MODELLING IMPORTANCE PREFERENCES IN CUSTOMER SATISFACTION SURVEYS

ABSTRACT Customer satisfaction measurement, through MUSA model, provides the analysts with the highest and lowest

MODELLING IMPORTANCE PREFERENCES IN CUSTOMER

SATISFACTION SURVEYS

E Grigoroudis (1) , Y.Politis (1) , O Spyridaki (1) and Y Siskos (2) ,

(1) Technical University of Crete Decision Support Systems Laboratory University Campus, 73100 Chania, Greece Tel +30-8210-37346 / Fax +30-8210-64824 Email: vangelis@ergasya.tuc.gr (2) University of Piraeus Department of Informatics Karaoli Dimitriou 80, 18534 Piraeus, Greece Tel +30-10-4142260 / Fax +30-10-4142264

Email: ysiskos@unipi.gr

ABSTRACT

Customer satisfaction measurement, through MUSA model, provides the analysts with the highest and lowest performance indicators, pointing out the leverage opportunities and the weaknesses of the company An extension of the MUSA methodology for modelling customer importance preferences for service characteristics is presented in this paper Several approaches

in the context of multiobjective linear programming are examined, which give the ability to compare derived and modelled weights of the satisfaction dimensions and to introduce the principles of Kano’s model to MUSA methodology Finally, the results of an application of the MUSA extension to an educational organization are presented in this paper

Key words: Customer satisfaction analysis, MUSA method, Satisfaction importance modelling,

Kano’s model

1 INTRODUCTION

To reinforce customer orientation on a day-to-day basis, a growing number of companies choose customer satisfaction as their main performance indicator However, customer satisfaction must be translated into a number of measurable parameters directly linked to several aspects of a company’s products/services or else it will remain an abstract and intangible notion Measurement will provide the analysts with the highest and lowest performance indicators, pointing out the leverage opportunities and the weaknesses of the company

It often happens that derived importance by a preference disaggregation model differs from the stated importance By the term stated importance we refer to the importance that is given to each criterion by the customers It is not unreasonable to say that customers tend to rate every criterion as important, when asked freely (Naumann and Giel, 1995)

The aim of this paper is to present an extension of the MUSA methodology that helps modelling customer importance preferences for service characteristics This approach gives the ability to compare derived and modelled weights of the satisfaction dimensions and to extrapolate valuable results The results of an application of the MUSA extension to an educational organization, give an example of the differentiation between derived and stated importance This paper is organised in 4 sections Section 2 presents briefly the mathematical background of importance preferences modeling Here are presented the basic principles of Kano’s model and

MUSA method, as well as a summary description of customers’ preferences importance modelling with MUSA The main results of the application for the customers of an educational organization are presented in section 3 Section 4 summarizes some concluding remarks, along with the basic advantages of the MUSA extension

SATISFACTION

2.1 Kano’s model for customer satisfaction

In many cases customer satisfaction has been seen mostly as a one-dimensional construction – the higher the perceived product quality, the higher the customer’s satisfaction and vice versa But fulfilling the individual product/service requirements to a great extent does not necessarily imply a high level customer satisfaction It is also the type of requirement which defines the perceived product/service quality and thus customer satisfaction A characteristic example of this situation is the assessment of customer satisfaction for a pen point (Vavra, 1997) If the flow of the ink is not sufficient (or it is more than needed), customers will state a high level of dissatisfaction On the other hand, if the flow of the ink is sufficient, it is possible that the customers will not state a high level of satisfaction, considering that the particular attribute is a necessary and expected feature of the product

In his model (see figure 1), Kano (1984) distinguishes between three types of product/service requirement which influence customer satisfaction in different ways when met Based on the Kano model, it can be recognized that customer satisfaction is more than one-level issue as traditionally viewed It may not be enough to merely satisfy customers by meeting their basic and spoken requirements under current highly competitive environments One main reason is that nowadays there are so many similar products for customers to choose from in the marketplace The three types of product/service requirements in the Kano model are (Kano 1984):

Must-be requirements

The must be requirements are basic criteria of a product/service If these requirements are not fulfilled, the customer will be extremely dissatisfied On the other hand, as the customer takes these requirements for granted, their fulfillment will not increase his satisfaction Fulfilling the must-be requirements will only lead to a state of “not dissatisfied” The customer regards the must-be requirements as prerequisites, he takes them for granted and therefore does not explicitly demand them Must-be requirements are in any case a decisive competitive factor, and

if they are not fulfilled, the customer will not be interested in the product/service at all

One dimensional requirements

With regard to these requirements, customer satisfaction is proportional to the level of fulfillment – the higher the level of fulfillment, the higher the customer’s satisfaction and vice versa One-dimensional requirements are usually explicitly demanded by the customer

Attractive requirements

These requirements are the product/service criteria which have the greatest influence on how satisfied a customer will be with a given product/service Attractive requirements are neither explicitly expressed nor expected by the customer Fulfilling these requirements leads to more than proportional satisfaction If they are not met, however, there is no felling of dissatisfaction

It must be noticed that the specific classification of customer requirements to one of the above categories is dynamic and affected from the competitiveness of the market Thereby, an

Customer satisfied

Customer dissatisfied

Requirement fulfilled

Requirement

not fulfilled

Must-be requirements

- implied

- self-evident

- not expressed

- obvious

Attractive requirements

- not expressed

- customer-tailored

- cause delight

One dimensional requirements

- articulated

- specified

- measurable

- technical

Source: Berger et al., 1993

Figure 1: Kano’s model of customer satisfaction

The advantages of classifying customer requirements by means of the Kano method are very

clear (Matzler et al., 1996, Matzler and Hinterhuber, 1998):

• Product requirements are better understood: the product/service criteria which have the greatest influence on the customer’s satisfaction can be identified Classifying product/service requirements into must-be, one dimensional and attractive dimensions can

be used to focus on

• Priorities for product development It is, for example, not very useful to invest in improving must-be requirements which are already at a satisfactory level but better to improve one-dimensional or attractive requirements as they have a greater influence on perceived product/service quality and consequently on the customer’s level of satisfaction

• Kano’s method provides valuable help in trade-off situations in the product development stage If two product requirements cannot be met simultaneously due to technical or financial reasons, the criterion which has the greatest influence on customer satisfaction can

be identified

• Must-be, one-dimensional and attractive requirements differ, as a rule, in the utility expectations of different customer segments From this starting point, customer-tailored solutions for special problems can be elaborated which guarantee an optimal level of satisfaction in the different customer segments

• Discovering and fulfilling attractive requirements creates a wide range of possibilities for differentiation A product which merely satisfies the must-be and one-dimensional

requirements is perceived as average and therefore interchangeable (Hinterhuber et al.,

1994)

• Kano’s model of customer satisfaction can be optimally combined with quality function deployment A prerequisite is identifying customer needs, their hierarchy and priorities (Griffin and Hauser, 1993) Kano’s model is used to establish the importance of individual product/service features for the customer’s satisfaction and thus it creates the optimal prerequisite for process-oriented product development activities

2.2 Satisfaction and customer loyalty

There have been extensive studies about the linkage between satisfaction and customer loyalty

As many researchers suggest, customer loyalty is a combination of both behaviours and attitudes (Dick and Basu, 1994; Oliver 1996; Allen and Rao, 2000; Jacoby, 1978) This means that loyal customers are those who have a favourable attitude and repeated purchases as well Oliver (1996) defines loyalty as a strong commitment of customers that will repeat the purchase

or will continue to be customers of a product or a service in the future, no matter what the impact of various situations or the efforts of marketing that aims to the change of customers’ purchase behaviour are In most cases, customer satisfaction is a necessary but not sufficient condition for loyalty Satisfaction is directed specifically at product/service characteristics, and may be relatively more dynamic measure In contrast, customer loyalty is a broaden, more static attitude toward a company in general, and it may include both rational and emotional elements

In any case, it is generally accepted that loyalty is affected by customer satisfaction in direct or indirect way (Vavra, 1997; Oliver 1996; Allen and Rao, 2000)

There are several types of loyalty according to the market conditions or the customer attachment toward a product/service Furthermore, different levels of customer loyalty exist in relation to the degree of positive commitment (Hill, 1996) The most common acceptable measures of loyalty are customer retention (repurchase intention) and willingness to recommend the product/service to other consumers

Of much interest is the work of Oliva et al (1992, 1995) were there is an attempt to study and

analyse the correlation of customer loyalty with customer satisfaction, by using the basic principles of catastrophe theory

The MUSA model is based on the principles of multicriteria analysis, using ordinal regression techniques The main objective of the MUSA method is the aggregation of individual judgments into a collective value function via a linear programming disaggregation formulation The assumption is made that client’s global satisfaction depends on a set of criteria or variables representing service characteristic dimensions

According to the model, each customer is asked to express his/her preferences, namely his/her global satisfaction and his/her satisfaction with regard to the set of discrete criteria MUSA

assesses global and partial satisfaction functions Υ * and Χ i * respectively, given customers’

judgments Υ and Χ i The method follows the principles of ordinal regression analysis under constraints using linear programming techniques (Jacquet-Lagrèze and Siskos, 1982; Siskos and Yannacopoulos, 1985; Siskos, 1985) The ordinal regression analysis equation has the following form (Table 1 presents model variables):











=

=

∑

∑

=

=

1

1

1

n

i

i

n

i

*

i

i

*

b

X

b

Y

where the value functions Y and are normalised in the interval [0, 100], and is the

weight of the i-th criterion

i

Table 1: Variables of the MUSA method

Y : client’s global satisfaction

α : number of global satisfaction levels

y m : the m-th global satisfaction level (m=1, 2, , α)

n : number of criteria

X i : client’s satisfaction according to the i-th criterion (i=1, 2, …, n)

α i : number of satisfaction levels for the i-th criterion

x i k : the k-th satisfaction level of the i-th criterion (k=1, 2, , α i)

Y * : value function of Y

y *m : value of the y m satisfaction level

X * i : value function of X i

x i *k : value of the x i k satisfaction level

The normalisation constraints can be written as follows:









=

=

=

=

,n , , i=

x

,

x

y

,

y

i

*α

i

i

*α

*

… 2 1 for 100

0

100

0

1

1

(2)

Furthermore, because of the ordinal nature of Y and the following preference conditions

are assumed:

i

X









−

=

⇔

≤

−

=

⇔

≤

+ +

+ +

1 2

1 for

1 2

1 1 1

1 1

i k

i k i k

*

i

k

*

i

m m m

*

m

*

,α , , k x

x x

x

,α , , m y

y y

y

…

≺

…

(3) where ≺ means “less preferred or indifferent to”

The MUSA method infers an additive collective value function , and a set of partial

satisfaction functions from customers’ judgements The main objective of the method is to

achieve the maximum consistency between the value function Υ and the customers’

judgements Υ

*

Υ

*

* i

Χ

Based on the modelling presented in the previous section, and introducing a double-error

variable, the ordinal regression equation becomes as follows:

− +

=

+

−

Y ~

i

*

i

i

*

1

(4)

where Y ~ * is the estimation of the global value function Y , and and are the

overestimation and the underestimation error, respectively

Equation (4) holds for customer who has expressed a set of satisfaction judgements For this

reason a pair of error variables should be assessed for each customer separately (Figure 2)

0

100

Y

Figure 2: Error variables for the j-th customer

Removing the monotonicity constraints, the size of the previous LP can be reduced in order to

decrease the computational effort required for optimal solution search This is effectuated via

the introduction of a set of transformation variables, which represent the successive steps of the

transformation equation can be written as follows (see also Figure 3):

i

Χ









−

−

=

−

−

=

+

+

, ,n , i=

, ,α , k=

x b x

b

w

, ,α , m=

y

y

z

i k

i i k

i

i

ik

m

* m

*

m

2 1 and 2

1 for

1 2

1 for 1

1

(5)

It is very important to mention that using these variables, the linearity of the method is achieved

since equation (4) presents a non-linear model (the variables Υ and , as well as the

coefficients b should be estimated)

i

Χ

i

y 1 y 2 y m y α

0

100

Y

Y

0

100

i

i

b

w

i− 1 α

i

i

b

i

i

b

Figure 3: Transformation variables z m and w ik in global and partial value functions

According to the aforementioned definitions and assumptions, the basic estimation model can

be written in a linear program formulation as it follows:

[min]F = ∑

=

− + +

M

j 1 j j

σ σ

under the constraints

∑

∑∑ −

=

− +

=

−

=

= +

−

− 1

1

1

1

1

0

j

ji t

m

j j m n

i

t

k

ik z

∑−

=

=

1

1

100

a

m m

∑∑

=

−

=

=

n

i

a

k ik

i

w

1

1

1

100

0

≥

m

z , wik ≥ 0, ∀ m , , i k

0

≥

+

j

j

where M is the number of customers

The preference disaggregation methodology consists also of a post optimality analysis stage in order to face the problem of multiple or near optimal solutions The MUSA method applies a heuristic method for near optimal solutions search (Siskos, 1984) The final solution is obtained

by exploring the polyhedron of near optimal solutions, which is generated by the constraints of

the above linear program During the post optimality analysis stage of the MUSA method, n

linear programs (equal to the number of criteria) are formulated and solved Each linear program maximizes the weight of a criterion and has the following form:

∑−

=

=

1

[max] i

a

k

ik

w

F , for i = 1 , 2 , , n

under the constraints

+

all the constraints of LP (6)

where ε is a small percentage of F *

The average of the optimal solutions given by the n LPs (7) may be considered as the final

solution of the problem The model provides collective global and partial satisfaction functions

as well as average satisfaction indices and weights that represent the relative importance of each

criterion/subcriterion

4

Figure 4)

Figure 4: Clauses of customers’ importance preferences

4.1 Weight estimation using ordinal regression techniques

between the derived importance

(where M is the number of

following co

MODELLING PREFERENCES FOR CRITERIA IMPORTANCE

In order to model customers’ preferences, customers are asked, via a specialized questionnaire,

to place each one of the satisfaction criteria in one of the following categories: C1 = very important criterion, C2 = important criterion, C3 = less important criterion

Considering that C1, C2, C3 are ordered in a 0 to 100% scale, there are two preference thresholds

T1 and T2, which define the % rate, which distinguishes each one of the three categories (see

The main purpose of this approach is the comparative analysis

of the criteria through the MUSA method and the stated importance given by the customers In order to estimate the stated importance of the criteria, which is a qualitative variable, a linear program is formulated The program calculates the two preference thresholds T1, above which a criterion is considered very important, and T2, below which a criterion is considered less important In this way, the importance of each criterion according to customers’ preferences can

be assessed and compared with the results of the MUSA method

For each criterion i =1,2, n and each customer j = 1 , 2 , , M

customers and n is the number of criteria) we set the nstraints:

• If bˆij∈C1, that is customer j considers criterion i ‘very important’ then:

∑

=

1

1

i

t it

−

100 Τ1 + Sij 0

• If C2, that is customer j considers criterion i ‘important’ then:

a

ij

bˆ ∈

a

∑

=

1

1

i

t it

−

100 Τ1 - Sij- 0

=

1

1

i

a

t it

w -100 Τ2 + Sij 0 ≥

• If b ˆ ∈ij C3, that is customer j considers criterion i ‘less important’ then:

∑−

=

1

1

i

a

t it

where Sij+ and Sij- are the overestimation and underestimation error, respectively, for the i-th criterion of the j-th customer, C1, C2, C3 are the customers’ preference categories, T1 and T2 are the preference thresholds, αi is the number of satisfaction scale levels for i criterion, and wit is a MUSA variable

The final linear program is:

j i

ij

ij S

S

under the constraints

∑

∑

∑

∑

−

=

−

−

=

+

−

=

−

−

=

+

∈

≤

−

−

∈

≥ +

−

≤

−

−

∈

≥ +

−

1

1

3 2

2 1

1

1

1

1

1

1 1

ˆ , 0 100

ˆ 0 100

0 100

ˆ , 0 100

i

i

i

i

a

t

ij ij

it

ij a

a

t

ij it

a

t

ij ij

it

C b S

T

w

C b S T

w

S T

w

C b S

T

w

∀i=1,2, ,n and j =1,2, ,M (8)

∑∑

=

−

=

=

n

i

a

k

ik

i

w

1

1

1

100

≥

2

≥

− 2

1 T

After the solution of LP (8) a post optimality analysis follows, where n linear programs (equal

to the number of criteria), are formed and solved Those linear programs maximize the weights

bi of the criteria and have the following form:

∑−

=

=

1

[max] i

a

k

ik

w

F , for i=1,2, ,n

under the constraints

ε

+

all the constraints of LP (8)

where F *is the optimal solution of the objective function of LP (8) and ε is a small percentage of

F *

4.2 Extension of the MUSA model

The main purpose of this analysis is to examine whether additional information about the weights of the criteria can improve the results of the MUSA method The examination of possible improvement is done through the Average Stability Index (ASI) ASI is the mean value

of the normalized standard deviation of the estimated weights bi and is calculated as follows:

( )

∑ ∑ ∑

=

−













−

−

i

n j

n j

j i

j i

n

b b

n n

ASI

1 1

2

1 2

1 100

1

the j-th post-optimality analysis LP (Grigoroudis and Siskos, 2002)

j i

At first, it is examined the following Mulltiobjective Linear Programming (MOLP) problem:

[min]F1=∑

=

− + +

M

j 1 j j

σ σ

j i

ij

ij S S

under the constraints

all the constraints of LP (6)

In a Mulltiobjective problem it is pointless to try to find out a solution which will optimize all the criteria of the objective functions simultaneously, considering that, in most of the cases, the criteria are competitive, that is the optimal value of one criterion is not optimal for the other A basic tool for the representation of the competitiveness among multiple objective functions is the pay-off matrix This table represents the values that the multiple objective functions take when optimizing the value of one of these objective function

This multiobjective problem could be solved according to any MOLP method Here the following heuristic method is chosen:

Stage A: Solution of the following linear program

=

− + +

M

j

j j

1

σ σ

under the constraints

all the constraints of LP (6)