Biologically Inspired Approaches to Strategic Service Design

Abstract

This article introduces nature-inspired modeling to strategic service management. It determines optimal service diversification through an evolutionary mechanism of natural selection and population genetics as well as a model of cooperative behavior and collective intelligence in swarms. Specifically, we design and implement Genetic and Particle Swarm Optimization algorithms to stated-preference data derived from a conjoint experiment measuring consumer preferences for service attributes in a retail setting. The proposed procedure provides key insights to strategic service management such as optimal service design, optimal mix of service offerings in terms of consumer demand, and local adaptation of service portfolios. It demonstrates how diversification meets heterogeneous customer preferences and how localized solutions address cross-country differences. The findings suggest that variation in service portfolios elevates customer utility, in the sense that diversified offerings better match heterogeneous customer needs. In an intuitive fashion, consumer diversity is such that a uniform service portfolio is inferior to differentiated offerings, especially with regard to salient service attributes. The results also illustrate that localized diversification strategies are necessary for multistore, multimarket operations. Our method has valuable implications for managers aiming to improve how they design their services. A new tool is introduced which handles tangible and intangible service elements and allows service design optimization by predicting which elements create the most compelling service contexts from a customer perspective. The tool also facilitates localized diversification decisions by adapting critical service attributes to local markets. Bio-inspired models shed new light on marketing phenomena and reveal opportunities for empirical research.

Keywords

evolutionary algorithm swarm intelligence optimal service design service diversification

Introduction

Diversification is an important marketing strategy that provides an operational way to address heterogeneous markets. Firms strive to make market-driven, empirically determined, diversification decisions and develop portfolios of optimally diversified offerings. Diversification of services is increasingly common, especially in contexts where success depends heavily on customer experience and satisfaction. For example, retailers develop and operate multiple store types, each offering a unique service experience, to better satisfy increasingly heterogeneous shopping patterns and habits of their customers. Preferences for service attributes may vary across customers and within individuals across time. Thus, a key challenge for service strategists is to identify the best combination of service offerings that would maximize customer utility and satisfaction.

Research in the area of service design has been mainly concerned with managerial and administrative aspects (Berry and Lampo 2000; Froehle et al. 2000; Martin and Horne 1993; Patrício, Fisk, and Falcão e Cunha 2008; Tax and Stuart 1997). The analysis of service design from a customer-preference perspective has received far less attention in the literature, despite its importance for the development of successful, customer-driven offerings (Heineke and Davis 2007; Patrício, Fisk, and Falcão e Cunha 2008; Pullman and Moore 1999; Roth and Menor 2003). Even a cursory look through the service literature shows that there is a lack of research in the area of optimal service design and optimal service diversification. On the other hand, optimal product design is an established area in the empirical literature and researchers have applied various methods to determine optimal product lines for manufacturing firms.

The present article determines optimal service design and optimal service diversification through an evolutionary mechanism of natural selection and population genetics as well as a model of cooperative behavior and collective intelligence in swarms. Both bio-inspired models are introduced for the first time in the service literature, while the Swarm Intelligence mechanism is introduced for the first time in the empirical marketing literature. Furthermore, in contrast to existing studies on optimal product design, which maximize a firm’s market share or profit, the models used in the present study maximize customer utility from services.

In particular, we develop and implement a Genetic algorithm (GA) as well as a Particle Swarm Optimization (PSO) algorithm to obtain solutions for the optimal combination of service offerings and the spatial distribution of service offerings across different markets. To illustrate our approach, we use a hybrid conjoint experiment to collect preference data from two different markets. Then, we use our bio-inspired algorithms to determine optimal service offerings for each market. The adaptation to local market conditions is a rather natural extension for this biological framework, in which the interplay between service mix and market environment is most explicit. To the best of our knowledge, this is the first empirical application of biologically inspired models in services. We demonstrate the application of our bio-inspired methodology in a retail service setting where, more so than any other service sector, the literature has completely failed to address issues of optimal service diversification. Therefore, the novelty of this study is two-fold and resides in both the managerial problem and the research methodology.

The rest of the article is organized as follows. The next section summarizes the theoretical background and the aim of the present work. The third section is concerned with the service diversification problem and outlines our combined experimental and optimization method. Subsequently, we present the hybrid conjoint experiment that measures consumer preferences for service attributes. The fifth section introduces evolutionary systems and presents our GA. This is followed by an overview of Swarm intelligence and the development of our PSO algorithm. The seventh section presents and discusses empirical results. The last two sections summarize the article and discuss important implications for managers and researchers.

Theoretical Background

Literature on Service Design

Existing studies on service design have addressed mainly administrative issues such as service design processes (e.g., Fließ and Kleinaltenkamp 2004; Hill et al. 2002; Johnson et al. 2000), redesign of service offerings (e.g., Berry and Lampo 2000), experience-centric services (Zomerdijk and Voss 2010), service encounters (e.g., Cook et al. 2002; Gupta and Vajic 2000; Gwinner et al. 2005; Tansik and Smith 2000), success and failure in service design (e.g., Cooper and de Brentani 1991; Cooper et al. 1994; de Brentani 1995; de Brentani and Cooper 1992; Martin and Horne 1993), and physical or virtual “servicescapes” (Vilnai-Yavetz and Rafaeli 2006).

The issue of service design from a customer-preference perspective has received little if any attention, despite its importance for the development of successful, customer-driven offerings (Heineke and Davis 2007; Patrício, Fisk, and Falcão e Cunha 2008; Pullman and Moore 1999; Roth and Menor 2003). In particular, the literature reflects a lack of research in the area of optimal service design and optimally diversified service portfolios. Brown (2012) is the only study that specifically addresses optimal diversification of service portfolios. Brown introduces modern financial portfolio analysis to managing service portfolios and uses the criteria of risk and return as well as efficiency frontiers to determine optimal service portfolios for three prominent hotel firms. To our knowledge, this is the only such study in the empirical literature.

The service design problem possesses some unique aspects that differentiate it from the product design problem. For example, the design of services requires the systematic management of service experience aspects related to sensory and human elements, such as the physical environment and the fellow customers, and generally a set of both tangible and intangible elements of the service delivery system (Pullman and Gross 2004; Zomerdijk and Voss 2010).

Some service elements that potentially affect customer experiences have been individually examined in the literature, such as the physical or virtual “servicescape” (Vilnai-Yavetz and Rafaeli 2006) and the human elements in service encounters (Cook et al. 2002; Gwinner et al. 2005). However, more work is needed to advance customer-driven service design in an integrated fashion (Heineke and Davis 2007; Patrício, Fisk, and Falcão e Cunha 2008; Roth and Menor 2003; Zomerdijk and Voss 2010). To design a new or improved service, managers have to determine which service elements are important to customers, and whether or not the service provider is capable of delivering those elements (Berry and Lampo 2000; Tax and Stuart 1997). Therefore, it would be desirable to develop a method that (a) designs optimal new services by integrating both tangible and intangible service elements (e.g., intangible service elements related to human aspects of the service experience, such as the type of clientele), and (b) predicts which service elements create the most compelling service contexts from a customer perspective.

Literature on Service Diversification

New service design and service diversification are of course interrelated. Service diversification is a means to meet heterogeneous needs of a diverse clientele, an issue which is crucial for an organization’s survival (Knudsen, Roman, and Ducharme 2005).

Generally considered, diversification strategies can be either related or unrelated (e.g., Keep, Hollander, and Calantone 1996). In related diversification, a firm may choose to add new and modified versions of a current offering to expand its potential market by focusing on segments with different preferences for specific attributes. For example, a supermarket chain that intends to expand its customer base may choose to diversify its services by developing new store formats that serve different shopping habits and differ from each other regarding the relative importance of customer satisfaction drivers (Malthouse et al. 2004). In practice, supermarket chains manage to expand their customer base by delivering diversified services through their store formats to meet heterogeneous shopping habits and preferences (González-Benito 2005). For example, British retailer Tesco diversifies its services across four distinct store types of its store portfolio to provide different shopping experiences: Tesco Extra (hypermarket offering maximum food and nonfood range), Tesco Superstore (large store offering a full range of many nonfood products), Tesco Metro (high street store located in large city centers), and Tesco Express (petrol station forecourt stores selling a range of convenience products).

The present article introduces a new method for diversifying services and combining multiple service designs. The method also incorporates a spatial perspective and addresses localized diversification strategies. The issue of local adaptation has attracted considerable attention in the area of international marketing but far less attention in retail services (e.g., McGoldrick 1998; Severin, Louviere, and Finn 2001). On the other hand, retail services become more internationalized. Retailers transcend geographical borders, and cross-country differences may limit the transferability of retail strategies. Research suggests that differences among international markets in terms of ethnic, economic, demographic, cultural, social, and linguistic bases may result in different expectations from service providers and different perceptions of the actual service delivered (e.g., Lee and Ulgado 1997; Mattila 1999). Cross-country differences may require service adaptation and localized diversification strategies.

In this article, we develop and implement optimal service design methods by combining an experimental, conjoint-based approach with bio-inspired artificial intelligence. The proposed methodology is capable of addressing a wide range of service environments and illustrated in a retail service context.

The aim of this article is two-fold. First, the article introduces bio-inspired methods to service research and Swarm Intelligence to the broader area of marketing by demonstrating their usefulness for strategic decision making. Second, it considers the problem of optimal service design and the interrelated problem of service diversification, for which very little is currently known from empirical studies. It should be emphasized that there has been no previous attempt to address this specific issue in the existing literature, besides the seminal study by Brown (2010) already mentioned.

Product Design

The literature on optimally diversified offerings in product line design has been surveyed by Lancaster (1990), Krishnan and Ulrich (2001), and Ramdas (2003), and reflects two distinct research streams, namely, the engineering and the marketing stream. Some researchers have also tried to integrate both streams into the design of either a single product (e.g., Besharati et al. 2006; Luo et al. 2005; Luo, Kannan, and Ratchford 2008; Michalek, Feinberg, and Papalambros 2005) or a line of products (e.g., D’Souza and Simpson 2003; Farrell and Simpson 2009; Heese and Swaminathan 2006; Jiao and Zhang 2005; Kumar, Chen, and Simpson 2009; Li and Azarm 2002; Luo 2011; Michalek et al. 2006).

In the engineering stream, researchers focus on platform management and strive for balance between the commonality of the product platform and the individual product’s engineering performance (e.g., Rai and Allada 2006). The engineering perspective manifests the manufacturing orientation of existing optimal product line research. In the marketing stream, researchers usually employ conjoint or simulated data and search for an optimal or near-optimal product line by selecting discrete attribute levels (e.g., Balakrishnan, Gupta, and Jacob 2004; Belloni et al. 2008; Kannan, Pope, and Jain 2009; Nair, Thakur, and Wen 1995; Steiner and Hruschka 2003). The application of conjoint analysis provides an index for each attribute level (i.e., partworth), which represents a relative measure of utility. Based on these estimations, the application of an optimization algorithm finds the product line that maximizes market share.

The optimization problem can become too complex and thus alternative heuristic procedures have been proposed, including dynamic programming (Kohli and Sukumar 1990), beam search (Nair, Thakur, and Wen 1995), and Lagrangian relaxation with branch and bound (Belloni et al. 2008). Recently, nature-inspired approaches have been introduced to the problem, including GAs (Alexouda and Paparrizos 2001; Steiner and Hruschka 2003; Balakrishnan, Gupta, and Jacob 2004) and Ant Colony Algorithms (Albritton and McMullen 2007).

Empirical Determination of Optimal Service Diversification

The steps toward the determination of optimal service diversity are as follows. First, some experimental or observational method determines consumer preferences for service attributes. This study uses conjoint analysis to estimate partworth values of store characteristics. Optimal decisions could be reached through an explicit enumeration of idiosyncratic utilities if conjoint profiles consisted of a limited number of attributes and attribute levels. Thus, the optimal solution of small experimental designs can be found through a full search of the entire solution space by applying complete enumeration processes. In more complex problems involving a greater number of attributes and attribute levels, the complete enumeration procedure can become formidable. The classic optimization problem becomes quite impractical and, inevitably, one turns to alternative heuristic approaches.

In particular, our conjoint problem involves 12 store attributes with 29 levels grouped into four facets (see Appendix). The derived measures of individual preferences (i.e., partworth values) evaluate alternative profiles of the hypothetical store. For a single-store portfolio, the number of possible solutions (combinations of attribute levels) is 2⁷ × 3⁵= 31,104. A complete enumeration procedure can identify the optimal solution that corresponds to the profile with the highest choice share. However, for a portfolio of two different stores, the number of possible solutions is 9 × 10⁸, whereas for three different stores the number is greater than 10¹². The extremely large number of possible solutions of a multiple store selection problem makes the classic optimization problem infeasible. Actually, Kohli and Krishnamurti (1989) proved that similar share of choices problems (market share maximization) belong to the class of Nondeterministic Polynomial Time (NP)-hard problems, which means that a full search of the entire solution space is practically infeasible in tractable time. Belloni et al. (2008) use discrete optimization methods that combine Lagrangian relaxation with Branch and Bound and find an optimal solution after 1 week of computation time. This article shows how GAs and PSO represent nature-inspired, efficient search methods for such difficult optimization problems.

Determining Preferences for Service Attributes

Data and Variables

The literature of service design emphasizes the importance of identifying which service attributes create the most compelling offerings and how such salient characteristics can be used to enhance customer experience, satisfaction, and loyalty (e.g., Gustafsson and Johnson 2004; Zomerdijk and Voss 2010). Consumer preferences for service characteristics have to be empirically determined by some experimental or observational method. In this article, we design a hybrid conjoint experiment that can handle a large number of attribute levels. Conjoint analysis has been used in service settings including academic services (e.g., Howard and Sobol 2004) and digital television services (e.g., Song, Jang, and Sohn 2009).

In this study, we consider 12 supermarket service attributes with 29 levels grouped into four categories; called facets (see Appendix). The first facet includes all elements related to merchandise’s tangible and intangible aspects (e.g., quality of fresh foods and vegetables, variety of product categories, presentation of the products). The second facet includes experience-sensory elements of the physical environment (e.g., layout and size). The third facet includes intangible service elements related to human aspects of the service experience (e.g., size and type of clientele; Zomerdijk and Voss 2010). The fourth facet includes convenience characteristics of the service facility (e.g., distance, parking). Each attribute ranges from two to three levels, resulting in a total of 29 levels. The facet and attribute selection is based in the literature of service design as well as a series of in-depth interviews with consumers. Exploratory in-depth interviews are commonly used at the initial stages of conjoint experiments for the specification of variables (Hair et al. 2006).

To incorporate a spatial perspective to our method and address localized diversification strategies, the conjoint experiment was carried out in two European Union (EU) member states, namely, United Kingdom (Market A) and Greece (Market B) and administered during personal interviews at the homes of 116 participants after arranging an appointment. In Table 1, some basic demographic characteristics of our sample are presented.

Table 1.

Demographic Characteristics.

Demographic Variable	Total Sample (in %)	Market A (in %)	Market B (in %)
Gender
Male	38	20	53
Female	62	80	47
Age
Up to 34 years old	56	33	77
35–49	24	33	16
50 years old or more	20	33	7
Annual household income
Up to £10,000	20	20	21
£10,000–£25,000	31	24	37
£25,000–£45,000	38	48	29
£45,000–£60,000	6	6	5
£60,000 or more	6	2	8

Questionnaire Design

We adopted and adapted the Individualized Hybrid Model, proposed by Green and Krieger (1996). The questionnaire includes two main tasks, a compositional (self-explicated) and a decompositional task (full-profile evaluations). During the compositional task, respondents (1) rate each attribute level (from 0 = not desirable at all to 10 = extremely desirable), (2) allocate 100 points across the attributes of each facet so as to reflect their relative importance, and (3) allocate 100 points across all four facets so as to reflect their relative importance. In the compositional task, the respondent had to state their preferences for various supermarket features, based on a single comparison rather than on multiple ones. During the decompositional task, respondents evaluate a set of six complete designs and rate the likelihood of visiting each hypothetical supermarket for their main shopping (from 0% = definitely would not buy from this supermarket to 100% = definitely would buy from this supermarket). Prior to the actual survey, questionnaires were checked in terms of their clarity and effectiveness to extract the appropriate information in a pilot sample of 20 respondents.

The Experimental Design

The set of 6 profiles that was randomly assigned to be rated by each participant belongs to a fractional factorial design of 36 profiles. We used a subset of all possible multifactor stimuli combinations by generating a 36-profile, asymmetric fractional factorial design (Hair et al. 2006). A blocking design was then used to group the 36 profiles of the fractional factorial design into six sets (or blocks) of 6 profiles each. Each respondent was randomly assigned to rate the profiles of a block. Block 1 included descriptions of large supermarkets with wide aisles that operate 24 hours a day; Block 2 included descriptions of supermarkets with well-presented merchandise but of relatively low quality; Block 3 included descriptions of supermarkets with wide assortment that provide food and nonfood products; Block 4 included descriptions of supermarkets that usually are not crowded with people and have good parking facilities; Block 5 included descriptions of supermarkets that are always crowded with people and have merchandise that is not very well presented; and Block 6 included descriptions of supermarkets that provide only food products and have poor parking facilities.

Service Diversification Through Evolution

GAs belong to the class of Evolutionary algorithms and mimic the approach of nature in the evolution of species (e.g., Hruschka 2008; Safarzynska and van den Berg 2010). Possible solutions to our problem (i.e., store-format portfolios) are represented by chromosomes, whose genetic material (genes) consists of service attributes. Each profile consists of 12 service attributes, which take discrete values 0/1 for the two-level attributes and 0/1/2 for the three-level attributes. For a uniform service portfolio, each chromosome represents a single service design and has a length of 12 genes. For a diversified portfolio, each chromosome has a length of m × 12 genes, where m is the number of service designs.

The algorithm begins with the creation of an initial population P(0) of n chromosomes at random. Then, the algorithm evaluates each chromosome based on its utility value and assigns the derived value as the chromosome’s fitness. Next, the reproduction process begins with the creation of the mating pool. The algorithm employs the binary tournament selection, a semi-random process where a pair of chromosomes is selected with replacement from population P(t), and the one with the higher fitness is chosen to participate in the mating pool. The process is repeated n times to choose n candidate parent chromosomes. After creating the mating pool, the algorithm randomly selects without replacement n/2 pairs of chromosomes to be candidate parents. A pair of new chromosomes (offspring) is created by mating the pair of parents through the application of a crossover operator with a probability p_c . In crossover, the parent chromosomes combine their genetic material in a certain way. The mutation process follows, which creates new genetic material and enables the algorithm to search new areas of the solution space that crossover would never explore. The algorithm employs a mutation operator where each gene of the newly created chromosomes is selected with probability p_m and its value is altered.

The process is then repeated from the evaluation step, until a convergence criterion is met. We chose the moving average criterion. The algorithm terminates when the average fitness of the best three chromosomes remains constant for five consecutive generations, indicating that further optimization is unlikely. In implementing our GA, it is important to select the most appropriate values for the population size (n), the crossover probability (p_c ), and the mutation rate (p_m ). Table 2 shows a 4 × 4 × 4 full factorial design with four values for each of the three parameters.

Table 2.

Genetic Algorithm: Alternative Values for Population Size, Crossover Probability, and Mutation Rate.

Parameter	Values
Population size	100	150	200	250
Crossover probability	0.6	0.7	0.8	0.9
Mutation rate	0.01	0.04	0.07	0.10

Note. Five replications are performed for each of the 64 combinations of the three parameters (320 runs in total). For population sizes more than 150, there is no gain in performance, while the best average performance is achieved for p_c = .9 and p_m = .04.

A sensitivity analysis tests several values for one parameter while holding the other two parameters constant. Ten replications are performed in each case. First, the sensitivity analysis evaluates eight population sizes in the range (120, 190) at increments of 10 chromosomes with p_c = .9 and p_m = .04. Populations greater than 160 do not improve the algorithm’s performance. Second, the sensitivity analysis evaluates seven crossover values (0.84, 0.86, 0.88, 0.9, 0.92, 0.94, 0.96) with n = 160 and p_m = .04. The algorithm has the best performance for p_c = .96. Finally, the sensitivity analysis tests five different mutation rates (0.03, 0.035, 0.04, 0.045, 0.05) with n = 160 and p_c = .96. The algorithm has the best performance for p_m = .035.

Service Diversification Through Swarm Intelligence

Swarm Intelligence

In nature, we observe biological collective phenomena whose structures and processes exceed in complexity the perceptual, physical, and cognitive abilities of the participating organisms. Such phenomena emerge from the collective work of individual organisms (e.g., bees, birds, and fishes) that interact locally and execute simple behaviors based on limited information. It should be emphasized that individual agents do not follow any centralized plan to coordinate their actions. The research area of swarm intelligence is concerned with groups of such simple agents that can collectively solve problems that are too complex for a single agent and display robustness and adaptability to environmental variation (Floreano and Mattiussi 2008).

Particle Swarm Optimization

In particular, PSO is a biologically inspired optimization framework developed by Kennedy and Eberhart (1995). PSO is, in essence, a machine-learning technique that creates computational intelligence by exploiting simple analogues of collective behavior found in nature, such as bird flocking and fish schooling. PSO mimics nature in exploring collective problem solving without centralized control or the provision of a global plan. More specifically, PSO is an iterative population-based algorithm that works with a group (swarm) of particles (e.g., birds), which collectively move on the d-dimensional real space in search of the global optimal solution of the problem. The location of each particle i corresponds to a candidate solution of the problem represented by a vector ${\vec{x}}_{i} \in ℜ^{d}$ :

{\vec{x}}_{i} = (x_{i 1}, x_{i 2}, . . ., x_{i d}), i = 1, 2, . . ., n, x \in ℜ,

where n is the number of particles in the swarm (population size). The particles “fly” within the problem space by changing their locations in each iteration of the algorithm. The PSO algorithm begins with a random placement of the particles on the problem space (i.e., initialization of the particles’ locations), and an evaluation of their fitness (performance) on the objective function. Thus, each particle obtains a location and a performance score. In the PSO algorithm, each particle communicates its current performance to other particles, knows the best performing particle of the swarm, and remembers its own past best performance. This enables the particles to follow the “leader” while maintaining their position in the swarm. An iterative procedure follows where each particle moves on the search space by following both its current personal best position (solution) ${\vec{p}}_{i}$ , as well as the position of the best particle of the swarm ${\vec{p}}_{g}$ , with some random permutations. That is, in each algorithm’s iteration, particle i alters its location (i.e., candidate problem solution) by taking into account the values of ${\vec{p}}_{i}$ and ${\vec{p}}_{g}$ .

The first particle swarms evolved out of bird-flocking simulations, and PSO topologies were based on proximity in the search space. However, besides being computationally intensive, this kind of communication structure had undesirable convergence properties and was soon abandoned. In this study, we use the gbest topology (for “global best”) where the best particle in the entire population influences the target particle. While this may be conceptualized as a fully connected graph, in practice it only means that the algorithm needs to keep track of the best result that has been found, and the particle that found it (Poli, Kennedy, and Blackwell 2007).

The rate of the particle’s position change is represented by its velocity ${\vec{v}}_{i}$ . The velocity and the position of particle’s i dimension d are adjusted in each algorithm’s iteration with the use of the following functions:

v_{i d} (t + 1) = v_{i d} (t) + c_{1} \times r n d_{1} \times (p b e s t_{i d} - x_{i d} (t)) + c_{2} \times r n d_{2} \times (p b e s t_{g d} - x_{i d} (t)) .

x_{i d} (t + 1) = x_{i d} (t) + v_{i d} (t + 1),

where t is the iteration number, rnd₁ and rnd₂ are two random functions in the range (0, 1), and pbest_id and pbest_gd are dimension’s d values of the ${\vec{p}}_{i}$ and ${\vec{p}}_{g}$ , respectively. The two positive constants c₁ and c₂ control the weights of the “cognition” part, which simulates the private thinking of the particle itself, and the “social” part that simulates the collaboration among particles (Kennedy 1997). They are both usually set to 2, which on average makes the two weights to be 1. When all particles have completed their move, their fitness is evaluated and the algorithm proceeds to the next iteration. Thus, the swarm, like a flock of birds or a school of fish collectively foraging for food, is likely to move close to an optimum of the fitness function (Poli, Kennedy, and Blackwell 2007). The optimization process terminates after a predetermined number of iterations, or when a convergence criterion has been met. Figure 1 illustrates the process.

Figure 1.

Particle swarm optimization algorithm flowchart.

Solving the Optimal Service Design Problem Through PSO

We now turn to deal explicitly with our service design problem. Our PSO algorithm is developed in the MATLAB environment. While the PSO was originally developed for continuous optimization functions, it has been successfully used in many classical discrete problems such as the traveling salesman problem (Clerc 2004; Marinakis and Marinaki 2010; Shi et al. 2007), the vehicle routing problem (Ai and Kachitvichyanukul 2009; Zhu et al. 2006, the minimum spanning tree problem (Goldbarg et al. 2006; Guo et al. 2007), and the integer programming problem (Laskari, Parsopoulos, and Vrahatis 2002). We represent a problem’s solution (i.e., a store-type portfolio) with the use of a binary vector, each dimension of which represents an attribute level. A value of 1 denotes that the specific level is assigned to the corresponding attribute. Since exactly one level must be assigned to each attribute, the solution’s vector is divided into parts, each describing a single attribute. Within each part, exactly one dimension must take a value of 1 and all the others must take a value of 0. For instance, in a portfolio that contains a single store with four service attributes each taking three levels, the vector x = (0 0 1| 0 0 1| 0 1 0| 1 0 0) represents a profile where the third level is selected in attributes one and two, the second level is selected in attribute three, and the first level is selected in attribute four. Under this representation schema, the particles move on a search space of d = m × l dimensions, where m is the number of service designs in the portfolio and l is the total number of attribute levels per design.

In our data set, each profile consists of 12 attributes. The total number of attribute levels is 29. If we are looking for the best single-type portfolio, the position of each particle corresponds to a single profile and is represented by a vector ${\vec{x}}_{i}$ = (x_i ₁, x_i ₂, . . . , x_i ₂₉). If we are looking for a multiple-store solution, we concatenate the different profiles into a single particle with a size of d = m × 29.

When PSO is applied to discrete domains, which is the case in our problem, a mapping rule must be employed for converting the particle’s position (which is a vector of real numbers) to a vector that represents a solution of the discrete problem (binary in our case). For this, we adopt the Smallest Position Value mapping rule, a heuristic used in the application of PSO to the Single Machine Total Weighted Tardiness problem (Tasgetiren et al. 2004). To convert a real vector that represents a particle’s position to a binary vector that represents a problem’s solution, we assign a value of 1 to the dimension with the smallest real value within each particle’s part and assign a value of 0 to the remaining dimensions. For example the particle,

x = (20 .5 8 .54 11 .44| 24 .54 1 .52| 0 .88 2 .35| 8 .27 4 .1 - 1 .3| 5 .21 1 .95| - 2 .74 1 .65| 5 .78 12 .8 3 .15| 2 .45 3 .21| - 5 .63 0 .48 7 .11| 7 .21 0 .35 4 .36| - 0 .58 2 .63| 1 .51 3 .49),

corresponds to the solution

x^{'} = (010 | 01 | 10 | 001 | 01 | 10 | 001 | 10 | 100 | 010 | 10 | 10) .

This solution represents a single store profile with the following configuration: Quality of fresh foods and vegetables is at the second level, presentation of products is at the second level, variety of product categories is at the first level, prices are at the third level, store brand is at the second level, assortment is at the first level, store size is at the third level, number of shoppers is at the first level, location is at the first level, distance is at the second level, opening hours are at the first level, and parking facilities are at the first level (see Appendix for attribute level descriptions).

Hence, the value of each particle’s dimension is a measure of the relative probability that the corresponding attribute level will be selected. The smaller the value of a particle’s dimension the more likely it is that the corresponding level will be selected for the attribute. The particle’s velocity defines the change in the probability between iterations.

The algorithm begins with the creation of an initial population P(0) of n particles, that is, P(0) = {x₁ (0), . . . , x_n (0)}, where x_i (0), i = 1, . . . , n, corresponds to the ith particle of the initial population (iter = 0). We generate the particles at random, since there is no prior knowledge about potential good solutions that should be included in the initial population. Subsequently, we evaluate each particle according to the objective function and assign the derived value as the particle’s fitness. To calculate the fitness score of a particle that represents a profile x, we first estimate the utility value of profile x for each individual y. The utility value is the sum of the partworth values v of y that correspond to the service attribute levels of profile x, that is, $U_{y x} = \sum_{k} v_{y k}$ , where k = 1, . . . , 12.

Next, we aggregate the utility values of x across the entire sample to compute the overall customer preference f_x provided by profile x, that is $f_{x} = \sum_{i} U_{x i}$ . When the solution contains more than one-store format, the algorithm assigns to each individual the maximum utility service design and aggregates the utilities of all consumers to compute the solution’s fitness.

The velocity for each particle’s dimension is updated using Equation 1, and the dimensions of the particle’s position are updated using Equation 2. The process is then repeated from the evaluation step until a prespecified number of iterations is completed.

Different population sizes (n) as well as different values for the maximum number of iterations (i) are tested in the three-store portfolio problem for the entire data set. Table 3 illustrates an 8 × 8 full factorial design with 8 values for each of the two parameters.

Table 3.

Particle Swarm Optimization: Alternative Values for Population Size, and Number of Iterations.

Parameter	Values
Population size	20	40	60	80	100	120	140	160
Number of iterations	400	450	500	550	600	650	700	750

Note. Five replications are performed for each of the 64 combinations of the two parameters (320 runs in total). For more than 550 iterations, there is no gain in performance. We employ the minimization of the total number of evaluations (population size times the number of iterations) criterion (Eberhart and Shi 2001) for choosing the right population size, which is similar to the elbow criterion used in cluster analysis. The best average performance is achieved for n = 60, since populations greater than 60 require much more iterations for the algorithm to converge and substantially increase CPU time with only a marginal improvement in the algorithm’s performance. A sensitivity analysis tests several values for one parameter while holding the other parameter constant. Ten replications are performed in each case. First, the sensitivity analysis evaluates five population sizes in the range (50, 70) at increments of five particles with i = 550. The algorithm has the best performance for n = 55. Second, the sensitivity analysis evaluates nine values for the maximum number of iterations (510, 520, 530, 540, 550, 560, 570, 580, 590) with n = 55. For number of iterations greater than 520, the algorithm’s performance does not improve.

Results

Consumer Preferences for Supermarket Service Elements

Table 4 presents the estimated consumer preferences for the various supermarket service elements. It reports the mean partworths of the total sample and each market separately. The results suggest that merchandise is the most important facet for the whole sample (mean importance is 40.01), followed by the facet of convenience (mean importance is 32.34). With regards to service attributes, distance from the supermarket is by far the most important characteristic (mean importance 16.72), followed by price (13.23), parking facilities (11.39), quality of fresh foods and vegetables (11.34), supermarket size (7.95), presentation of the products (7.20), opening hours (6.37), variety of product categories (6.36), number of customers (5.72), assortment (5.61), store brands (5.08), and location in terms of neighborhood image (3.05). With regards to market differences, Merchandise characteristics are less important in Market A (37.85) as compared to Market B (41.89), while layout is more important in Market A (15.34) as compared to Market B (10.91).

Table 4.

Estimated Consumer Preferences for Supermarket Facets, Attributes, and Levels.

	Total Sample			Market A			Market B
Facet, Attribute, and Level	Relative Facet Importance	Relative Attribute Importance	Attribute Level Part-Worth Values	Relative Facet Importance	Relative Attribute Importance	Attribute Level Part-Worth Values	Relative Facet Importance	Relative Attribute Importance	Attribute Level Part-Worth Values
Merchandise	40.01			37.85			41.89
1. Quality of fresh foods and vegetables		11.34			9.36			13.06
Level 1			−0.144			−0.120			−0.165
Level 2			−0.068			−0.058			−0.077
Level 3			0.076			0.044			0.103
2. Presentation of the products on the shelves		7.20			8.20			6.32
Level 1			−0.033			0.010			−0.071
Level 2			−0.210			−0.197			−0.222
3. Variety of product categories		6.36			7.51			5.35
Level 1			−0.020			0.008			−0.045
Level 2			−0.150			−0.159			−0.142
4. Level of prices		13.23			10.52			15.59
Level 1			0.119			0.034			0.194
Level 2			0.081			0.044			0.114
Level 3			−0.149			−0.129			−0.167
5. Store brands		5.08			5.27			4.92
Level 1			−0.149			−0.129			−0.168
Level 2			−0.211			−0.195			−0.226
6. Assortment		5.61			5.02			6.12
Level 1			−0.059			−0.087			−0.036
Level 2			−0.163			−0.158			−0.167
Layout	12.97			15.34			10.91
7. Supermarket size		7.95			10.39			5.82
Level 1			0.329			0.387			0.278
Level 2			0.364			0.417			0.319
Level 3			0.113			0.145			0.085
Clientele	14.67			14.93			14.45
8. Number of customers		5.72			5.84			5.61
Level 1			0.157			0.126			0.183
Level 2			−0.138			−0.162			−0.118
9. Location image		3.05			3.16			2.94
Level 1			−0.106			−0.110			−0.103
Level 2			−0.075			−0.085			−0.066
Level 3			−0.165			−0.185			−0.148
Convenience	32.34				16.42			16.99
10. Distance		16.72	0.381			0.343			0.413
Level 1			0.147			0.180			0.117
Level 2			−0.072			−0.053			−0.088
Level 3					6.84			5.95
11. Opening hours		6.37	0.091			0.090			0.092
Level 1			0.039			0.061			0.020
Level 2					11.47			11.32
12. Parking facilities		11.39	0.209			0.156			0.255
Level 1			−0.124			−0.154			−0.097
Level 2

Performance of the Bio-inspired Models

We compare the two bio-inspired models with a complete enumeration approach and simulated annealing. The latter is an established optimization algorithm inspired by the physical process of annealing in metallurgy. Belloni et al. (2008) compared nine methods on the optimal product line design problem and found that Simulated Annealing (SA) displayed the best performance. SA begins with the creation and evaluation of a random solution of the problem (store-type portfolio), followed by a series of random attribute changes. All the changes that improve the objective function (overall customer preference) are accepted. In order for the algorithm to escape from potential local minima, some changes that decrease the objective function are also accepted with a probability: $P_{a c c e p t a n c e} = e x p (\frac{C P - C P^{'}}{T})$ , where CP and CP’ are the customer preference after and before the change, respectively, and T a controlling parameter analogous to the “temperature” in the physical process of annealing. T begins with a large value where many negative changes are accepted favoring a wide search of the problem space (global search) and is gradually decreased to a very small value (local search) as the algorithm proceeds (this process is called the cooling schedule).

The four approaches find solutions for one, two, or three different store designs in the entire data set and each region separately. Twenty replications are performed in each case, except of course for the complete enumeration approach is implemented only. Each replication provides a final population of 160 chromosomes for the GA, and 55 particles for the PSO, along with their fitness scores, from which one can choose the best or any other solution with fitness close to the best. With regard to the SA algorithm, we divide the problem into 24 time stages for the one- and two-store problems and 28 time stages for the three-store problem, and test 8,000 and 10,000 attribute changes in each stage, respectively. Following the Belloni et al. (2008), SA implementation we initially set T equal to 21 in the first case and 1,443 in the second case and calculate each subsequent value (next time stage) by multiplying the existing value by 0.9 and 0.8, respectively. The best solution found after the 1,920,000 and 2,800,000 attribute changes is returned as the final solution of the problem. The global optimal solutions are the ultimate yardstick for assessing the quality of the derived solutions. In the single store portfolio problems, the global optimum can be found and verified using a complete enumeration of the search space. However, in the multiple store portfolio problems we cannot implement a full search, since it requires an excessive amount of computer memory space. Hence, we use the IBM ILOG CPLEX Optimization Studio (version 12.2) which constitutes a state-of-the-art mathematical programming solver for optimization problems of high complexity. Specifically, we use the TOMLAB edition of the CPLEX software that runs in the MATLAB environment. A comparison of the four implementations is illustrated in Table 5.

Table 5.

Performance Comparison of the Algorithms.

	Genetic Algorithm	Particle Swarm Optimization	Simulated Annealing	CPLEX
Two-store portfolio
Average CPU time	2.17 sec	13.08 sec	24.23 sec	8 hr
Number of times the global optimum was reached	60/60	60/60	60/60
Three store portfolio
Average CPU time (sec)	5.73 sec	16.22 sec	44.69 sec	113 hr
Number of times the global optimum was reached	54/60	56/60	56/60

CPU = Central Processing Unit.

The CPLEX optimizer reached the global optimal solution after 8 hours in the two-store portfolio problem (9 × 10⁸ possible solutions) and after 4½ days in the three-store portfolio problem (10¹² possible solutions), while the other three approaches solved the problems in just a few seconds. The results demonstrate that both GA and PSO represent efficient search methods for even the largest real conjoint optimization problems such as the Courtyard by Marriott concept that included a total of 167 attribute levels and 10⁹ possible solutions (Wind et al. 1989). Concerning the best solution found, the two biologically inspired approaches display comparable performance with the state-of-the-art SA algorithm. However, they both have the advantage of providing the decision maker with a wide range of near optimal solutions (final population of chromosomes and particles), whereas the SA returns a single final solution.

Derived Optimal Solutions

As expected, both proposed heuristic mechanisms derive the same optimal solutions for each problem. Tables 6 and 7 report the derived store portfolios, the utility for each profile, the portfolio fitness indices (overall portfolio utility), and the percentage of customers assigned to each profile for the global optimal solutions of each problem.

Table 6.

Derived Optimal Service Portfolios.

Solution	Profile		Utility	Choice Share (%)	Portfolio Fitness Index
Single-store portfolio	1st	200000101000	122.93	100	122.93
Two-store portfolio	1st	200100001000	72.87	50.86	140.16
Two-store portfolio	2nd	200000101010	67.29	49.14	140.16
Three-store portfolio	1st	200000101000	61.05	39.65	145.42
	2nd	200100001000	53.77	35.34
	3rd	200100100010	30.60	25.01

Table 7.

Locally Diversified Service Portfolios.

Solution	Profile		Utility	Choice Share (%)	Portfolio Fitness Index
Market A
Single-store portfolio	1st	200100101000	50.67	100	50.67
Two-store portfolio	1st	200000001000	34.86	55.56	61.74
Two-store portfolio	2nd	200100101010	26.88	44.44	61.74
Three-store portfolio	1st	200000101000	28.47	46.29	64.55
	2nd	200110001010	21.08	31.48
	3rd	200100100110	15.00	22.23
Market B
Single-store portfolio	1st	200000101000	72.79	100	72.79
Two-store portfolio	1st	200000101000	60.85	72.58	79.51
Two-store portfolio	2nd	201110000010	18.66	27.42	79.51
Three-store portfolio	1st	200000101000	45.66	51.61	83.26
	2nd	200100000000	22.02	27.42
	3rd	100000100010	15.58	20.97

Inspection of Table 6 reveals interesting patterns. The optimal service portfolios differ across the derived solutions. For example, the store type of the single-store solution is not included in the two-store solution, suggesting discrepancies between a uniform and a differentiated approach to the market. More specifically, the single-store solution represents a medium-sized uncrowded supermarket, which is located in an affluent neighborhood, a 5-minute distance away from its main shoppers. This supermarket offers well-presented, high-quality merchandise and a wide assortment of brands (both manufacturer and store brands) in both food and nonfood product categories at relatively low prices. Finally, it has good parking facilities and operates 24 hours a day. On the other hand, the two-store solution includes a large supermarket of average prices (store type 1) and a supermarket that operates 12 hours a day (store type 2). Similarly, the second store format of the two-store portfolio is not included in the three-store solution.

We now turn to derive localized portfolio solutions for each market. Inspection of Table 7 reveals that the optimal service portfolios derived for Market A differ significantly from their Market B counterparts.

The two markets share only a single common service design (store type 1 of their three-store portfolios) and differ in all other optimal service designs of their portfolios. The results illustrate that localized diversification strategies may be necessary for multistore, multimarket operations. For example, all Market B’s store portfolios include at least one low price store, but this is not the case in Market A. Furthermore, Market B’s three-store portfolio includes a supermarket of average quality (Service Design 3), but the latter service design does not appear in any of Market A’s solutions, which include stores of higher quality merchandise. Similarly, Market B’s two-store portfolio includes a food-only store (Service Design 2), but the latter service design does not appear in any of Market A’s optimal solutions. Finally, Market A’s three-store portfolio includes a supermarket that is 15 minutes away from its customers (Service Design 3), but the latter store type does not appear in any of Market B’s optimal solutions.

From the foregoing analysis, certain conclusions can be drawn. For example, consumers in Market B seem to be more price-conscious and less willing to travel long distances. Consumers in Market A seem to be more sensitive to quality issues and more prone to one-stop shopping. Finally, Market A is more heterogeneous as none of the derived service portfolios share any common service designs. Market B is less heterogeneous as the variant of the single-store solution is included in both diversified portfolios. It is easily understood that retailers have to use different diversification strategies in the two regions.

Despite the differences in the derived store portfolios suggesting a localized diversification approach, there are also some important similarities. For example, stores that offer limited merchandise, have a poorly presented assortment, are too crowded with customers, or have poor parking facilities are unacceptable in both markets and not included in any of the derived portfolios.

Note finally that store-type choice shares differ markedly within all diversified portfolios of both markets. The distribution of demand across the elements of a diversified service portfolio is asymmetric, suggesting that successful diversification requires identifying not only which service designs to combine but also the significance of each variant in terms of customer demand. In conclusion, the foregoing analysis provides evidence in support for localized service differentiation strategies and shows how service portfolios can be adapted to local markets.

Managerial Implications

The present article introduces bio-inspired modeling to strategic service management. It shows how the biological frameworks of evolution and swarm intelligence can be applied to preference data to address service diversification problems and identify optimal service portfolios.

The proposed approach is integrated and comprehensive. It provides key answers to strategic service management in a unified, empirically derived framework: optimal service design, optimal mix of service designs in terms of consumer demand, and local adaptation of service portfolios. The adaptation to local market conditions is a rather natural extension for this bio-inspired framework where the interaction of marketing mix with market environment is explicit and parallels the adaptability of biological organisms and processes to environmental variation. The foregoing analysis also illustrates the complexities associated with service diversification strategies and the need for data-driven decisions.

The findings imply that diversifying an existing service portfolio may take more than merely adding new variants. It would be necessary to modify current service designs in the sense earlier sketched.

The derived utility levels suggest compellingly that variation in service portfolios elevates customer utility and diversified offerings better match heterogeneous customer needs. In a very intuitive fashion, consumer diversity is such that a uniform service portfolio is inferior to differentiated offerings, especially with regard to salient service attributes.

The results also imply that localized diversification strategies may be necessary for multistore, multimarket operations. This is particularly relevant to multinational service firms that deal with considerable economic, social, and cultural differences. Consumers are sufficiently heterogeneous in their preferences for service design characteristics to justify localized strategies. To successfully develop such strategies, firms must address differences in customer preferences for service product and process attributes such as the ones considered in our illustrative application. We have shown how firms can design service facilities that maximize customer utility by optimizing diversification of service portfolios and adapting critical service attributes to local markets. It is worth emphasizing that the implementation of such adaptation strategies is easier in services than in manufacturing, since the economic benefits of standardization are less pronounced.

In conclusion, tangible and intangible design dimensions of the service delivery system such as the ones considered here are important determinants of customer experience, preference, satisfaction, and loyalty. The proposed methodology can assist in the design of customer-driven service delivery systems that improve customer experience, increase customer utility, and create satisfaction and loyalty.

Limitations and Research Implications

The literature in the area of conjoint analysis has grown at an impressive rate and has considered different types of conjoint tasks (i.e., rankings, ratings, and choices), alternative data collection techniques (i.e., compositional, decompositional, and hybrid approaches), as well as techniques to estimate structural parameters (e.g., Bayesian estimation methods). It would be desirable to extend the framework to other conjoint techniques such as adaptive conjoint analysis and choice-based conjoint. Of particular note is the comparison between hybrid conjoint and hierarchical Bayes conjoint in this combined experimental and bio-inspired framework.

To keep the scope within reasonable limits, we do not consider competitive pressure or the cost structure of a multistore, multimarket design. The extension one might consider is to include competition and cost variables and provide a more integrated analysis of the forces that shape optimal service portfolios.

Another direction to extend the present framework is to jointly consider experiential design areas such as back-office support, physical environment, service employees, service delivery process, and fellow customers. Similarly, it has been suggested that the importance of the experiential design areas may depend on the nature of the service being delivered (Zomerdijk and Voss 2010). Therefore, useful extensions may include different service sectors where success also depends heavily on customer experience, utility, and satisfaction such as hotels, restaurants, and health care.

From a theoretical viewpoint, the preceding analysis employs two biological metaphors derived from evolutionary and collective systems. Both assume an evolving population of candidate solutions and determine market-driven offerings that are evolved to match consumer preferences. The search in the evolutionary system of the GA is driven by competition, and the search in the collective system of swarm intelligence is driven by cooperation among candidate solutions (Floreano and Mattiussi 2008). In the evolutionary system of the GA, service offerings are evolving units and possess multiple characteristics of selective significance. Firms’ offerings compete for survival and the ones that are best adapted to the market environment succeed. Evolutionary modeling can be applied to similar decision problems such as product management (Hruschka 2008; Steiner and Hruschka 2003). In the collective system of PSO, individual agents interact locally and follow simple behavioral rules that create patterns of intelligent group behavior through self-organizing processes. Thus, PSO explores collective problem solving without centralized coordination of individual agents.

We hope that the preceding discussion demonstrates the potential for using biological models in service and marketing research and will stimulate work in this emerging and exciting field.

Footnotes

Appendix

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The first author acknowledges support from the Basic Research Funding Program of the Athens University of Economics & Business.

References

The Jin

Kachitvichyanukul

Voratas

(2009), “A Particle Swarm Optimization for the Vehicle Routing Problem with Simultaneous Pickup and Delivery,” Computers & Operations Research, 36 (5), 1693–1702.

Albritton

David M.

McMullen

Patrick R.

(2007), “Optimal Product Design Using a Colony of Virtual Ants,” European Journal of Operational Research, 176 (1), 498–520.

Alexouda

Georgia

Paparrizos

Konstantinos

(2001), “A Genetic Algorithm Approach to the Product Line Design Problem Using the Seller’s Return Criterion: An Exhaustive Comparative Computational Study,” European Journal of Operational Research, 134 (1), 165–178.

Balakrishnan

Sundar P. V.

Gupta

Rakesh

Jacob

Varghese

(2004), “Development of Hybrid Genetic Algorithms for Product Line Designs,” IEEE Transactions on Systems, Man, and Cybernetics, 34 (1), 468–483.

Belloni

Alexandre

Freund

Robert

Selove

Matthew

Simester

Duncan

(2008), “Optimizing Product Line Designs: Efficient Methods and Comparisons,” Management Science, 54 (9), 1544–1552.

Berry

Leonard L.

Lampo

Sandra K.

(2000), “Teaching an Old Service New Tricks: The Promise of Service Redesign,” Journal of Service Research, 2 (3), 265–275.

Besharati

Babak

Luo

Lan

Azarm

Shapour

Kannan

P. K.

(2006), “Multi-Objective Single Product Robust Optimization: An Integrated Design and Marketing Approach,” ASME Journal of Mechanical Design, 128 (4), 884–892.

Brown

James R.

(2010), “Managing the Retail Format Portfolio: An Application of Modern Portfolio Theory,” Journal of Retailing and Consumer Services, 17 (1), 19–28.

Clerc

Maurice

(2004), “Discrete Particle Swarm Optimization, Illustrated by the Traveling Salesman Problem,” in New Optimization Techniques in Engineering, Studies in Fuzziness and Soft Computing, Vol. 141, Onwubolu

Godfrey C.

Babu

B. V.

, eds. Berlin, Germany: Springer, 219–239.

10.

Cook

Lori S.

Bowen

David E.

Chase

Richard B.

Dasu

Sriram

Stewart

Doug M.

Tansik

David A.

(2002), “Human Issues in Service Design,” Journal of Operations Management, 20 (2), 159–174.

11.

Cooper

Robert G.

de Brentani

Ulricke

(1991), “New Industrial Financial Services: What Distinguishes the Winners,” Journal of Product Innovation Management, 8 (2), 75–90.

12.

Cooper

Robert G.

Easingwood

Christopher J.

Edgett

Scott

Kleinschmidt

Elko J.

Storey

Chris

(1994), “What Distinguishes the Top Performing New Products in Financial Services,” Journal of Product Innovation Management, 11 (4), 281–299.

13.

D’Souza

Bryan

Simpson

Timothy W.

(2003), “A Genetic Algorithm Based Method for Product Family Design Optimization,” Engineering Optimization, 35 (1), 1–18.

14.

de Brentani

Ulrike

(1995), “New Industrial Service Development: Scenarios for Success and Failure,” Journal of Business Research, 32 (2), 93–103.

15.

de Brentani

Ulrike

Cooper

Robert G.

(1992), “Developing Successful New Financial Services for Businesses,” Industrial Marketing Management, 21 (3), 231–241.

16.

Eberhart

Russell C.

Shi

Yuhui

(2001), “Particle Swarm Optimization: Development, Applications and Resources,” Proceedings of the 2001 Congress on Evolutionary Computation, Seoul, South Korea, 1, 81–86.

17.

Farrell

Ronald S.

Simpson

Timothy W.

(2009), “Improving Cost Effectiveness in an Existing Product Line Using Component Product Platforms,” International Journal of Production Research, 48 (11), 3299–3317.

18.

Fließ

Sabine

Kleinaltenkamp

Michael

(2004), “Blueprinting the Service Company: Managing Service Processes Efficiently,” Journal of Business Research, 57 (4), 392–404.

19.

Floreano

Dario

Mattiussi

Claudio

(2008), Bio-Inspired Artificial Intelligence. Theories, Methods, and Technologies. Cambridge: MIT Press.

20.

Froehle

Craig M.

Roth

Aleda V.

Chase

Richard B.

Voss

Christopher A.

(2000), “Antecedents of New Service Development Effectiveness: An Exploratory Examination of Strategic Operations Choices,” Journal of Service Research, 3 (1), 3–17.

21.

González-Benito

Oscar

(2005), “Spatial Competitive Interaction of Retail Store Formats: Modeling Proposal and Empirical Results,” Journal of Business Research, 58 (4), 457–466.

22.

Goldbarg

Elizabeth F. G.

de Souza

Givanaldo R.

Goldbarg

Marco C.

(2006), “Particle Swarm Optimization for the Bi-objective Degree-Constrained Minimum Spanning Tree,” Proceedings of IEEE Congress on Evolutionary Computation, Vancouver, BC, Canada, 420–427.

23.

Green

Paul E.

Krieger

Abba M.

(1996), “Individualized Hybrid Models for Conjoint Analysis,” Management Science, 42 (6), 850–867.

24.

Guo

Wenzhong

Chen

Guolong

Feng

Xiang

Lun

(2007), “Solving Multicriteria Minimum Spanning Tree Problem with Discrete Particle Swarm Optimization,” Proceedings of the Proceedings of the Third International Conference on Natural Computation (ICNC '07), Vol. 04, 471–478.

25.

Gupta

Sudheer

Vajic

Mirjana

(2000), “The Contextual and Dialectical Nature of Experiences,” in New Service Design, Fitzsimmons

James A.

Fitzsimmons

Mona J.

, eds. Thousand Oaks, CA: Sages, 33–51.

26.

Gustafsson

Anders

Johnson

Michael D.

(2004), “Determining Attribute Importance in a Service Satisfaction Model,” Journal of Service Research, 7 (2), 124–141.

27.

Gwinner

Kevin P.

Bitner

Mary J.

Brown

Stephen W.

Kumar

Ajith

(2005), “Service Customization through Employee Adaptiveness,” Journal of Service Research, 8 (2), 131–148.

28.

Hair

Joseph F.

Anderson

Rolph E.

Tatham

Ronald L.

Black

William

(2006), Multivariate Data Analysis. UpperSaddle River, NJ: Prentice Hall.

29.

Heese

Hans Sebastian

Swaminathan

Jayashankar M.

(2006), “Product Line Design with Component Commonality and Cost-Reduction Effort,” Manufacturing & Service Operations Management, 8 (2), 206–219.

30.

Heineke

Janelle

Davis

Mark M.

(2007), “The Emergence of Service Operations Management as an Academic Discipline,” Journal of Operations Management, 25 (2), 364–374.

31.

Hill

Arthur V.

Collier

David A.

Froehle

Craig M.

Goodale

John C.

Metters

Richard D.

Verma

Rohit

(2002), “Research Opportunities in Service Process Design,” Journal of Operations Management, 20 (2), 189–202.

32.

Howard

Mona W.

Sobol

Marion G.

(2004), “Combining Economic and Conjoint Analysis to Determine Optimal Academic Services,” Decision Sciences Journal of Innovative Education, 2 (1), 27–44.

33.

Hruschka

Harald

(2008), “Neural Nets and Genetic Algorithms in Marketing,” in Handbook of Marketing Decision Models, Wierenga

Berend

, ed. Berlin, Germany: Springer, 399–433.

34.

Jiao

Jianxin

Zhang

Yiyang

(2005), “Product Portfolio Planning with Customer-Engineering Interaction,” IIE Transactions, 37 (9), 801–814.

35.

Johnson

Susan Paul

Menor

Larry J.

Roth

Aleda V.

Chase

Richard B.

(2000), “A Critical Evaluation of the New Service Development Process,” in New Service Development, Fitzsimmons

James A.

Fitzsimmons

Mona J.

eds. Thousand Oaks, CA: Sage, 1–32.

36.

Kannan

P. K.

Pope

Barbara Kline

Jain

Sanjay

(2009), “Pricing Digital Content Product Lines: A Model and Application for the National Academies Press,” Marketing Science, 28 (4), 620–636.

37.

Keep

William W.

Hollander

Stanley C.

Calantone

Roger J.

(1996), “Retail Diversification in the USA. Are There Performance Benefits?,” Journal of Retailing and Consumer Services, 3 (1), 1–9.

38.

Kennedy

James

(1997), “The Particle Swarm: Social Adaptation of Knowledge,” Proceedings of the IEEE International Conference on Evolutionary Computation. Indianapolis, IN, Piscataway, NJ: IEEE Service Center, 303–308.

39.

Kennedy

James

Eberhart

Russell C.

(1995), “Particle Swarm Optimization,” Proceedings of the IEEE International Conference on Neural Networks, IV, Piscataway, NJ: IEEE Service Center, 1942–1948.

40.

Knudsen

Hannah K.

Roman

Paul M.

Ducharme

Lori J.

(2005), “Does Service Diversification Enhance Organizational Survival? Evidence from the Private Substance Abuse Treatment System,” Journal of Behavioral Health Services and Research, 32 (3), 241–252.

41.

Kohli

Rajeev

Sukumar

Ramamirtham

(1990), “Heuristics for Product Line Design Using Conjoint Analysis,” Management Science, 36 (12), 1464–1478.

42.

Kohli

Rajeev

Krishnamurti

Ramesh

(1989), “Optimal Product Design Using Conjoint Analysis: Computational Complexity and Algorithms,” European Journal of Operational Research, 40 (2), 186–195.

43.

Krishnan

Vijaya

Ulrich

Karl T.

(2001), “Product Development Decisions: A Review of the Literature, Management Science, 47 (1), 1–21.

44.

Kumar

Deepak

Chen

Wei

Simpson

Timothy W.

(2009), “A Market-Driven Approach to the Design of Platform-Based Product Families,” International Journal of Production Research, 47 (1), 71–104.

45.

Lancaster

Kelvin

(1990), “The Economics of Product Variety,” Marketing Science, 9 (3), 189–206.

46.

Laskari

Elena C.

Parsopoulos

Konstantinos E.

Vrahatis

Michael N.

(2002), “Particle Swarm Optimization for Integer Programming,” Proceedings of the 2002 Congress on Evolutionary Computation, Vol. 2, 1582–1587.

47.

Lee

Moonkyu

Ulgado

Francis M.

(1997), “Consumer Evaluations of Fast-Food Services: A Cross-National Comparison”, Journal of Services Marketing, 11 (1), 39–52.

48.

Hui

Azarm

Shapour

(2002), “An Approach for Product Line Design Selection under Uncertainty and Competition,” Journal of Mechanical Design, 124 (3), 385–392.

49.

Luo

Lan

(2011), “Product Line Design for Consumer Durables: An Integrated Marketing and Engineering Approach,” Journal of Marketing Research, 48 (1), 128–139.

50.

Luo

Lan

Kannan

P. K.

Ratchford

Brian T.

(2008), “Incorporating Subjective Characteristics in Product Design and Evaluations,” Journal of Marketing Research, 45 (2), 182–194.

51.

Luo

Lan

Kannan

P. K.

Besharati

Babak

Azarm

Shapour

(2005), “Design of Robust New Products under Variability: Marketing Meets Design,” Journal of Product Innovation Management, 22 (2), 177–192.

52.

Malthouse

Edward C.

Oakley

James L.

Calder

Bobby J.

Iacobucci

Dawn

(2004), “Customer Satisfaction across Organizational Units,” Journal of Service Research, 6 (3), 231–242.

53.

Marinakis

Yannis

Marinaki

Magdalene

(2010), “A Hybrid Multi-Swarm Particle Swarm Optimization Algorithm for the Probabilistic Traveling Salesman Problem,” Computers & Operations Research, 3 (37), 432–442.

54.

Martin

Claude R.

Horne

David A.

(1993), “Services Innovation: Successful versus Unsuccessful Firms,” International Journal of Service Industry Management, 4 (1), 49–65.

55.

Mattila

Anna S.

(1999), “The Role of Culture in the Service Evaluation Process,” Journal of Service Research, 1 (3), 250–261.

56.

McGoldrick

Peter J.

(1998), “Spatial and Temporal Shifts in the Development of International Retail Images,” Journal of Business Research, 42 (2), 189–196.

57.

Michalek

Jeremy J.

Feinberg

Fred M.

Papalambros

Panos Y.

(2005), “Linking Marketing and Engineering Product Design Decisions via Analytical Target Cascading,” Journal of Product Innovation Management, 21 (1), 42–62.

58.

Michalek

Jeremy J.

Ceryan

Oben

Papalambros

Panos Y.

Koren

Yoram

(2006), “Balancing Marketing and Manufacturing Objectives in Product Line Design,” Journal of Mechanical Design, 128 (6), 1196–1204.

59.

Nair

Suresh K.

Thakur

Lakshman S.

Wen

Kuang-Wei

(1995), “Near Optimal Solutions for Product Line Design and Selection: Beam Search Heuristics,” Management Science, 41 (5), 767–785.

60.

Patrício

Lia

Fisk

Raymond P.

Cunha

Joao Falcao e

(2008), “Designing Multi-Interface Service Experiences: The Service Experience Blueprint,” Journal of Service Research, 10 (4), 318–334.

61.

Poli

Riccardo

Kennedy

James

Blackwell

Tim

(2007), “Particle Swarm Optimization. An Overview,” Swarm Intelligence, 1 (1), 33–57.

62.

Pullman

Madeleine E.

Gross

Michael A.

(2004), “Ability of Experience Design Elements to Elicit Emotions and Loyalty Behaviors,” Decision Sciences, 35 (3), 551–578.

63.

Pullman

Madeleine E.

Moore

William L.

(1999), “Optimal Service Design: Integrating Marketing and Operations Perspectives,” International Journal of Service Industry Management, 10 (2), 239–260.

64.

Rai

Rahul

Allada

Venkat

(2006), “Agent-Based Optimization for Product Family Design,” Annals of Operations Research, 143 (1), 147–156.

65.

Ramdas

Kamalini

(2003), “Managing Product Variety: An Integrative Review and Research Directions,” Production Operations Management, 12 (1), 79–101.

66.

Roth

Aleda V.

Menor

Larry J.

(2003), “Insights into Service Operations Management: A Research Agenda,” Production and Operations Management, 12 (2), 145–164.

67.

Safarzynska

Karolina

van den Bergh

Jeroen C. J. M.

(2010), “Evolutionary Models in Economics: A Survey of Methods and Building Blocks,” Journal of Evolutionary Economics, 20 (3), 329–373.

68.

Severin

Valerie

Louviere

Jordan J.

Finn

Adam

(2001), “The Stability of Retail Shopping Choices Over Time and Across Countries,” Journal of Retailing, 77 (2), 185–203.

69.

Shi

Xiaohu H.

Liang

Yanchun Chun

Lee

Heow Pueh

Lung Lu

Chin

Wang

Q. X.

(2007), “Particle Swarm Optimization-Based Algorithms for TSP and Generalized TSP,” Information Processing Letters, 103 (5), 169–176.

70.

Song

Jiwoong

Jang

Taewon

Y. Sohn

(2009), “Conjoint Analysis for IPTV Service,” Expert Systems with Applications, 36 (4), 7860–7864.

71.

Steiner

Winfried

Hruschka

Harald

(2003), “Genetic Algorithms for Product Design: How Well do they Really Work?,” International Journal of Market Research, 45 (2), 229–240.

72.

Tansik

David A.

Smith

William L.

(2000), “Scripting the Service Encounter,” in New Service Design, Fitzsimmons

James A.

Fitzsimmons

Mona J.

, eds. Thousand Oaks, CA: Sage, 239–263.

73.

Tasgetiren

Fatih M.

Sevkli

Mehmet

Liang

Yun-Chia

Gencyilmaz

Gunes

(2004), “Particle Swarm Optimization Algorithm for Single Machine Total Weighted Tardiness Problem,” Evolutionary Computation, CEC2004 congress, Vol. 2, 1412–1419.

74.

Tax

Stephen S.

Stuart

Ian

(1997), “Designing and Implementing New Services: The Challenges of Integrating Service Systems,” Journal of Retailing, 73 (1), 105–134.

75.

Vilnai-Yavetz

Iris

Rafaeli

Anat

(2006), “Aesthetics and Professionalism of Virtual Servicescapes,” Journal of Service Research, 8 (3), 245–259.

76.

Wind

Jerry

Green

Paul E.

Shifflet

Douglas

Scarbrough

Marsha

(1989), “Courtyard by Marriott: Designing a Hotel Facility with Consumer-Based Marketing Models,” Interfaces, 19 (1), 25–47.

77.

Zhu

Qing

Qian

Limin

Yingchun

Zhu

Shanjun

(2006), “An Improved Particle Swarm Optimization Algorithm for Vehicle Routing Problem with time Windows,” Proceedings of the 2006 IEEE congress on evolutionary computation, Vancouver, Canada, July 16–21, 1386–1390.

78.

Zomerdijk

Leonieke G.

Voss

Christopher A.

(2010), “Service Design for Experience-Centric Services,” Journal of Service Research, 13 (1), 67–82.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.08 MB

0.00 MB