the rate of convergence of this model is slow. After the learning is over the average key factors of the remaining customers which were not used for training are applied in each case and the estimated satisfaction of the model is obtained for validation purpose. In Fig. 4(a) the plots of actual and predicted satisfaction values of each customer are shown. The comparative plot indicates that the linear model

w1(k) AD(k) CS(k) CN(k)

w2(k) w3(k)

-

∑

+ ∑

(k)

w4(k)

S(k)

ES(k) AV(k)

w5(k) e(k)

LMS Algorithm Fig. 1. Linear adaptive model for satisfaction prediction of customers.

Fig. 2. Comparison of convergence characteristics of prediction model for consumer satisfaction.

574

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

agrees quite well with the actual values of the customer. Using the weights attained after training the linear relation between the predicted satisfaction of any kth consumer and the associated five key factors is given as follows: ^ SðkÞ ¼ w1 ADðkÞ þ w2 CSðkÞ þ w3 CNðkÞ þ w4 ESðkÞ þ w5 AVðkÞ

ð2Þ

For consumer k, the key factors and other parameters of the model are AD(k)¼Adaptability, CS(k)¼Consumer Service, CN(k)¼Connectivity, ES(k)¼Employee Skills, AV(k)¼ Advertisement, e(k)¼Error Value, ^ S(k)¼Customer Satisfaction, S(k) ¼predicted customer Satisfaction. The five weight values w1 to w5 in Eq. (2) are obtained from the simulation study and are listed in Table 3. To further improve the performance of satisfaction prediction, we propose a non-linear satisfaction model which has been developed in the next subsection. Table 3 Weight values of linear model for satisfaction prediction. Weights

Values

w1 w2 w3 w4 w5

0.240 0.186 0.163 0.240 0.244

AD(k)

AD(k)

6.2. Development and validation of non-linear model for customer satisfaction prediction A non-linear adaptive model for prediction of customer satisfaction using trigonometric expansion scheme is shown in Fig. 3. In this case each of the input given in the linear model is expanded to seven terms by using trigonometric expansion. For example, the input AD(k) which represents the adaptability of kth consumer is expanded in a nonlinear manner to seven terms as[ AD(k),sin{π QUOTE AD(k)}, cos{ QUOTE πAD(k)},sin{2 QUOTE πAD(k)},cos{2 QUOTE πAD(k)},sin{3 QUOTE π AD(k)},cos{3 QUOTE π AD(k)}]. Seven expansions have been proposed because such a choice has been observed to provide best possible prediction performance during simulation. The total number of expanded values of the non-linear model becomes 7  5. Hence 35 number of random weights {w1(k),w2(k)……w35(k)} are initially chosen which are then updated using the same LMS learning algorithm as given in Eq.(1). In this case also the actual and the estimated satisfactions of each consumer are compared and the error term e(k) is used to update the weight values. The convergence characteristics obtained from this non-linear model is also plotted in Fig. 2 in dotted line for comparison purpose. It is observed that the residual mean square error in case of nonlinear model is lower than that of the linear model which indicates that the prediction performance of the nonlinear model would be better. Further the nonlinear model offers faster convergence compared to that provided by the linear model. After the training of model is complete the training

w1(k)

sin πAD(k)

w2(k)

cos πAD(k)

w3(k)

Sin2 πAD(k) cos2 πAD(k) Sin3 πAD(k) cos3 πAD(k)

w4(k) w5(k) w6(k) w7(k)

CS(k)

CN(k)

Σ (k)

Σ

+ S(k)

ES(k) e(k) AV(k) sin πAV(k) cos πAV(k)

AV(k)

w29(k) W30(k) w31(k)

Sin2 πAV(k)

w32(k)

cos2 πAD(k)

W33(k)

Sin3 πAD(k)

W34(k)

cos3 πAD(k)

w35(k)

w36(k)

+1

Bias input

LMS Learning Algorithm Fig. 3. Non-linear adaptive model for prediction of consumer satisfaction.

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

575

Fig. 4. Comparison of actual and predicted satisfaction of consumers during testing phase obtained by nonlinear model. (a) Comparison of actual satisfaction and estimated satisfaction obtained using linear model. (b) Comparison of actual satisfaction with estimated satisfaction obtained by non-linear model. Table 4 Weight values of the nonlinear model for satisfaction prediction. Weights

Values

Weights

Values

Weights

Values

Weights

Values

Weights

Values

w1 w2 w3 w4 w5 w6 w7

0.071 0.070  0.043  0.007 0.008  0.019  0.017

w8 w9 w10 w11 w12 w13 w14

0.073 0.056  0.042  0.010  0.006  0.042  0.016

w15 w16 w17 w18 w19 w20 w21

0.083 0.052  0.059  0.035 0.028  0.036  0.015

w22 w23 w24 w25 w26 w27 w28

0.090 0.041  0.082  0.040 0.037  0.018  0.013

w29 w30 w31 w32 w33 w34 w35 w36

0.092 0.038  0.086  0.007 0.001  0.040 0.106 0.105

scheme is withdrawn and the performance of the model is tested using the key factors of remaining customers. The actual and predicted responses obtained are shown in Fig. 4(b). The Average Percentage of Errors (APEs) obtained during testing of the linear and non-linear models are found to be 4.16 and 5.20, respectively. Comparison of the results of Fig. 4 and also of the APE indicate that

the satisfaction level predicted using the same inputs, the nonlinear model shows more accurate prediction performance compared to that of the linear one. This conclusion is also evident from the APE. The magnitude of each of 35 weights and the bias weight obtained after training is given in Table 4. Using these weights, the non-linear relation between five input factors and the predicted

576

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

satisfaction level is represented as follows: 3

sðkÞ ¼ w1 ADðkÞ ∑ ½w2m sin mπ:ADðkÞ þ w2m þ 1 cos mπ:ADðkÞ m¼1

6

þ w8 ðkÞCSðkÞ ∑ ½w2m þ 1 sin ðm  3Þπ:CSðkÞ m¼4

þ w2m þ 2 cos ðm  3Þπ:CSðkÞ 10

þ w15 ðkÞCNðkÞ ∑ ½w2m sin ðm  7Þπ:CNðkÞ m¼8

þ w2m þ 1 cos ðm  7Þπ:CNðkÞ 13

þ w22 ðkÞESðkÞ ∑ ½w2m þ 1 sin ðm  10Þπ:ESðkÞ m ¼ 11

þ w2m þ 2 cos ðm  10Þπ:ESðkÞ 17

þ w29 ADðkÞ ∑ ½w2 sin ðm  14Þπ:AVðkÞ m ¼ 15

þ w2m þ 1 cos ðm  14Þπ:AVðkÞ

ð3Þ

The weight values of the nonlinear model associated with (3) and Fig. 3 are listed in Table 4. The improved prediction of satisfaction obtained from the nonlinear model is non-linearly related to the satisfaction values. Thus Hypothesis-1 on the nonlinear relation between satisfaction and its associated five key factors is validated. 6.3. Development and validation of nonlinear model for customer loyalty prediction It is observed that the satisfaction prediction performance of non-linear model is better than that of the liner model. Accordingly in case prediction of loyalty level of customers, also a trigonometric expansion based non-linear adaptive model was developed. The block diagram of such a model for customer loyalty

RE(k) RI(k)

Adaptive Nonlinear Model

k)

-

+ ∑

L(k)

CU(k) e(k)

Learning Algorithm Fig. 5. Block diagram for prediction of consumer loyalty.

Five factors as inputs

Satisfaction prediction model

(k)

prediction is shown in Fig. 5. As stated earlier, the number of key factors in this case has been reduced to three which are reliability, repurchase intention and continuity in usage. For the kth consumer these values are represented as RE(k), RI(k) and CU(k), respectively. The actual and estimated customer loyalty values are represented by CL(k) and CL QUOTE (k), respectively. The term e(k) denotes the error value between the estimated and actual loyalties. As in case of satisfaction prediction model, the LMS learning algorithm is used to update the weights of the non-linear model. The term e(k) denotes the error value between the estimated and actual loyalties. The model given in Fig. 5 does not take satisfaction as one of the inputs. However as has been reported by many authors (Sivadass and Baker-Prewitt, 2000, Krishnamurthi and Raj, 1991, Clarke, 2001, Choi et al., 2008), the satisfaction value positively influences the loyalty level of a customer. Accordingly in the block diagram of Fig. 6, the predicted satisfaction value obtained from the non-linear satisfaction model of each customer is used as one additional input. All other symbols used in Fig. 6 are same as that used in Fig. 5. For kth consumer, the notation used are RE(k) ¼Reliability, RI (k)¼Repurchase Intention, CU(k) ¼Continuity in usage,L^ QUOTE (k)¼Estimated, consumer loyalty, L(k)¼ Actual consumer loyalty, e(k)¼Error between estimated and actual outputs loyalties. The expanded form of the non-linear model of Fig. 5 is shown in Fig. 7. The total non-linearly expanded terms in this case is 21. Hence there are 21 adjustable weights represented by {w1,w2(k), …. w21(k)} which are to be adjusted using the LMS learning algorithm. Similar to the previous model the training is performed by applying the average loyalty values as inputs and customer satisfaction values of each customer as the target value. The learning process is continued until the mean square error of the model is reduced to the lowest possible value. At this stage the training is discontinued and the performance of the resulting loyalty prediction model is tested using the data of the customers which are not used during training. In fact, two non-linear prediction models have been developed through simulation; one using five expanded terms and other using seven expanded terms. These are represented as NM1 and NM2. Figs. 8–10 show three sets of convergence characteristics of the models under different conditions. Fig. 8 shows the comparison of the convergence characteristics of models NM1 and NM2 without using satisfaction as input, whereas Fig. 9 shows the comparison between the same models with satisfaction as input. Fig. 10 shows the comparison of convergence performance of loyalty prediction models NM2 with and without satisfaction as input. Some interesting observations have been made. First, the NM2 model in which seven expansions

Estimated satisfaction value

(k)

Inputs for loyalty prediction

Loyalty Prediction model

(k)

∑

+ L(k) e(k)

Learning Algorithm Fig. 6. Block diagram for prediction of loyalty using predicted satisfaction as additional input to the model.

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

RE(k)

RE(k)

w2

cos πRE(k)

w3

cos2π RE(k) Sin3π RE(k) cos3πRE(k)

RI(k)

w4 w5 w6 w7 w8

sinπ RI(k)

W9

cos πRI(k)

W10

Sin2π RI(k)

RI(k)

w1

sin πRE(k)

Sin2 πRE(k)

cos2π RI(k) Sin3 πRI(k) cos3πRI(k)

577

w11

-

w12

Σ

Σ

w13

(k)

w14

+ CL(k) e(k)

CU(k) sinπ CU(k) cos πCU(k)

CU(k)

w15 w16 w17

Sin2 πCU(k)

w18

cos2 πCU(k)

w19

Sin3 πCU(k)

w20

cos3πCU(k)

w21

+1

Bias input

Learning Algorithm Fig. 7. Non-linear model for prediction of consumer loyalty.

have been used, the convergence performance becomes better when five input expansions are used. This indicates that the loyalty level is more non-linearly related to the input key factors reliability, repurchase intention and continuity in usage. The second observation is that when satisfaction is given as additional input to the non-linear model the convergence performance becomes still better. This demonstrates that the satisfaction value of a consumer positively influences the loyalty value. To show the effect of satisfaction on loyalty Fig. 10 is plotted which clearly shows that satisfaction has positive impact on the loyalty value of a consumer. Analysis of Fig. 8 clearly shows that when each input key factor is nonlinearly expanded to three terms each, the convergence characteristics remains at a lower level than that obtained in case of two term expansion scheme. This implies that higher term expansion model, which is more non-linear in nature though takes more time to train and yields more accurate estimation of loyalty value. Thus these findings clearly validate that the key factors such as reliability, repurchase intention and continuity in usage are nonlinearly related to customer loyalty. Further the loyalty value was estimated using a customer's own satisfaction value as additional input and the corresponding convergence performance is compared in Fig. 10. These plots demonstrate that the satisfaction input improves loyalty prediction value. The findings in Figs. 8–10 validate Hypothesis-3 that the satisfaction has positive influence on the loyalty.

The weight value of the non-linear loyalty prediction models with and without satisfaction inputs obtained from the simulation study with five terms expansion scheme are listed in Table 5. The resultant nonlinear equation obtained from the model is given as follows: 3

LðkÞ ¼ ∑ ½w2m sin mπ:REðkÞ þw2m þ 1 cos mπ:REðkÞ m¼1

6

þ ∑ ½w2m þ 1 sin ðm  3Þπ:RIðkÞ þ w2m þ 2 cos ðm  3Þπ:RIðkÞ m¼4 9

þ ∑ ½w2m þ 2 sin ðm  6Þπ:CUðkÞþ w2m þ 1 cos ðm  6Þπ:CUðkÞ m¼7

þ w1 REðkÞþ w8 RIðkÞ þ w15 CUðkÞ þ w22 SðkÞ

ð4Þ

In (4), the last term becomes zero, if the satisfaction input is not given to the model. Under such condition, the weight values obtained from the model are shown in Table 4.

7. Managerial implications The present study has provided few important contributions which have lot of relevance for managerial decision making. Conventionally various types of regression or structural equation models are used for the study of satisfaction and loyalty. Similar investigation in the past has employed linear techniques. But in

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

MSE

578

No. of experiments

MSE

Fig. 8. Comparison of learning characteristics of loyalty prediction models using five and seven trigonometric expansion schemes.

No. of experiments Fig. 9. Comparison of learning characteristics of loyalty prediction models estimated satisfaction value as additional input for two and three terms trigonometric expansion schemes.

practice the attributes which affect the satisfaction and loyalty are not directly related to either satisfaction or loyalty values determined by the customers. Hence earlier assessment techniques provide poor estimation performance. Most of the earlier methods are adaptive in nature. Further, it has been reported that satisfaction level of a consumer has positive effect on his/her loyalty level of the products/services. The previous literature does not indicate to what extent the satisfaction value of the consumer influences his/her loyalty to the brand service. The present study shows evidence of this influence. The first contribution of the paper is identification of

key factors such as adaptability, services offered, connectivity, skills of employees and advertisement of the service providers which mostly enhance customer satisfaction. Similarly, it is observed that the reliability, repurchase intention and continuity in usage significantly improve the loyalty levels of customers. To determine the inter relation between satisfaction and its associated five key factors another nonlinear model has been developed which takes satisfaction as input along with three key factors. The results show improvement in loyalty value prediction compared to that given by the three input based model. Using these proposed models the

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

579

Fig. 10. Comparison of convergence performance of loyalty prediction models using with and without satisfaction input.

Table 5 Weight values of the nonlinear prediction (with satisfaction input). Weights

Values

Weights

Values

Weights

Values

Weights

Values

w1 w2 w3 w4 w5 w6 w7

0.118 0.101 0.101 0.079 0.093 0.046 0.052

w8 w9 w10 w11 w12 w13 w14

0.054 0.054  0.094  0.094  0.038  0.071  0.045

w15 w16 w17 w18 w19 w20 w21

 0.046 0.006  0.017 0.043 0.010 0.027 0.007

w22 w23 w24 w25 w26 w27 w28 w29

0.002  0.066 0.012  0.053  0.041  0.008  0.030  0.021

levels of satisfaction as well as loyalty of new customer can be determined by asking some relevant questions on satisfaction and loyalty. There is no need to inquire from him/her that how much satisfied/loyal he/she is on his/her service providers. This study can be directly used to assess a customer if he is likely to churn in near future. This can be achieved if appropriate threshold levels are set on satisfaction and loyalty values. Those who have these values below the corresponding pre-specified values then those customers are likely to become churners. Using the proposed model and the information from the customers' base a service provider can find the number of possible churners during a given time. This information will help the service provider to plan its strategy to provide incentives on the key factors so that the number of churners will be reduced. Further, study can be made to analyze the effect of each of the key factors on satisfaction or loyalty using the nonlinear prediction models. In essence the models developed for the estimation of satisfaction and loyalty levels of customers will benefit the telecom service providers in understanding its business health and taking appropriate measures to boost its business.

8. Limitations and future scope The present investigation has few limitations. First the sampling issue, the data collected from the customers of various cities of Andhra Pradesh have been assumed to be applicable to all other

places of India. In fact this is not true. Even though every attempt has been made to choose the respondents from all sections of people but still we feel that improvement could have been made in this regard. Every respondent has been asked to provide his/her overall satisfaction and loyalty levels in a scale of one to five. It is a fact that accurately quantifying these values is difficult task for a consumer. Since estimation of satisfaction and loyalty is based on these data, there may be some discrepancies in estimated nonlinear models and the true value. The present investigation can be extended in many ways. The estimated results on satisfaction and loyalty can be used to predict probable churners by suitably fixing the threshold values of these two parameters. The decision makers can then analyze these results to plan their strategy to retain the customers. Nonlinear relation between the key factors of loyalty/ satisfaction can be established by similar studies which will be very useful for the service providers to identify key factors in influencing customers. Then they can be suitably grouped and appropriate measures can be taken to retain them.

9. Conclusion We have studied in this paper issues related to consumer satisfaction and loyalty particularly in the mobile telecom sector of India. Factor analysis on the attributes has been carried out and five and three key factors have been obtained in case of satisfaction and loyalty, respectively. Suitable nonlinear adaptive models have been developed for estimating these values using their corresponding average value for each key factor as inputs. Promising results have been obtained from the study. The positive effect of satisfaction on loyalty prediction has been shown in the simulation results. It is also demonstrated that the respective key factors are nonlinearly related to the satisfaction/loyalty levels of the customers. Hence, simple regression based technique would not provide accurate estimation of satisfaction and loyalty values. In short the paper has made few important contributions in estimation of loyalty and satisfaction levels and the results of this study would be highly useful for estimation of number of churners of service providers or companies for devising strategies in retaining them.

580

T. Rahul, R. Majhi / Journal of Retailing and Consumer Services 21 (2014) 570–580

References Au, N., Ngai, E.W.T., Cheng, T.C.E., 2008. Extending the understanding of end user information systems satisfaction formation: an equitable needs fulfillment model approach. MIS Q. 32 (1), 43–66. Chien, T.K., Chang, T.H., Su, C.T., 2003. Did your efforts really win customers' satisfaction? Ind. Manag. Data Syst. 103 (4), 253–262. Choi, J., Seol, H., Lee, S., Cho, H., Park, Y., 2008. Customer satisfaction factors of mobile commerce in Korea. Internet Res. 18 (3), 313–335. Clarke, K., 2001. What price on loyalty when a brand switch is just a click away? Qual. Market Res.: an Int. J. 4 (3), 160–168. Dedeoglu, A.O., 2004. The symbolic use of mobile telephone among Turkish consumers. J. Euromarketing 13 (2), 143–162. Donio, J., Massari, P., Passiante, G., 2006. Customer satisfaction and loyalty in a digital environment: an empirical test. J. Consumer Mark. 23 (7), 445–447. Eggert, A., Ulaga, W., 2002. Customer perceived value: a substitute for satisfaction in business markets. J. Business Industrial Market. 17 (2-3), 107–118. Fornell, C., 1992. A national customer satisfaction barometer: the Swedish experience. J. Market. 56 (1), 6–12. Gerpott, T.J., Rams, W., Schindler, A., 2001. Customer retention, loyalty and satisfaction in the German mobile cellular telecommunications market. Telecommun. Policy 25 (4), 249–269. Isiklar, G., Buyukozkan, G., 2007. Using a multi-criteria decision making approach to evaluate mobile phone alternatives. Comput. Stand. Interfaces 29, 265–274. Kassim, N.M., Abdullah, N.A., 2008. Customer loyalty in e-commerce settings: an empirical study. Electron. Markets 18 (3), 275–290. Kim, M.K., Park, M.C., Jeong, D.H., 2004. The effects of customer satisfaction and switching barrier on customer loyalty in Korean mobile telecommunication services. Telecommun. Policy 28 (2), 145–159. Krishnamurthi, L., Raj, S.P., 1991. An empirical analysis of the relationship between brand loyalty and consumer price elasticity. Market. Sci. 10 (2), 172–183. Lai, T.L., 2004. Service quality and perceived value's impact on satisfaction, intention and usage of short message service(SMS). Inf. Syst. Front. 6 (4), 353–368. Lam, S.Y., Shankar, V., Erramilli, M.K., Murthy, B., 2004. Customer value, satisfaction, loyalty and switching costs: an illustration from a business-to-business service context. J. Market. Sci. 32 (3), 293–311. Laurn, P., Lin, H.H., 2003. A customer loyalty model for e-service context. J. Electron. Commer. Res. 4 (4), 156–167. Liao, C., Palvia, P., Chen, J.L., 2009. Information technology adoption behavior life cycle: toward a technology continuance theory(TCT). Int. J. Inf. Manag. 29 (4), 309–320. Lim, H., Widdow, R., Park, J., 2006. M-loyalty: winning strategies for mobile carriers. J. Consum. Market. 23 (4), 208–218. Lin, H.H., Wang, Y.S., 2006. An examination of the determinants of customer loyalty in mobile commerce contexts. Inf. Manag. 43 (3), 271–282. Goode, Mark M.H., Davies, Fiona, Moutinho, Luiz, Jamal, Ahmad, 2005. Determining customer satisfaction from mobile phones: a neural network approach”. J. Market. Manag. 21 (7-8), 755–778.

Mittal, V., Ross, W., Baldasare, P., 1998. The asymmetric impact of negative and positive attribute-level performance on overall satisfaction and repurchase intentions. J. Market. 61 (1), 33–47. Narver, J.C., Slater, S.F., 1990. The effect of a market orientation on business profitability. J. Market. 54 (4), 20–35. Nasir, S., 2005. The development, change, and transformation of management information systems (MIS): a content analysis of articles published in business and marketing journals. Int. J. Inf. Manag. 25 (5) (442-257). Hansemark, Ove C., Albinsson, Marie, 2004. Customer satisfaction and retention: the experiences of individual employees. Manag. Serv. Quality 14 (1), 40–57. Oliver, R.L., 1981. Measurement and evaluation of satisfaction processes in retail settings. J. Retail. 57 (3), 25–48. Oliver, R.L., 1999. Whence consumer loyalty? J. Market. 63 (4), 33–44 (special issue). Ozcan, Y.Z., Kocak, A., 2003. Research note: a need or a status symbol? Use of cellular telephones in Turkey. Eur. J. Commun. 18 (2), 241–254. Pan, Y., Sheng, S., Xie, F., 2012. Antecedents of customer loyalty: an empirical synthesis and re-examination”. J. Retail. Consum. Serv. 19 (1), 150–158. Reichheld, F.F., Schefter, P., 2000. E-loyalty: your secret weapon on the web. Harv. Bus. Rev. 78 (4), 105–113. Semejin, J., Van Riel Allard, C.R., Van Birgelen, M.J.H., Streukens, S., 2005. E-services and offline fulfillment: how e-loyalty is created. Manag. Serv. Qual. 15 (2), 182–195. Seo, D., Ranganathan, C., Babad, Y., 2008. Two-level model of customer retention in the US mobile telecommunications service market. Telecommun. Policy 32 (3-4), 182–196. Simsek, G.G., Noyan, F., 2009. A model of brand loyalty in Turkish cellular phone market. METU Stud. Dev. 36 (1), 121–159. Sivadass, E., Baker-Prewitt, J.L., 2000. An examination of the relationship between service quality, customer satisfaction and store loyalty. Int. J. Retail Distrib. Manag. 28 (2), 73–82. Tsoukatos, E., Rand, G.K., 2006. Path analysis of perceived service quality, satisfaction and loyalty in Greek insurance. Manag.Serv. Qual. 16 (5), 501–519. Turel, O., Serenko, A., 2006. Satisfaction with mobile services in Canada: an empirical investigation. Telecommun. Policy 30 (5-6), 314–331. Walsh, G., Dinnie, K., Wiedmann, K.P., 2006. How do corporate reputation and customer satisfaction impact customer defection? a study of private energy customers in Germany. J. Serv. Market. 20 (6), 412–420. Wang, Y.S., Liao, Y.W., 2007. The conceptualization and measurement of mcommerce user satisfaction. Comput. Hum. Behavior 23 (1), 381–398. Webster, F.E., 1992. The changing role of marketing in the corporation. J. Res. Market. 56 (4), 1–17. Widrow, B., Stearns, S.D. Adaptive signal processing, Pearson education, 2002. Zeithaml, V.A., Berry, L.L., Parasuraman, A., 1996. The behavioral consequences of service quality. J. Market. Res. 60 (2), 31–46. Zhaohua Deng, Yaobin Lu, Wei, Kwok Kee, Zhang, Jinlong, 2010. Understanding customer satisfaction and loyalty: an empirical study of mobile instant messages in China. Int. J. Inf. Manag. 30, 289–300.