Integrating conjoint analysis with quality function deployment to carry out customer-driven concept development for ultrabooks

(1)

Integrating conjoint analysis with quality function deployment to carry out

customer-driven concept development for ultrabooks

Chih-Hsuan Wang

⁎

, Chih-Wen Shih

Department of Industrial Engineering & Management, National Chiao Tung University, Taiwan

a b s t r a c t

a r t i c l e i n f o

Article history:

Received 19 February 2013 Received in revised form 12 May 2013 Accepted 24 July 2013

Available online 3 August 2013 Keywords:

Concept generation Prototype evaluation Conjoint analysis Quality function deployment DEMATEL

TOPSIS

A hybrid framework integrating conjoint analysis (CA) with quality function deployment (QFD) is presented to incorporate customer preferences into the process of product development. In particular, the proposed framework constitutes two sequential phases, namely, concept generation based on CA and prototype eval-uation based on QFD. In addition, product features are characterized by customer requirements (CRs) and functional attributes (FAs). By means of DEMATEL (Decision Making and Trial Laboratory), the impacts of FAs on CRs are systematically identiﬁed to visualize their causalities. Instead of utilizing FAs, the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is employed to assess potential prototypes in terms of CRs.

1. Introduction

In an era of global customization, owing to dynamically changing customer desires coupled with rapid advances in manufacturing tech-nologies, today's marketplaces are full of various product offerings. Back to 2011, Ultrabooks are designed to feature reduced size (less than 2.1 cm thick) and weight (less than 1.5 kg) without compromising system performance and battery life. Thus, low-power Intel processors with integrated graphics and unibody chassis are used toﬁt larger bat-teries into smaller cases. Different from past products like netbooks and notebooks, ultrabooks would be very thin, quite slight, and could also accommodate tablet features such as a touch screen and long battery life[29,30]. Obviously, the Ultrabook directly competes against Apple's MacBook Air, which has similar product speciﬁcations, but runs the ker-nels of Apple OS (and is capable of running Microsoft Windows). In order to avoid fatal mistakes before implementing practical product strategies, companies need to deliberately understand what customers want and desire for capturing customer preferences or customer perceptions.

In practice, new product development defined as a process of transforming an identified market opportunity into profitable prod-uct(s) for sale[4,26], usually consists of a sequence of steps in which an enterprise could employ it to accomplish the goal of commercializa-tion. Typically, the NPD process consists of the following six phases, such as initial planning, concept development, system-level design,

detail design, testing and refinement, and production ramp-up[24]. Among them, the phase of concept development is of critical impor-tance because it does not only impact on downstream activities of the whole process, but also influence NPD's overall success, significantly. In particular, the process of concept development includes a couple of representative activities: (1) identifying customer needs, (2) concept generation, (3) concept selection, (4) cost analysis, (5) prototype test-ing, and (6) benchmarking analysis[14,15]. In this paper, we particular-ly focus on two critical activities, nameparticular-ly, concept generation and concept selection. Needless to say, product development without incor-porating customer involvement into the process of concept develop-ment is doomed to failure since huge gaps might exist between perceived customer requirements (CRs) and con_{figured functional} attri-butes (FAs).

Specifically, five main schemes are commonly adopted for concept evaluation, including utility theory, analytical hierarchy/network pro-cess (AHP/ANP), graphical methods, fuzzy logic approaches, and QFD matrices[2,3]. Apparently, most of the above methods are fully reliant on subjective human assessment or experts' domain knowledge. For in-stance, a pairwise comparison among two alternatives is often applied to a respondent by asking the following question: How much degree is concept A preferred to concept B with respect to a specific dimension? Ap-parently, due to lack of concrete product features, the AHP[21]seems to be quite ambiguous in practice. Suppose there are n criteria at a hierar-chy, we need to completen n−1ð₂ Þtimes of pairwise comparisons for deriv-ing their importance[20]. Obviously, when the number of criteria or competitive alternatives is over seven, its feasibility is highly doubtful for respondents to reach a consistent result. Thus, instead of using the AHP/ANP based schemes, this study integrates the conventional QFD ⁎ Corresponding author at: 1001 University Road, Hsinchu 30013, Taiwan. Tel.: +886 3

5712121 57310; fax: +886 3 5722392.

http://dx.doi.org/10.1016/j.csi.2013.07.010

Contents lists available atScienceDirect

Computer Standards & Interfaces

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c s i

(2)

with conjoint analysis (CA) to incorporate customer preference and utility into the decision-making process of concept development. In ad-dition, arti_{ficial neural networks (ANNs) and clustering techniques have} been sorely utilized or fused together to helpfirms achieve product con-ceptualization, product definition and product customization[7,28].

For clarity,Table 1brie_{ﬂy compares our proposed approach with} other past studies. Unfortunately, most of the previous studies rarely ex-plore the impacts of FAs on CRs and hence it is quite challenging for them to assess product alternatives in a customer-driven way. Followed by[5,10,14,15,22,23,25], a market-oriented approach is presented and several crucial issues are addressed below:

● Based on the QFD platform, product features are characterized by perceived customer requirements (CRs) and conﬁgurable functional attributes (FAs) and a systematic approach is offered to identify the causal impacts of FAs on CRs,

● With respect to distinct segments, CA is employed to extract cus-tomer utilities of FAs for generating design concepts in a cuscus-tomer- customer-driven way,

● With consideration of manufacturing costs, prototype alternatives are prioritized in terms of market-oriented CRs for offering manage-rial implications.

In particular, two fundamental design phases are emphasized in this study: phase 1 for concept generation and phase 2 for prototype evalua-tion. The rest of this paper is structured as follows.Section 2brie_ﬂy over-views conjoint analysis and quality function deployment. Section 3

presents the proposed framework. An industrial example regarding conﬁguring varieties of ultrabooks is illustrated inSection 4. Concluding remarks areﬁnally drawn inSection 5.

2. Overview of quality function deployment and conjoint analysis In response to customer desire and much shorter product life cycle than ever, launching attractive products faster than competitors can as-sistﬁrms in not only acquiring larger market share but also reducing de-velopment lead time, signiﬁcantly. In practice, however, manufacturing companies are often struggling with the dilemma of increasing product variety or controlling manufacturing complexity [15,25]. In other words, to survive in a wide range of market segments, companies are now more aware that seeking an optimal balance between enhancing product varieties and controlling manufacturing complexities is the key to staying ahead of competitors. In order to help an enterprise better optimize product varieties at the marketplace, typical schemes includ-ing product family architecture, product platform design, and product module mix have been widely presented[7,18].

In addition, two fundamental design phases are emphasized in product conceptualization, including product definition (aiming at establishing a product platform and relevant product family as well as design alternatives) and product customization (focusing on transferring specific customer desires into corresponding product design alternatives). Speci_{fically, product conceptualization can be} practically implemented by a set of customer requirements (CRs) and functional attributes (FAs). To effectively fulfill customer satis-faction, an enterprise needs to understand how product offerings are preferred and perceived in terms of market-oriented CRs. Mean-while, to acquire new opportunities and to survive among distinct segments, an enterprise requires working out potential concepts or profitable prototypes through a series of decision-making processes

[13,17,26].

In order to facilitate research gap between product configuration and concept development, a hybrid framework combining quality func-tion deployment (QFD) with conjoint analysis (CA) is adopted in this study. As we know, the conventional QFD is good at interpreting intan-gible CRs in terms of measurable FAs for performing product de_finition, yet, it is deficient in realizing product customization, especially when incorporating customer preference or customer perception into the process of product development[5,10,22]. Furthermore, without con-sidering the interrelationships between CRs and FAs, it might be prob-lematic to give priorities to design alternatives. Thus, in this paper, CA is incorporated into the QFD and their details are briefly overviewed later. For convenience, an overall comparison between AHP, CA, and QFD is shown inTable 2 [14,16,24].

2.1. Quality function deployment (QFD)

Quality function deployment[1]originated in Japan has been widely applied to various industries for product development, service design, and competitor benchmarking. Basically, customers' desires on a specif-ic product or servspecif-ice can be represented by a set of intangible customer requirements (CRs) and thus a series of functional attributes (FAs) that impact on CRs need to be realized for accomplishing successful product development or service design. Typically, the conventional QFD consists of the following four phases[6,25]: phase one translates customer re-quirements into functional attributes; phase two translates functional attributes into part characteristics; phase three translates part charac-teristics into manufacturing operation, and phase four translates manufacturing operations into production requirements. In particular, phase one— the QFD or the so-called HOQ (house of quality), provides a communication platform to fuse diverse opinions among cross-functional team members (seeFig. 1).

Table 1

A comparison between the proposed method and other existing studies.

Market segmentation Identifying the causalities between CRs and FAsa

Concept generation and prototype evaluation Proposed method Customers' affordable prices

(pricing policies)

QFD and DEMATEL Customer-driven by integrating CA with TOPSIS Ayağ[2], Ayağ and Özdemir[3] Not applicable Not applicable Reliant on domain experts

Chaudhuri and Bhattacharyya[5] Not applicable QFD Integrating CA with integer programming

Chen et al.[7] Respondents' ages, gender, and skill levels

Not applicable, only considering FAs Customer-driven by fusing CA with Kohonen association

Fogliatto et al.[10] Choice menus Not applicable Stated preferences

Işıklar and Büyüközkan[13] Not applicable Not applicable, but including product and interface features

Combining AHP with TOPSIS Jiao and Zhang[15] Not applicable, but considering

manufacturing costs

Not applicable, only considering FAs CA

Lin et al.[17] Not applicable Not applicable, only considering FAs Combining AHP with TOPSIS

Liu and Hsiao[18] Not applicable ANP Goal programming

Sereli et al.[22] Not applicable QFD Not applicable

Yan et al.[28] Not applicable Not applicable, only considering FAs General sorting and FCM clustering

a

(3)

In order to connect intangible CRs with measurable FAs, the weights of CRs and FAs could be modiﬁed and derived as follows[25,27]:

WtCRj¼ Xm i¼1 WtFAiRij 0 ; 1≤i≤m; 1≤ j≤n; ð1Þ Rij 0_¼ Xm k¼1 Rjk γki Xm i¼1 Xm k¼1 Rjk γki ; ð2Þ

where WtCRjand WtFAirepresent the weight of CRjand FAi, respectively. Meanwhile, we assume that n customer requirements and m functional attributes exist in the QFD, Rjkmeans the dependences between FAiand

CRj(Rij' is a normalized matrix), andγkidenotes the correlations among

FAs. Once the weights of CRs are attained, they might be used to form a basis of market segmentation or incorporated into the evaluation sys-tem for prioritizing design alternatives.

2.2. Conjoint analysis (CA)

Conjoint analysis[19]is one of the most popular techniques to mea-sure diverse customers' preferences among multi-attributed products or services[7,10,15,23]. When a product is decomposed into indepen-dent multi-attributes, its overall utility could be obtained by aggregating part-worth utilities of the attributes with their associated levels. In brief, CA is in nature, a process of making trade-offs among limited alterna-tives that are characterized by various combinations of functional attri-butes. After gathering respondents' preference rankings among product alternatives, two critical measures could be obtained through CA: the importance degrees of functional attributes and the part-worth utilities of attributes associated with speciﬁc levels. Apparently, CA could be applied to analyzing customer individuals or previously-segmented groups for the purpose of target marketing. In this study, customer pref-erence is deﬁned by the perceived importance degrees of FAs and the

extracted part-worth utilities are utilized to form a basis of concept generation.

For simplification, let us illustrate a simple case to explain its ap-plicability. Suppose that a product is characterized by six attributes (e.g., A1–A6) associated with specific levels (also seeFig. 2), intuitively, a maximal number of 144 (3 × 2 × 2 × 3 × 2 × 2) combinations may be possibly generated. When deriving their part-worth utilities among respondents, it is practically infeasible to ask an evaluator to prioritize 144 alternatives at a time. Hopefully, by means of fractional factorial de-sign[11], the required alternatives for prioritizing can be significantly reduced to only 16 orthogonal samples. For convenience, a general form of CA for an alternative can be briefly modeled as follows:

Uk¼ β0þ Xm i¼1 Xn j¼1 uijk; ð3Þ

where Ukis alternative k's overall utility,β0is a regularized constant, uijk

is the part-worth of alternative k associated with attribute i and level j, m represents the number of attributes, and n denotes the number of corresponding levels. In order to derive the importance degree of func-tional attributes, it is commonly believed that attributes with a larger range of part-worth values should have a greater impact on the overall utility. Therefore, the relative weight (Wi) of attribute i can be obtained

by normalizing its range (Ri) of part-worth utility:

Wi¼ Ri X i Ri ; where Ri¼ Max j uij − Min j uij : ð4Þ 3. Proposed techniques

Referring toFig. 3, several techniques including CA (conjoint analysis), DEMATEL (decision making and trial laboratory) and TOPSIS (technique for order preference by similarity to ideal solution) are well fused into the QFD to perform market-oriented concept generation and prototype evalua-tion. For convenience, their details are operated and described as follows: ● Initially, the QFD is employed to separate product features into ei-ther perceived CRs (customer requirements) or conﬁgurable FAs (functional attributes),

● Secondly, based on customers' affordable prices, the entire market is divided into two distinct segments, namely, the business segment and the home segment,

● With respect to two identiﬁed segments, CA is respectively applied to derive respondents' perceived importance weights of FAs and ex-tract customer utilities of FAs for generating potential concepts, ● By virtue of the DEMATEL, the causal dependences of FAs on CRs are

identiﬁed and thus the importance weights of CRs can be derived, ● Finally, with the aid of TOPSIS, the priorities of selected prototypes

are systematically assessed and determined in a market-oriented manner.

Table 2

An overall comparison between QFD, CA, and AHP.

QFD CA AHP

Basic principle Mapping customer requirements

into functional attributes

Making trade-offs among simpliﬁed alternatives

Conducting pair-wise comparisons among alternatives

Handling multi-leveled product attributes Limited Good Good

Source of gathering product information Customers/experts Customers Most experts

Extracting customer preference for product features Not applicable Good Good

Recognizing correlations between CRs and FAs Good Not applicable Not applicable

Functional Attributes (FAs) Customer Requirements (CRs) Correlations among FAs Interrelationships between CRs and FAs

Marketing Benchmarking

Technical Benchmarking

(4)

3.1. Use of DEMATEL to derive the interdependences between FAs and CRs DEMATEL (decision making and trial laboratory), developed by the science and human affairs program of the Battelle Memorial Institute of Geneva Research Centre[9], is able to visualize the complex interre-lationships among interdependent factors. Through converting their causal relationships of the whole system into an intelligible structure model, the DEMATEL could distinguish all factors into either the trans-mitter group which impacts on other factors or the receiver group which is inﬂuenced by other factors[25,26]. By using a 5-point rating scale (i.e. 1: very low, 2: low, 3: moderate, 4: high, 5: very high), domain experts are required to quantify their inﬂuential measures among fac-tors and its details are described as:

● Generating the direct-relation matrix: Suppose there are m CRs and n FAs for the QFD framework, then, a (m + n) × (m + n) matrix A with elements of aijisﬁlled to denote the impact of factor i exerted

on factor j. Here, note that all diagonal elements of matrix A are ini-tially set by zero.

● Normalizing the direct-relation matrix: the normalized matrix B could be obtained through the above-mentioned matrix A (see Eqs.(5)_–(6)), in which B¼ k A ð5Þ k¼ Min 1 max j X mþn j¼1 aij ; 1 max j X mþn i¼1 aij 0 B B B B @ 1 C C C C A: ð6Þ

● Deriving the total-relation matrix: the total-relation matrix M can be derived via Eq.(7), where I denotes an identity matrix.

M¼ B þ B2

þ B3

þ ⋯ ¼ B I−Bð Þ‐1 ð7Þ

Speciﬁcally, the interdependences between FAs and CRs could be extracted through the total-relation matrix.

● Displaying a causal diagram through distinguishing the transmitter group T from the receiver group R:

Ti¼ Xn j¼1 Mij; ð8Þ Rj¼ Xn i¼1 Mij; ð9Þ

where T is the sum of rows of the total-relation matrix while R is the sum of columns.

A causal diagram is visualized by portraying the dataset comprising (T + R, T− R), where the horizontal axis represents “T + R” and the vertical axis denotes“T − R”. Intuitively, the “T + R” named “promi-nence” reveals how much important the factor is. On the other hand, the “T − R” named “inﬂuence” classiﬁes the factor into either the cause group or the effect group. In simple words, the factors in the * * L3 * * * * * * L2 * * * * * * L1 A6 A5 A4 A 3 A 2 A 1 D 3 l26 l22 21 D 2 l16 l12 l11 D 1 D 144 A6 A5 A4 A 3 A 2 A1 6 D 2 D 1 D 16 A6 A5 A4 A 3 A 2 A1

Utility decomposition into separate part-worth values of product attributes with associated levels

Concept generation through various combinations of attributes with

their multi-specification levels

Prototype evaluation (trade-off decisions) after processing by

fractional factorial design

Fig. 2. Simplifying concept generation through conjoint analysis.

Employing QFD to separate product features into customer requirements (CRs) and functional attributes (FAs)

Using customers’ affordable prices (companies’ pricing policies) to divide the entire market into distinct segments

Conducting CA to derive the weights of FAs and extract customer utilities

Utilizing DEMATEL to identify the dependences between CRs and FAs

Employing TOPSIS to carry out product evaluation in terms of market oriented CRs

(5)

cause group are acting as a“dispatcher” since they impact on the others. By contrast, the factors in the effect group are playing as a“receiver” be-cause they are affected by the others.

3.2. Use of TOPSIS to prioritize design concepts and product prototypes Originated from Hwang and Yoon[12], TOPSIS (technique for order preference by similarity to ideal solution) was proposed to seek an opti-mal solution which is the most closest to the“PIS” (positive ideal solu-tion) but the farthest from the“NIS” (negative ideal solution). Suppose there are m alternatives and n attributes (criteria), the TOPSIS are oper-ated by the following procedures[13,17]:

● Generating a decision matrix. A m × n decision matrix (i.e. m repre-sents the number of alternatives and n denotes the number of attri-butes), X, consists of the elements of xijrepresenting the performance

rating of the ith alternative with respect to the jth attribute. ● Construing a normalized decision matrix. To reduce the scale effect

among various dimensions, the normalized matrix Y is obtained as: yij¼

xij

Max

i xij

; for ∀xij∈ thesetof “benefit” attributes; ð10Þ

yij¼

Min

i xij

xij ; for ∀xij∈ thesetof “cost” attributes:

ð11Þ

Here, the“beneﬁt” attributes possess the property of “the-larger-the-better” while the “cost” attributes own the characteristic of “the-smaller-the-better”.

● Searching for the elements of “PIS” (S+_{) and}_{“NIS” (S}−_{) by using:}

Sj þ_{¼ Max} i yijjj¼ 1; 2; ::::ng; n ð13Þ Sj−¼ Mini yijjj¼ 1; 2; ::::ng; n ð14Þ where Sjþ=Sj−denotes the jth element of S+/S−, respectively.

● Measuring a weighted distance from alternative i to the “PIS” and the “NIS”: Di þ_¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn j¼1 wj yij−Sj þ 2 v u u t _{; i ¼ 1; 2; ::::m} ð15Þ Di−¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn j¼1 wj yij−Sj− 2 v u u t _{; i ¼ 1; 2; ::::m;} ð16Þ

where wjrepresents the weight of attribute j.

● Conducting a ranking index (RI) for competing alternatives (prototypes):

RIi¼

Di−

Diþþ Di−; i ¼ 1; 2; ::::m:

ð17Þ

In terms of market-oriented CRs, the weights of CRs derived through the DEMATEL are incorporated into the TOPSIS for evaluating compet-ing prototypes.

4. An illustrative example of assessing various prototypes of ultrabooks

At the Intel Developer Forum in 2011, four Taiwan ODMs showed prototype ultrabooks that used Intel's Ivy Bridge processors which only consume 17 W default thermal power. Meanwhile, Intel tries to

enhance the slumping PC markets against rising competition from tab-let computers such as the iPad, which are typically powered by the ARM-based architectures[29,30]. Originally, for fast stimulating market sales, Intel plans to set a“below $1000” price for ultrabooks. However, the presidents of Acer and Compal think that this goal could be dif ﬁcult-ly achieved if Intel did not lower the price of its CPU chips. In 2011, an Intel manager stated that market analysis should be carefully carried out to investigate customer preference for screen size since it might in-ﬂuence some of the reluctance to switch to ultrabooks. Actually, 11– 12 inch ultrabooks might replace/substitute the markets of smart pads and netbooks to some degree while 13–14 inch ultrabooks currently dominate at least 50% models for the high-end segment.

A large-scale Taiwanese OEM/ODM company planned to precisely relate customer requirements (CRs) to functional attributes (FAs) prior to launching its next-generation ultrabooks. After looking at the speci_{fications of current products and consulting experienced experts,} six representative CRs and FAs associated with multi-levels and compo-nent costs are shown inTable 3. In addition to six critical FAs, it is noted that other common components like mother board, graphics card, Wi-Fi chip, front camera, keyboard, and the pre-installed operating system, are configured in an ultrabook and their entire cost is roughly estimated around 7100 in $TWD. More specifically, the questionnaires as well as relevant supporting commercial packages are illustrated in Table 4. Based on customers' affordable prices (companies' pricing policies), the target market is partitioned into two segments, namely, the busi-ness segment (pricing around $30,000 in TWD) and the home segment (pricing around $24,000 in TWD), respectively. For differentiating launched ultrabooks from Apple's products, the above pricing policies are set with consideration of MacBookAir's high-end prices (between $36,000 and $40,000 in TWD).

4.1. Concept generation based on customer utilities of FAs

Initially, a total number of 160 respondents are invited to complete marketing surveys on ultrabooks. Speciﬁcally, 54% of the respondents consist of high-tech engineers (aged between 26 and 40) in the Hsinchu Science Park. And the remaining 46% is composed of graduate students in our university (aged between 22 and 30). All of the invited respon-dents are pre-screened to assure having experiences in using ultrabooks. After taking the multi-levels of FAs into account, intuitively, there are 144 (3 × 2 × 2 × 3 × 2 × 2) design concepts which might be possibly generated. In particular, fractional factorial design originated from the concept of design of experiments[11]is employed to reduce 144 possible concepts to 16 representative samples, as indicated by

Table 5. Moreover, to reduce the serial position effect like primacy and recency biases in psychological decision making[8], we suggest evalua-tors to focus on three key FAs when making the trade-offs among vari-ous FAs. Furthermore, to speed up the ranking process, they need toﬁrst

Table 3

Critical CRs and FAs for characterizing an ultrabook.

CR (customer requirement) FA (functional attribute) Associated levels ($TWD) R1 System performance

(benchmarking test)

A1 CPU (type) A11—i7 series ($5800) A12—i5 series ($4400) A13—i3 series ($3500) R2 Booting response time (s) A2 RAM capacity (GB) A21—8GB ($1800) A22—4GB ($1000) R3 Operation duration (h)

A3 Hard disk (type) A31—SSD (3600) A32—SATA (2400) R4 Weight (kg) A4 Body material

(type)

A41—Carbon ﬁber ($3900) A42—Mg/Al alloy ($2400) A43—Common ($1500) R5 Thickness (cm) A5 Screen size (in.) A51—13–14 in. ($3800)

A52—11–12 in. ($2800) R6 Manufacturing cost (in $TWD) A6 Battery capacity (mAH) A61—8000 mAH ($3000) A62—4000 mAH ($2000)

(6)

determine their most preferredﬁve alternatives, then select their most disgusting_{ﬁve ones, and leave the remaining six to the last.}

After distinguishing the business segment from the home segment, conjoint analysis is applied to these two groups to extract their per-ceived importance weights of FAs and associated part-worth utilities for the purpose of concept generation. By observing customers' per-ceived importance weights (similar to customer preferences),Table 6

displays a diverse pattern for the priorities of FAs: A1≻ A4 ≻ A3 is favored by the business segment while A1≻ A5 ≻ A2 is preferred by the home segment. Apparently, the business group concerns more on CPU (A1), hard disk (A3), and body material (A4) than the remaining FAs. In contrast, the home group pays more attention to CPU (A1), RAM capacity (A2), and screen size (A5). Meanwhile, customer utilities of FAs (corresponding to associated levels) are further processed and extracted inTable 7. Very interestingly, the business segment favors smaller screen size (11″–12″) while the home segment prefers larger screen size (13″–14″). Finally, through prioritizing overall customer preference (OCP), the topﬁve design concepts are selected and shown inTable 8(see Eq.(18)).

OCP¼ X i¼1;4 X3 j¼1xijuijþ X i¼2;3;5;6 X2 j¼1xijuij; ð18Þ s:t:X3 j¼1 xij¼ 1; i∈ 1; 4f g; three‐levelFAsð Þ; X2 j¼1 xij¼ 1; i∈ 2; 3; 5; 6f g; two‐levelFAsð Þ:

4.2. Prototype evaluation in terms of market-oriented CRs

In order to derive the weights of CRs, the interdependences be-tween FAs and CRs need to be systematically identiﬁed. Referring toFig. 4, the DEMATEL is applied to invited experts to consult their judgments on the interdependences between CRs and FAs. For in-stance, a simple question is usually performed as follows: How much inﬂuence does FAi(1≤ i ≤ m) impact on CRj(1≤ j ≤ n) in a

5-point rating scale? As shown inTable 9, the interdependences be-tween FAs and CRs are utilized to derive the weights of CRs and they are consecutively incorporated into the TOPSIS ranking. In order to visualize the complicated interrelationships among all fac-tors, Table 10 lists the four main scores of CRs and FAs: active score, passive score, prominence score, and influence score. Hence, a structural diagram could be accordingly portrayed inFig. 5. Appar-ently, all FAs (denoted by the symbol of“square”) are categorized into the“cause” group because of having “positive” influence. Con-versely, all CRs (denoted by the symbol of“diamond”) are classified into the“effect” group due to having “negative” influence.

At the phase of concept generation, recall that only the topfive prior-ities of potential concepts are screened out for speeding up the entire pro-cess. Now, at the phase of prototype evaluation, they are configured into potential prototypes and passed to the testing center for measuring their performances of CRs. By means of the TOPSIS, a market-oriented ap-proach is adopted to assess selected prototypes with respect to the busi-ness segment (seeTable 11) and the home segment (see Table 12), respectively. At afirst glance of the weights of CRs, it is found that an order of R6≻ R2 ≻ R3 presents a pattern for the significant CRs and that means, three CRs including“manufacturing cost”, “response time”, and “operation duration” are mostly concerned in the minds of customers. Thereafter, rather than merely considering engineering-oriented FAs, se-lected prototype are systematically assessed and reprioritized in terms of market-oriented CRs.

Table 4

Illustration of simpliﬁed questionnaires.

Schemes Associated questions Respondents Software Market

segmentation

● How much is your maximally accepted price when purchasing an ultrabook ($30,000 or $24,000 in TWD)?

Consumers Not applicable

CA ● What are your most preferred 5 alternatives (ranking from 1 to 5)? ● What are your most disgusting 5 alternatives (ranking from 12 to 16)? ● What are the intermediate 6 alternatives (ranking from 6 to 11)?

Consumers SPSS

DEMATEL ● How much inﬂuence does FAi

impact on CRj(ranging from 1 to 5)?

Experts MATLAB TOPSIS ● Not Applicable (N/A), through

performing prototype testing.

N/A MATLAB

Table 5

Orthogonal questionnaire design characterized by functional attributes. CPU type Memory capacity Hard disk Body material Screen size Battery capacity

1 i5 8 GB SATA Common 13–14 in. 4000 mAH

2 i7 4 GB SATA Common 11–12 in. 4000 mAH

3 i7 4 GB SATA Carbonﬁber 11–12 in. 8000 mAH

4 i7 8 GB SATA Mg/Al alloy 11–12 in. 8000 mAH

5 i5 4 GB SSD Mg/Al alloy 11–12 in. 4000 mAH

9 i3 8 GB SSD Common 11–12 in. 8000 mAH

10 i7 4 GB SSD Common 13–14 in. 8000 mAH

12 i7 4 GB SSD Carbonﬁber 13–14 in. 4000 mAH

13 i3 8 GB SSD Carbonﬁber 11–12 in. 4000 mAH

16 i5 8 GB SATA Carbonﬁber 13–14 in. 8000 mAH

Table 6

Extracted importance weights of FAs for two distinct segments.

Entire Business Home

A1 CPU type 0.292 0.326 0.266

A2 RAM capacity 0.129 0.076 0.171

A3 Hard disk type 0.161 0.192 0.137

A4 Body material 0.162 0.202 0.131

A5 Screen size 0.113 0.045 0.166

A6 Battery capacity 0.142 0.158 0.129

Count 160 48 62

Table 7

Mean customer utility of FAs with associated levels.

Attributes Speciﬁcations Entire Business Home

CPU type i7 series 1.387 1.853 1.026

i5 series −0.207 −0.346 −0.099 i3 series −1.181 −1.507 −0.928 RAM capacity 8GB 0.525 0.391 0.628 4GB −0.525 −0.391 −0.628 Hard disk SSD 0.717 0.991 0.504 SATA −0.717 −0.991 −0.504

Body material Carbonﬁber 0.558 0.741 0.417

Mg/Al alloy 0.334 0.596 0.131

Common −0.893 −1.338 −0.548

Screen size 13–14 in. 0.244 −0.231 0.611

11–12 in. −0.244 0.231 −0.611

Battery capacity 6000 mAH 0.624 0.816 0.476

(7)

Finally, let us illustrate the optimal combination of FAs (A1–A6) for the identi_{ﬁed two segments. Referring to both}Tables 3 and 8

again, the top winner P1 for the business segment is characterized by“i7-CPU (A11), 8G-RAM (A21), SSD-HD (A31), carbon ﬁber-body material (A41), 11–12 inch screen (A52), and 8000 mAh battery (A61)”. While for the home segment, the top winner becomes P2 which corresponds to “i5-CPU (A12), 8G-RAM (A21), SATA-HD (A32), Mg/Al alloy-body material (A42), 13–14 inch screen (A51), and 8000 mAh battery (A61)”. For each segment, similar explanations can be generalized to all selected prototypes (P1–P5) to characterize their conﬁgurations. Apparently, owing to diverse customer desires and different pricing policies, two distinct segments display their own preference structures through differentiating the priorities of pre-selected prototypes.

5. Concluding remarks and future research

Today, manufacturing companies are inevitably to face the trade-offs between enhancing product varieties and controlling manufacturing costs. Despite that many studies have been presented to address this issue, however, most of them are fully reliant on experts' assessments without tacking customer preferences or customer utilities into ac-count. In order to overcome the above-mentioned shortcoming, this paper presents a hybrid framework which integrates QFD (quality func-tion deployment) with CA (conjoint analysis). In particular, without in-curring tedious pairwise comparisons between product features or

among prototype alternatives, this study contributes to this domain by demonstrating the following merits: (1) concept generation is conducted in a customer-driven way (through CA), (2) the complicated interdependences between FAs and CRs are systematically identified (through DEMATEL), and (3) prototype evaluation is carried out in a market-oriented manner (through TOPSIS). Furthermore, an industrial example regarding configuration varieties of ultrabooks for different segments is demonstrated to justify the validity of our proposed frame-work. In future studies, Kano model, Kansei engineering, or artificial neural networks could be considered and integrated with our frame-work for accommodating other product features (i.e. dichotomous func-tional attributes or esthetic factors) or for predicting customers' dynamic desires.

Acknowledgments

The authors would particularly thank two anonymous referees' helpful comments. This paper isﬁnancially supported by Taiwan Na-tional Science Council under Grant NSC-101-2410-H-009-002. Table 8

Suggested potential prototypes by means of CA.

Business (costb 30,000 in $TWD) Home (costb 24,000 in $TWD)

P1 P2 P3 P4 P5 P1 P2 P3 P4 P5

A1 A11 A11 A11 A11 A11 A12 A12 A13 A13 A12

A2 A21 A21 A21 A21 A22 A21 A21 A21 A21 A21

A3 A31 A31 A31 A31 A31 A32 A32 A31 A32 A32

A4 A41 A42 A41 A42 A41 A42 A42 A43 A43 A43

A5 A52 A52 A51 A51 A52 A51 A52 A51 A51 A51

A6 A61 A61 A61 A61 A61 A62 A61 A62 A61 A62

FA1 . . . . FAn CR1 . . . . CRm FA1

. FAn

n

n correlation matrix n m dependence matrix

CR1

. CRm

n

m zero matrix m m zero matrix

Fig. 4. Input of the direct-relation matrix for the DEMATEL.

Table 9

Identiﬁed interdependences between FAs and CRs through DEMATEL.

R1 R2 R3 R4 R5 R6 A1 0.290 0.258 0.129 0.258 A2 0.194 0.194 0.065 A3 0.161 0.065 0.065 0.129 A4 0.258 0.194 0.226 A5 0.194 0.161 0.194 A6 0.323 0.194 0.065 0.129 Table 10

Visualizing a causal diagram between CRs and FAs via DEMATEL. Active score Ti Passive score Rj Prominence score Ti+ Rj Inﬂuence score Ti− Rj A1 0.935 0.935 0.935 A2 0.452 0.452 0.452 A3 0.419 0.419 0.419 A4 0.677 0.677 0.677 A5 0.548 0.548 0.548 A6 0.710 0.710 0.710 R1 0.484 0.484 −0.484 R2 0.613 0.613 −0.613 R3 0.710 0.710 −0.710 R4 0.677 0.677 −0.677 R5 0.258 0.258 −0.258 R6 1.000 1.000 −1.000

Caudal Effect Diagram

-1.500 -1.000 -0.500 0.000 0.500 1.000 1.500 0.000 0.500 1.000 1.500 Prominence Influence CRs FAs A1 R4 R3 A5 A3 A2 R5 R6 R1 R2 A4 A6

Fig. 5. The cause and effect diagram between CRs and FAs.

Table 11

Prototype evaluation for the business segment. CRs (weights) Business P1 P2 P3 P4 P5 R1 (0.158) 3000 3000 3000 3000 2700 R2 (0.187) 2.50 2.50 2.50 2.50 3.00 R3 (0.165) 6.00 6.00 5.00 5.00 6.00 R4 (0.148) 1.20 1.35 1.30 1.50 1.20 R5 (0.071) 1.50 2.00 1.50 2.00 1.50 R6 (0.272) 28,000 26,500 29,000 27,500 27,200 Similarity 0.963 0.924 0.908 0.880 0.918 Priorities 1 2 4 5 3

(8)

References

[1] Y. Akao, Quality Function Deployment: Integrating Customer Requirements into Product Design, Springer, Cambridge, MA, 1990.

[2] Z. Ayağ, A fuzzy AHP-based simulation approach to concept evaluation in a NPD en-vironment, IIE Transactions 37 (4) (2005) 827–842.

[3] Z. Ayağ, R.G. Özdemir, A hybrid approach to concept selection through fuzzy analytic network process, Computers and Industrial Engineering 56 (2009) 368–379.

[4] G. Büyüközkan, O. Feyzioğlu, A fuzzy-logic-based decision making approach for new product development, International Journal of Production Economics 90 (2004) 27–45.

[5] A. Chaudhuri, M. Bhattacharyya, A combined QFD and integer programming frame-work to determine attribute levels for conjoint study, International Journal of Pro-duction Research 47 (23) (2009) 6633–6649.

[6] L.K. Chan, H.P. Kao, A. Ng, M.L. Wu, Rating the importance of customer needs in quality function deployment by fuzzy and entropy methods, International Journal of Production Research 37 (11) (1999) 2499–2518.

[7] C.H. Chen, L.P. Khoo, W. Yan, PDCS—a product deﬁnition and customization system for product concept development, Expert Systems with Applications 28 (2005) 591–602.

[8] D.R. Cooper, P.S. Schindler, Business Research Methods, 11th ed. McGraw-Hill Irwin, New York, 2006.

[9] E. Fontela, A. Garbus, The DEMATEL Observer: Battelle Institute, Geneva Research Centre, 1976.

[10] F. Fogliatto, J.C. Giovani, D. Siverira, Mass customization: a method for market seg-mentation and choice menu design, International Journal of Production Economics 111 (2008) 606–622.

[11] J.F. Hair, W.C. Black, B.J. Babin, R.E. Anderson, Multivariate Data Analysis: a Global Perspective, 7th edition Pearson, New Jersey, 2010.

[12] C.L. Hwang, K. Yoon, Multiple Criteria Decision Making: Methods and Applications, Springer-Verlag, New York, 1981.

[13] G. Işıklar, G. Büyüközkan, Using a multi-criteria decision making approach to evaluate mobile phone alternatives, Computers Standards & Interfaces 29 (2007) 265–274.

[14] J. Jiao, C.H. Chen, Customer requirement management in product development: a review of research issues, Concurrent Engineering 14 (3) (2006) 173–185. [15] J. Jiao, Y. Zhang, Product portfolio planning with customer-engineering interaction,

IIE Transactions 37 (2005) 801–814.

[16] Z. Kallas, F. Lambarraa, J.M. Gil, A stated preference analysis comparing the analytical hierarchy process versus choice experiments, Food Quality and Preferences 22 (2011) 181–192.

[17] M.C. Lin, C.C. Wang, M.S. Chen, C.A. Chang, Using AHP and TOPSIS approaches in customer-driven product design process, Computers in Industry 59 (2008) 17–31.

[18] E. Liu, S.W. Hsiao, ANP-GP approach for product variety design, International Journal of Advanced Manufacturing Technology 29 (2006) 216–225.

[19] R.D. Luce, J.W. Tukey, Simultaneous conjoint measurement: a new scale type of fun-damental measurement, Journal of Mathematical Psychology 1 (1964) 1–27. [20] I. Palcic, B. Lacic, Analytical hierarchy process as a tool for selecting and evaluating

projects, International Journal of Simulation Model 8 (1) (2009) 16–26. [21] T.L. Saaty, The Analytic Hierarchy Process, McGraw Hill, New York, 1980. [22] Y. Sereli, P. Kauffmann, E. Ozan, Integration of Kano's model into QFD for multiple

product design, IEEE Transactions on Engineering Management 54 (2) (2007) 380–390.

[23] J. Song, T. Jang, S.Y. Sohn, Conjoint analysis for IPTV service, Expert Systems with Applications 36 (2009) 7860–7864.

[24] K. Ulrich, S. Eppinger, Product and Development, McGraw-Hill, New York, 2008. [25] C.H. Wang, J.N. Chen, Using quality function deployment for collaborative product

design and optimal selection of module mix, Computers and Industrial Engineering 63 (4) (2012) 1030–1037.

[26] C.H. Wang, O.Z. Hsueh, A novel approach to incorporate customer preference and perception into product conﬁguration: a case study on smart pads, Computers & Standards Interface 35 (5) (2013) 549–556.

[27] G.S. Wasserman, On how to prioritize design requirements during the QFD planning process, IIE Transactions 25 (3) (1993) 59–65.

[28] W. Yan, C.C. Chen, M.D. Shieh, Product concept generation and selection using sorting technique and fuzzy C-means algorithm, Computers and Industrial Engi-neering 50 (2006) 273–285.

[29] Ultrabook Development Trends, DigiTimes, 2012, (http://www.digitimes.com/ supply_chain_window/story.asp?datepublish=2012/7/9&pages=VL&seq=201). [30] the website of Wikipedia,http://zh.wikipedia.org/wiki/Ultrabook, (in Chinese).

Chih-Hsuan Wang is an assistant professor in the Depart-ment of Industrial Engineering & ManageDepart-ment, National Chiao Tung University (NCTU), Taiwan. Prior to joining NCTU, he has been a faculty member in the Department of Marketing, National Chung Hsing University and an adjunct assistant professor at the National Taiwan University. During 2003 to 2005, he has been a research scholar at the University of Tennessee and Texas A&M University, respec-tively. He has published several SCI papers in IIE Transactions, IJPR, C&IE, CS&I, JIM, and ESWA. His research interests include product development, operation management, service sci-ence, and business intelligence. Since 2006, he also served a session chair for the IEEM and APIEMS international confer-ences.

Chih-Wen Shih is currently a PhD candidate at the Depart-ment of Industrial Engineering and ManageDepart-ment, National Chiao Tung University, Taiwan. His research interests are human–computer interaction, ambient intelligence, service science, business intelligence and supply chain management, etc. For the past 5 years he has been working most of the time in the ICT enabled applications and services. Currently, he is the project manager at Institute for Information Industry in Taiwan.

Table 12

Prototype evaluation for the home segment. CRs (weights) Home P1 P2 P3 P4 P5 R1 (0.168) 2500 2500 2300 2100 2500 R2 (0.189) 4.0 4.0 3.5 5.0 4.0 R3 (0.179) 5.0 6.8 6.5 7.0 5.0 R4 (0.144) 1.6 1.5 1.8 1.8 1.8 R5 (0.051) 2.0 2.0 2.2 2.2 2.2 R6 (0.268) 23,900 23,900 23,300 23,100 23,000 Similarity 0.868 0.939 0.921 0.848 0.859 Priorities 3 1 2 5 4