You are on page 1of 16

Information Sciences 177 (2007) 1567–1582

www.elsevier.com/locate/ins

A two phase multi-attribute decision-making approach for
new product introduction
Cengiz Kahraman a, Gu¨lc¸in Bu¨yu¨ko¨zkan

b,*

, Nu¨fer Yasin Atesß

a

a

b

Istanbul Technical University, Department of Industrial Engineering, Macka 34367, Istanbul, Turkey
Galatasaray University, Department of Industrial Engineering, Ciragan Caddesi No. 36, Ortako¨y 34357, Istanbul, Turkey
Received 6 September 2006; accepted 7 September 2006

Abstract
This study aims at improving the quality and effectiveness of decision-making in new product introduction. New product development has long been recognized as one of the corporate core functions to be competitive on an increasingly competitive global market. However, developing new products is a process involving risk and uncertainty. In order to solve this
stochastic problem, companies need to evaluate their new product initiatives carefully and make accurate decisions. For
this reason, a systematic decision process for selecting more rational new product ideas is proposed. Basically, two stages of
decision-making are described: the identification of nondominated new product candidates and the selection of the best
new product idea. These stages are composed of an integrated approach based on a fuzzy heuristic multi-attribute utility
method and a hierarchical fuzzy TOPSIS method. Finally, an application is given to demonstrate the potential of the
methodology. 
2006 Published by Elsevier Inc.
Keywords: New product idea selection; Fuzzy sets; Heuristics; Multi-attribute utility function; TOPSIS method

1. Introduction
In an increasingly competitive global market, companies must be better at developing new products. New
product development (NPD) has long been recognized as one of the corporate core functions. Particularly for
the companies with short product life cycles, it is important to develop new products and new product platforms that fulfill reasonable demands on quality, performance, and cost quickly and safely.
NPD which translates an idea into a tangible physical asset is structured around well-defined phases.
Each phase encloses many decision points, where management decides about the future of the project
[26]. One of the most critical decisions when managing NPD projects is the new product idea selection
[1,4]. Several researchers conclude that it is difficult for managers to end the NPD projects once they are

*

Corresponding author. Tel.: +90 212 227 4480; fax: +90 212 259 5557.
E-mail address: gbuyukozkan@gsu.edu.tr (G. Bu¨yu¨ko¨zkan).

0020-0255/$ - see front matter  2006 Published by Elsevier Inc.
doi:10.1016/j.ins.2006.09.008

configuration. 3.23. and cultural environment situation. Kahraman et al.22. Analytical hierarchy process [43].28. / Information Sciences 177 (2007) 1567–1582 begun [12. The next section presents an overview of the decision-making techniques used in a NPD process. However. and commercial issues. the last section contains some concluding remarks and possible perspectives. related to their economic.38. selected. NPD can be defined as a process including many ‘‘generic decision’’ points. In this study. and Fuzzy logic [19].33]. This is the main motivation of this study. Some of them may be classified in that way: Probabilistic models [37]. The decision-makers are actively involved in the decision process of the proposed approach. social. To help consumers and decision-makers easily express their judgments. and project schedule.13–15.19. and the decision rules are constructed through serious discussions between the decision-makers and the analysts. An application of the approach is described in Section 4 through a case study. For example. the problem of NPD could be adequately modeled by the maximization of single-objective models. Although different organizations can make different choices and may use different methods. NPD decisions contain considerable amount of uncertainty causing elements. it is critical to use a structured approach that can minimize the risks of NPD projects. In the following. Behavioral approaches [37]. EUT has been criticized for limiting its application to a single attribute: the pay-off (or wealth).11. this two phase approach is presented. we propose an integrated approach with two phases based on a fuzzy heuristic multi-attribute utility method and the hierarchical fuzzy TOPSIS method to make more rational selection decisions. Similar to all decision problems. management.4. Options pricing theory [18]. all of them make decisions about a collection of issues such as the product concept. procurement. etc. where each of them must be evaluated. Many authors have shown the complexity of the NPD decision-making process through empirical studies that have demonstrated that they consider more than one attribute in their utility functions [3. Finally.36.1568 C. tactic. The details of the proposed integrated approach are given in Section 3. These techniques can be used exclusively or in a hybrid way.35]. A two phase approach for selection among new product ideas 3.1. usually conflicting ones. 1. The second phase involves a more detailed analysis to select the best one among the remaining nondominated alternatives. Real-life observations refute this simplification. Scoring models and checklists [20. distribution arrangements. While considering the decision points in the whole NPD process. architecture. we propose a two phase integrated decision-making approach based on fuzzy heuristic multi-attribute utility and hierarchical fuzzy TOPSIS methods as given in Fig.39]. numerous decision tools and techniques have been developed to assist managers in making better screening decisions in an uncertain environment [10. heuristic information is obtained through a question and answer system. The first phase of our approach evaluates all the alternatives to eliminate the alternatives clearly dominated by the others. in addition to the expected profit.33]. which confuse the decision-maker to reach the targeted performance [2. Consequently. and prioritized. Expected utility theory (EUT) was a first step in the direction of broadening the profit maximizer assumption and including higher moments of the expected profit. the linguistic variables should be used to evaluate the importance of customer needs and ratings of new product alternatives. An overview for new product decision methods NPD has a vast working area and it addresses different strategic. According to this principle. and operational managerial levels in the organization. which focuses on making an efficient new product idea selection. Sensitivity analysis [34]. . The fuzzy set theory is a useful tool for this aim.32]. The rest of the paper is organized as follows. Uncertainty arises from both internal and external multiple sources including technical. We also present an industrial application to demonstrate the potential of this integrated approach. where a dialogue between the decision-makers and the analysts takes place. Thus. In this article. Heuristic multi-attribute utility function approach One of the basic principles of the classical economic theory is that decision-makers behave as profit maximizers. 2. All the related studies suggest that NPD decision-making processes are driven by various criteria.

List all alternatives to be evaluated. Identify the new product evaluation attributes and establish a universe of discourse for each of them. where U is the Multi-attribute utility function (MAUF) and rj is the attribute j. we resort to Multi-attribute utility theory (MAUT). construct heuristic decision rules for each group. 7. List the selected most appropriate alternatives and measure along the attributes. Identify the interest groups of people in the decision environment 2. Obtain the scores for selected alternatives in phase I with respect to each sub evaluation criterion. 8. The proposed approach. 5. Identify the weights of main and sub evaluation criteria with respect to goal and main criteria respectively. Fig. 3. 1. Interact with decision makers to identify their heuristics and according to them. the utilities of n attributes from different alternatives are captured in a quantitative way via a utility function represented mathematically as U = U(r1. . Thus. Thus. r2. . The aim of MAUT is to reduce a decision problem with multiple criteria to a cardinal function that ranks alternatives according to a single criterion. PHASE II 6. decisions under MAUT are made by maximizing U and responding to the set of objectives that are simultaneously aimed at by the decisionmaker. PHASE I 4. Rank the alternatives from the best to the worst and select the best new product idea for the company. Kahraman et al. . The level of achievement of each attribute can usually be expressed mathematically as a function of the decision variables. Recognizing the convenience of including several objectives to simulate evaluator or decision-maker behavior. an approach largely developed by Keeney and Raiffa [25]. in order to overcome the limitations of the single-attribute utility function.C. rn). . / Information Sciences 177 (2007) 1567–1582 1569 INITIALIZATION PHASE 1. 9. . Use the Hierarchical FuzzyTOPSIS Algorithm to obtain the relative closeness to ideals.

xm1 xij . This study is based on the Chen and Hwang’s [6] fuzzy TOPSIS method since the other fuzzy TOPSIS methods are derived from this method with minor differences. dij) shown in Fig. 7 . . . of m · n dimension is defined as in Eq.. xmj  Xn 3    x1n 7 .. ~ j ¼ ðaj .2. vj . a decision matrix. A1 2 X1  Xj x11    x1j 6 6 .. In addition. the steps of fuzzy TOPSIS developed by Chen and Hwang [6] are given. D. Table 1 gives the comparison of these fuzzy TOPSIS methods.2. xj is a cost attribute j =xij . The problem is solved using the following steps. First.1570 C. Kahraman et al.  Am  . Efstathiou and Rajkovic [16] and Efstathiou [17] argued that the MAUF could not be practically obtained by the combination of single-attribute utility functions because of the dependency among attributes. (1).. 6. If xij is fuzzy. The decision matrix must first be normalized so that the elements are unit-free. . which is therefore relatively cheap and available to any researcher.. 6 4. 7 . (2). this is a software-based strategy. it is represented by a trapezoidal number as xij = (aij. Therefore a heuristic approach is needed to define the MAUF [6]. . . THEN . The comparison includes the computational differences among the methods. 7 7 7    xin 7 7 7 . 6 .. Then the method chooses an alternative with the maximum similarity to the ideal solutions [41]..1. . The method is based on the concept that the chosen alternative should have the shortest distance from the positive-ideal solution and the longest distance from the negative-ideal solution. we use linear scale transformation as follows: ( xij =xj 8j . Since decision data may be numerically and/or linguistically expressed. cij. The fuzzy weights can be described by Eq. xj is a benefit attribute rij ¼ ð3Þ x 8j . dj Þ w ð2Þ Algorithm. some fuzzy TOPSIS methods were developed in the literature. 3. The utility function is represented then in the ‘‘IF . Step 1. . bij. the fuzzy set theory must be incorporated in this heuristic approach. Hierarchical fuzzy TOPSIS methodology 3. 2. In the last 15 years. . It was developed by Hwang and Yoon [21]. but heuristic methods are commonly used because they are much faster than exact algorithms. TOPSIS defines an index called similarity (or relative closeness) to the positive-ideal solution and the remoteness from the negative-ideal solution. . To avoid the complicated normalization formula used in classical TOPSIS. / Information Sciences 177 (2007) 1567–1582 A heuristic method is an algorithm that gives only an approximate solution to a given problem.’’ decision rule format. Sometimes we are not able to formally prove that this solution actually solves the problem. In the following. 7 5 ð1Þ    xmn where xij " i. j may be crisp or fuzzy. Normalize the decision matrix. 6 6 D ¼ Ai 6 6 xi1 6 . . bj . Fuzzy TOPSIS TOPSIS (Technique for order preference by similarity to ideal solution) views a multi-attribute decisionmaking problem with m alternatives as a geometric system with m points in the n-dimensional space.

.0 0. d ij Þ. (3). 1. when xij is fuzzy. bj . Trapezoidal fuzzy numbers. Kahraman et al. This matrix is obtained using vij ¼ rij wj 8j .2 0. 6 6 6 ri1    6 6.3 0.. Obtain the weighted normalized decision matrix.1 0 wj xij αj aij bij Linear normalization Modified Manhattan distance Vector normalization Linear normalization Manhattan distance βj cij χj δj d ij Fig. x j ¼         ða j . its corresponding rij must be fuzzy. respectively. we have   8 a b c d > < xij ðÞxj ¼ dijj . 6. (4). 2. Let xij ¼ ðaij . 7 7 . . 4.. Eq. A1 . rij . [40] Chu and Lin [9] Zhang and Lu [44] Triangular Fuzzy numbers Crisp values μ (x )μ 1. 1) and (0. 7 7 7    rin 7 7 . we can rewrite the decision matrix in Eq. .. .7 0. D0 ¼ A i . its corresponding rij must be crisp. cj . bij .5 0.C. j ð6Þ .8 0. 0. bj . . 1) and (0. cij . d i j d ij cij bij aij Step 2. 1. cj . respectively Chen [5] Chu [8] Tsaur et al.. 0). d j Þ and xj ¼ ðaj .9 0. / Information Sciences 177 (2007) 1567–1582 1571 Table 1 Comparison of fuzzy TOPSIS methods Source Attribute weights Type of fuzzy numbers Ranking method Normalization method Chen and Hwang [6] Liang [29] Fuzzy numbers Fuzzy numbers Fuzzy numbers Fuzzy numbers Crisp values Trapezoidal Lee and Li’s [27] generalized mean method Trapezoidal Chen’s [7] ranking with maximizing set and minimizing set Linear normalization Manhattan distance Triangular Triangular Chen [5] assumes the fuzzy positive and negative ideal solutions as (1. respectively Liou and Wang’s [30] ranking method of total integral value with a = 1/2 Zhao and Govind’s [45] center of area method Triangular Kaufmann and Gupta’s [24] mean of the removals method Triangular Chen’s [5] fuzzy positive and negative ideal solutions: as (1.. 7 .6 0. bijj .. 6 . d j Þ. In the formula above xj and x j represent the largest and the lowest scores. aij j j rij ¼ ð5Þ      > : x ðÞxij ¼ ai . (3) is then replaced by the following fuzzy operations. ci . bi . 5 ð4Þ    rmn When xij is crisp. 2 X1  r11    6. 0). 0. By applying Eq.4 0. cij . (1) as in Eq. Am rm1  Xj r1j . rmj  Xn 3    r1n .

respectively. and the negative ideal solution (NIS).. bj . Obtain the separation measures S i and S  i .. Eq. S i ¼ n X D ij . ð10Þ  A ¼ ½v 1 .. . In case   of fuzzy data. 5 vmj  ð9Þ vmn Step 3. vn . . vj and vj may be obtained through some ranking procedures. 6 . (8) is used when the jth attribute is a cost attribute. (17). Am vm1    Xj v1j . . . In the classical case. separation measures are defined as n X Dij . . . j ð17Þ . The vj and v j are the fuzzy numbers with the largest generalized mean and the smallest generalized mean. the difference between two fuzzy numbers lvij ðxÞ and lvj ðxÞ (based on Zadeh’s study [42]) is explained as given in Eq. m ð14Þ j¼1 For crisp data. 4 . 2 X1  A1 v11    . vj . 7 7    vin 7 7 . / Information Sciences 177 (2007) 1567–1582 When both rij and wij are crisp. vn .  dj ð7Þ d j cj bj aj    ai b c d i i i aj . . m ð13Þ S i ¼ j¼1 and.  bj . . . Chen and Hwang [6] use Lee and Li’s ranking method [27] for comparison of fuzzy numbers. (7) and (8) can be summarized as in Eq. Step 4. (9). i ¼ 1. 7 7 . . we find a vijwhose greatest mean is vj and lowest mean is v j . 7 7 .  Xn 3    v1n .. A*. . vj and vj are obtained straightforwardly. Kahraman et al. Obtain the positive ideal solution (PIS).. . For crisp data. dj ð8Þ vij ¼ rij ðÞwj ¼ d ij cij bij aij Eq. . 6 v ¼ Ai 6 6 vi1    . 6 6.. i ¼ 1. PIS and NIS are defined as A ¼ ½v1 . . (6) may be replaced by the following fuzzy operations: ! aij bij cij d ij vij ¼ rij ðÞwj ¼ aj .  vj . Eq. . The result of Eqs. The generalized mean for fuzzy number vij 8j is defined as Mðvij Þ ¼ a2ij  b2ij þ c2ij þ d 2ij  aij bij þ cij d ij ½3ðaij  bij þ cij þ d ij Þ ð12Þ For each column j. For fuzzy data. . . while when either rij or wij (or both) are fuzzy. vij . 6 6. ð11Þ   where vj ¼ maxi vij and v j ¼ mini vij .. the difference measures Dij and D ij are given as Dij ¼ jvij  vj j ð15Þ  D ij ¼ jvij  vj j ð16Þ The computation is straight forward.1572 C. . vij is crisp. . (7) is used when the jth attribute is a benefit attribute. ^ Dij ¼ 1  fsupblvij ðxÞ lvj ðxÞcg ¼ 1  Lij x 8i. ..

3. / Information Sciences 177 (2007) 1567–1582 μ(x) vj* vij 1. and it has less number of questions to consider. SAn2 . its algorithm should be developed. Step 5. and also provides a well expression of the decision problem to the decision-maker with its hierarchical model. 4 will be considered. Hierarchical TOPSIS is easy to implement. While developing it.. 4.. SA2r2 ... On the other hand. SApl MAn . D ij are crisp numbers. Meanwhile the well known and widely used multi-attribute decision-making method. This index is used to combine S i and S  i indices calculated in Step 4.C. Those methods evaluate the alternatives with respect to only main attributes with a single level. 1573 Lij x Fig. and its calculations are very tedious. Compute the relative closeness to ideals. SAnr n . j ð18Þ x Note that both Dij .2. Hierarchical fuzzy TOPSIS The classical TOPSIS method does not consider a hierarchical structure between main attributes and subattributes as it is in the additive weighting and weighted product methods. The hierarchical TOPSIS is developed to benefit both from the superiority of the hierarchical structure of AHP and easiness of implementation of TOPSIS. where Lij is the highest degree of similarity of vij and vj . The value of Lij is depicted in Fig. considers a hierarchical model which gives the ability of taking into consideration more information and provides superiority to AHP among many other multi-attribute decision-making methods. The hierarchy considered in fuzzy TOPSIS algorithm. The common property of these methods is their easiness of implementation. 3. Since this study suggests an integrated approach based on a hierarchical fuzzy TOPSIS method. GOAL MA1 SA11 SA12 . The derivation of Lij.... this method consumes a lot of time at the implementation phase because of many questions are asked for comparisons among alternatives and attributes.. the difference between lvij ðxÞ and lvj ðxÞ is defined as ^ D ðxÞcg ¼ 1  Lij ij ¼ 1  fsupblvij ðxÞ lv j 8i. Also its calculations are less tedious and faster than AHP. 3. AHP. SAn1 Fig. Kahraman et al. they can be combined: i   Ci ¼ S i =ðS i þ S i Þ ð19Þ The alternatives are ranked in descending order of the Ci index.. MA2 SA1r1 SA21 SA22 MAP . Since S i and S  are crisp numbers.2. the hierarchy in Fig.. Similarly.

(22). 4 0 0  0  ~n w    MA3 n  0 7  0 7 7 7 . Kahraman et al. 6 6 6w ~  0 6 1r1 0 6 60 ~ 21    0 w 6 6 ~ 22    0 w 60 6 6. and s respondents. 7 . 6 6 60 0  0 6 6 60 0  0 6 6 6 . Each main attriPn bute has ri sub-attributes where the total number of sub-attributes m is equal to i¼1 ri . 7 7  0 7 7 7 7 0 7 7 7 ~ n1 7  w 7 7 ~ n2 7  w 7 7 7 . 7 . The second matrix ðeI SA Þ represents the weights of the sub-attributes with respect to the main attributes. 7 7  0 7 7 7 7 ... The weights vector obtained from eI MA are written above this eI SA as illustrated in Eq. ~1 w SA11 SA12 . .. . SApl . 5 ~ nrn  w ð22Þ ~ pl is the arithmetic mean of the weights assigned by the respondents and it is calculated by Eq. 6 . 6 . . . is constructed from the weights of the main attributes with respect to the goal.. k alternatives. . (20). .. m sub-attributes... . . . 6. eI SA ¼ SA2r2 .1574 C... . 6.. . . ð20Þ MAn ~ p is the arithmetic mean of the weights assigned by the respondents and is calculated by Eq. 6 . . 6.. . n ð21Þ w s where ~ qpi denotes the fuzzy evaluation score of pth main attribute with respect to goal assessed by the ith respondent... (23). 2. p ¼ 1. 6 6 60 ~ 2r2    0 w 6 6 6... 7 7  0 7 7 7  0 7 7 7  0 7 7 7 .. (21): where w Ps ~ qpi ~ p ¼ i¼1 .. . . / Information Sciences 177 (2007) 1567–1582 Assume that we have n main attributes. SAnrn ~2 w  ~p w 2MA1 MA2    MAp ~ 11 0  0 w 6 6w  0 6 ~ 12 0 6 . where w Ps ~ qpli ~ pl ¼ i¼1 ð23Þ w s where ~ qpli is the weight of lth sub-attribute with respect to pth main attribute assessed by the ith respondent. .. eI MA MA1 MA2 2Goal3 ~1 w .. The first matrix ðeI MA Þ. ~2 7 6w 7 ¼ 6 7 MAp 4 w ~p 5 . . . SAn1 SAn2 . . 7 7 . ~n w . given by Eq. SA1r1 SA21 SA22 . . 0 6 6 60 ~ pl 0  w 6 6 6. .... . 7 . 6 . .

.. . / Information Sciences 177 (2007) 1567–1582 1575 The third matrix ðeI A Þ is formed by the scores of the alternatives with respect to the sub-attributes.. After obtaining ~cqpl s. 0. 6 . 0.. ~ pl w . the hierarchical structure is ready to be included to the fuzzy TOPSIS algorithm described in Section 3. . . ~c2nrn 7 7 7 . . ~ nrn w .3. . 7 7 . . ~ck1r1 A1 . 6 .. . .2. . . . . ~c11r1 6 A2 6 ~c211 ~c212 . the choice of triangular fuzzy numbers is made for the sake of simplicity and understandability. 6 . 50. 0. 4 Ak ~ck11 ~ck12 . .6. . ~cq1r1 6 . . . SApl . Table 2 The importance degrees Very low Low Medium High Very high (0. . The other types may increase the computational complexity without substantially affecting the significance of the results.. . . ~cqpl is the arithmetic mean of the scores assigned by the respondents and it is calculated by Eq. ... (27) Ps ~ qqpli ~cqpl ¼ i¼1 ð27Þ s where ~ qqpli is the fuzzy evaluation score of qth alternative with respect to lth sub-attribute under pth main attribute assessed by ith respondent.2. 0. SAnr3n . ~cknrn ð24Þ where e pl ¼ W n X ð25Þ ~ pw ~ pj w j¼1 Since wpj = 0 for j 5 l. we can use Eq. (24).C. . . ~c21r1 6 . 1) Table 3 The scores Very poor Poor Fair Good Very good (0..2) (0. 100) . 80. 7 . ~ckpl . 6 . . 0.. 40) (30. ~c2pl .. ~cqpl . 0. .4) (0. 20. .. . Table 2 will be used to determine the importance degree of each main attribute with respect to the goal and each sub-attribute with respect to the main attributes. SA1r1 ~c111 ~c112 . 1) (0. (25) e pl ¼ w ~ pw ~ pl W ð26Þ In eI A . (26) instead of Eq. .. . . . ~ 11 w ~ 12 w ~ 1r1 .5.. . 20) (0.8. 70) (60. . The linguistic terms represented by triangular fuzzy numbers for scoring the alternatives are given in Table 3. Among the types of fuzzy numbers.1. . ~c1pl . ~cqnrn 7 7 7 . 6 eI A ¼ 6 Aq 6 ~cq11 ~cq12 . ~c1nrn 7 . . 7 . . . 0. .8. The weights vector obtained from eI SA are written above this eI A as in Eq. 100) (80.. 5 . .7) (0.. 100. . w SA 2 11 SA12 . 1. Kahraman et al. .. 0. ..

These factors are identified from the review of the related literature and from the interviews with some domain experts. A simple feasible starting point for evaluating any investment is to consider the investment’s revenues and risks [31]. . medium (M). Evaluation environment and attributes The proposed framework was applied to the one of the largest automotive manufacturers in Turkey. 5. benefits and strategic impact. which may establish intangible profits for the company depending on the creativity and innovative features of investments [1]. we obtained the decision rules. 4. A scale of 0–100 is used for risk and strategic impact while a monetary scale of $0–$500. medium high (MH). one from financial department and a top-level manager came together to evaluate 14 NPIs.1. four from marketing department. Application of the proposed framework 4.000 is used for benefits.2. Fig. three from design and engineering department. Hierarchical structure of decision criteria in NPI evaluation. NEW PRODUCT IDEA EVALUATION Benefits Strategic Impact Profitability Strategic value Efficiency Business impact Risks Systematic risks Unsystematic risks Financial Managerial Technical Personnel Fig. very low to low (VLL). the objective is to evaluate potential benefits and risks associated with NPIs in a detailed way. These ideas are submitted from different employees and customers between July 2004 and December 2004. Fig. 6 gives the membership functions for the aspects risk and strategic impact. As an example. For this reason. In the second phase. / Information Sciences 177 (2007) 1567–1582 4.1576 C. 5. Heuristic MAUF approach The aspects considered in NPI evaluation are risk. Each of these aspects is treated as a fuzzy set. Through a careful discussion. However. The conversion scale consists of the linguistic variables very low (VL). and very high (VH). The hierarchical structure of decision attributes in NPI selection is shown in Fig. Kahraman et al. we have chosen three main attributes for the first phase of the study: NPI related risks. high to very high (HVH). low (L). high (H). one of the decision rules for the utility function of ‘‘Very Good’’ as follows: – If the benefit of alternative X is ‘‘very high’’ and its strategic impact is ‘‘very high’’ and its risk is ‘‘very low’’ then the utility value of the alternative X is ‘‘very good’’. NPD investments have also a special characteristic. and strategic impact. medium low (ML). bounded to a predetermined interval. 7 illustrates the membership function for the aspect benefit. benefit. and characterized by a single possibility distribution. A group of nine decision-makers.

After identifying all decision rules. 4. μ (x ) VL VLL L ML M 150 200 250 MH H HVH 300 350 400 VH 1. Kahraman et al.000) Fig. we listed all alternatives and evaluated them according to the determined rules. 6. 7. The membership function for the evaluation aspect risk and strategic impact. Five alternatives (NPIs 1. 9. Membership function for the evaluation aspect benefit.0 0 50 100 450 500 x (× $1. / Information Sciences 177 (2007) 1567–1582 1577 μ (x) VL VLL L ML M MH H HVH 60 70 80 VH 1. which have the highest utility values.0 x 0 10 20 30 40 50 90 100 Fig. Then we obtained the utility values of each alternative as given in Table 4. were identified for a detailed analysis. 11 and 14). Table 4 The linguistic evaluation of the alternatives with respect to each criterion Alternatives Benefits Strategic impact Risks Utility NPI-1 NPI-2 NPI-3 NPI-4 NPI-5 NPI-6 NPI-7 NPI-8 NPI-9 NPI-10 NPI-11 NPI-12 NPI-13 NPI-14 High to very high Low High to very high Very high High Medium Medium–high High Very high High to very high Very high Medium–high Medium Very high Very high Very low to low Medium–good High High Medium–high Medium Very high High High Very high Medium Low High to very high Low Medium Very low to low Low Medium–low Low Medium–low High Very low to low Low Medium–low Medium Medium–high Low Very good Medium–poor Good Very good Good Medium–good Medium Good Very good Good Very good Medium–good Medium–poor Very good .C.

00.32. 0. 0. 0. 0.00.42.00.58. 5.87) (0. daij j j j rij ¼ ð28Þ     ai bi d i > : x j ðþÞxij ¼ d ij . b. 0. 0.00. 0. Taking into consideration the hierarchy given in Fig. 8 cij  a  > > < 1  b þ c  a  b for bij < b ij ij 8i. The evaluations from all nine respondents were received and eI MA . Kahraman et al. 0. 0. the equations based on trapezoidal fuzzy numbers in the fuzzy TOPSIS algorithm given in Section 3.00. c Þ and v j ¼ ða . Then. respectively. (30) and (31).00.86.00) (0.47. four sub-sub-attributes and five alternatives.00) (0.00. respectively.67.97) . 0.1 are rewritten for TFNs.00. b. 0.00. 0.00) (0. 0.00. 0. 0. 0.00. 0. 0. 0.00. it can be easily seen that Eq.62. j cij  a >  > for b < b :1   ij b þ cij  a  bij ð30Þ ð31Þ    where vj ¼ ða .97) (0. 0. b . Prepared questionnaires were applied to the group of decision-makers. The model considered in the study has three main attributes. / Information Sciences 177 (2007) 1567–1582 4.00) (0. 0. Hierarchical fuzzy TOPSIS method In the second step.00) (0. 0.00) (0. (12) is reduced to Eq.1578 C. Eq. 0.00.63.66. Since a triangular fuzzy number (a.87) (0. a detailed questionnaire for fuzzy TOPSIS was prepared to receive the individual weights of main. 0.3. 0. 0. 0. 0.73.83) (0. 0.00) (0.53. bbij . the identified five NPIs by heuristic MAUF approach were evaluated using the hierarchical fuzzy TOPSIS. Since we preferred using triangular fuzzy numbers (TFNs).00) (0. c). c) can be represented in a trapezoidal form like (a.00. 0.00. 0. 0.00.00.00) (0. sub and sub-sub-attributes. (29) Mðvij Þ ¼ a2ij þ d 2ij  aij bij þ bij d ij ½3ðaij þ d ij Þ ð29Þ Dij and D ij are calculated by Eqs. (5) can be expressed as follows: 8   > < xij ðþÞxj ¼ adij .78. c Þ are the fuzzy numbers with the largest generalized mean and the smallest generalized mean. bij .00. the steps of the hierarchical fuzzy TOPSIS algorithm are executed.43.00) (0.80) (0. 0. b . aij Then. 0.80) (0.00.63. 0. eI SA and eI A were obtained as in Tables 5–8.00.61.00. 0. six sub-attributes.00. 0. b. 0.00. j Dij ¼ c  aij >  > :1  for b < b ij bij þ c  aij  b 8  c  aij  > > < 1  b þ c  a  b for b < bij ij ij  Dij ¼ 8i. Table 5 eI MA Goal Benefits Strategic impact Risks (0.53.83.83) (0.93) Table 6 eI SA Profitability Efficiency Strategic value Business impact Systematic risks Unsystematic risks Benefits Strategic impact Risks (0.00) (0. 0. 0.73. 0.00) (0. 0.00.

87) (0. 60.00) (0. 70. 94) (59. 78.020 0. 72) (42. 79) (36. 94) (58. 62. 11. The results of the sensitivity analysis are given in Table 10 and the graphical representation of these results is illustrated in Fig. 78) Table 9 Distances from the ideal solution NPI-1 NPI-4 NPI-9 NPI-11 NPI-14 S i S i Ci Normalized Ci 0. Four states in total are examined: One of the main attributes has the highest . 75) (49.5.00. 9. NPI-4.56.00) (0.298 1.87) (0. 63. 63. The obtained results indicate that NPI-1. 88. 83. 0.913 0. 48.00) (0. 88) (32. A ranking among the reduced set of alternatives is the result of the second phase. 85.754 0.49. 4.441 0. and 14 were to be considered in the second phase for a detailed evaluation by the fuzzy hierarchical TOPSIS approach. 79) (52. 67. 96) (45. 94) (62.49.774 0. 0.00. Results From a set of 14 NPIs. a sensitivity analysis is made. Although NPI-9 is selected in the first phase. The reduced set of alternatives selected with respect to the evaluations is analyzed by the hierarchy developed in the initialization phase. 4. 47) (33. 0. 86) (48. 90. 76. 66) (42. 92) (63.00. 75. 64.4.546 0. 95) (62.69. it is not eligible to invest it because of its relatively low score. The first phase determined that NPIs 1. 73) (32. 82) (43. 81) (47. 75) (50. 59. 73) Personnel risks (43. and NPI-14 are the first NPIs to invest.69.00. 73) (49. 61. 73) Managerial risks (51. 0. 75) Technical risks (48. 69. 0. 91) (32. 73) (24. 73) (69. 54.36. 63) (46. 73) NPI-1 NPI-4 NPI-9 NPI-11 NPI-14 Financial risks (51. 56) (49. Sensitivity analysis To analyze the attitudes of the alternatives under different weights of main attributes. 61.282 0.165 0.024 0. 62. 64. 8. 42. 0.278 0. 68. 48. 58.290 0. 0. 0. 58. 4. 83) (61. 75) (50.00.C. first a rough evaluation is made with heuristic MAUF approach using the rules defined by decision-makers.255 Table 9 shows the distances from the ideal solution for each NPI and the normalized values making the results easy to interpret.007 0.952 0. 66. 97) (60. 0.00.00) (0. 82) (55.886 0.39. 78) (55. 48.414 0. 0. 61. 66) (39. 66) (49. 87. 0. 78) (55. 0. 56. 31. 69. The results in Table 9 indicate that NPI-1 achieves the highest score whereas NPI-9 has the lowest. 0.59.213 0. / Information Sciences 177 (2007) 1567–1582 1579 Table 7 eI SSA Financial Technical Managerial Personnel Systematic risks Unsystematic risks (0. 0.00. 60. 0. 80. 78. 74.691 0.73) Table 8 eI A Proficiency Efficiency Strategic impact Business value NPI-1 NPI-4 NPI-9 NPI-11 NPI-14 (44. 66. 78) (47.77) (0.00. 67. 88) (56. Kahraman et al. 69) (55.681 0. 0.

It is clear that not only the importance of sub and sub-sub-attributes but also the importance of main attributes are very influential to the selection process.334 0. For each state.20) (0.80.00.021 0.80.70) SI R (0. 0. / Information Sciences 177 (2007) 1567–1582 Table 10 Sensitivity analysis B (0.50. 0.246 0. normalized relative closeness to ideals (Ci) are computed. possible weight whereas the other two have the minimum (this creates three states).412 0. 1.045 0. 1. 0.20) (0.20) (0. The sensitivity application shows us prioritization of NPI-1. 0. 400 0. 000 4 3 NPI-1 NPI-4 2 NPI-9 1 NPI-11 es at St NPI-14 Fig. and risks. Kahraman et al.00) (0. In order to reduce these risks and uncertainties. 0. developing new products is a risky process involving uncertainty.154 0.266 0.091 0.00. It is clear that the selection among NPIs is a difficult and sensitive issue which has quantitative and qualitative aspects.20) (0. The approach is based on the premise that the selection among NPIs should be viewed as a product of benefits.049 0.00) (0.233 0.236 0.20) (0. 0.70) Normalized Ci (0.00. 1. complexity. 0. On the other hand. 1. strategic impact. 0. an integrated approach based on fuzzy heuristic MAUF and hierarchical fuzzy TOPSIS method is proposed to improve the quality and effectiveness of decision-making in new product introduction. 200 0.50. and 14 are effected by decision-makers’ perceptions about main attributes. 0.30. 0. The developed two phase multi-attribute decision-making .70) States NPI-1 NPI-4 NPI-9 NPI-11 NPI-14 1 2 3 4 0. and imprecision. In this paper.262 0. the success of a product highly depends on the efficiency and effectiveness of the NPD process.00. 0.109 0.80.00.20) (0.30. 0. A particularly important key to the adequacy of a NPD process is the way in which it is capable of handling uncertainty.00.00. 300 0.00) (0. In State 1 where the benefit aspect has the largest weight. 0.300 0. 1. Sensitivity analysis. 100 0. 0.156 0. The first phase of the evaluation consists of judgments of decision-makers with respect to these three main attributes. 0.00.139 0. 8.243 Normalized Ci 0.278 0. companies need to evaluate their new product initiatives carefully and make accurate decisions.00.257 0.00. 1. 4. Conclusion In today’s fast and dynamic markets for innovative products.1580 C. NPI-11 performs close to NPI-14. whereas the second phase considers a detailed hierarchy of attributes and alternatives.50. 0.00. 500 0. 0.00.30. The fourth state is considered for the equal mid-value weights.00.169 0.00. 5. 0.00.

Krishnan. Feyziog˘lu. The incorporation of objective and subjective aspects into the assessment of information systems. Faulkner. [3] G.J. Hannis. Computer and Mathematics with Applications 15 (1988) 887–896. Li. Accelerating the new product introduction with intelligent data mining. [11] S. IEEE Transaction on Systems. New product development: the performance and time-to-market tradeoff. the other single level multiattribute methods like outranking methods and multi-attribute utility theory can be developed to compare with the results of fuzzy hierarchical TOPSIS. International Journal of Advanced Manufacturing Technology 21 (2003) 284–290.). Chen. Feyziog˘lu.H. 2005.S. Kaufmann. Bu¨yu¨ko¨zkan. Comparison of fuzzy numbers based on the probability measure of fuzzy events. [16] J. [9] T. O 784–801. A. Land. For further research. Journal of Product Innovation Management 11 (1994) 397–417. [4] P. [15] R. European Journal of Operational Research 163 (2005) [33] M. [20] D. Criteria employed for go/no-go decisions when developing successful highly innovative products. Wets (Eds. An interactive approach for selecting R&D project. R. [27] E. J. An empirical study on the drivers of management control systems’ design in new product development. Ranking fuzzy numbers with integral value. Fuzzy MCDM based on ideal and anti-ideal concepts.J.A.A. Chen. [25] R. [2] G. M. [35] J.C.-C.G. [26] V.P. Cohen. Fuzzy Mathematical Models in Engineering and Management Science.I.T. Yoon.J.S. Doctor.B. Ho. R. Supper (Eds. Fuzzy Sets and Systems 50 (1992) 247. .. Raiffa. Lincoln. Decisions with Multiple Objectives: Preferences and Value Trade Offs. [12] R. Fuzzy Sets and Systems 17 (1985) 113–129. [23] P. A. V. Chen. Rodrı´guez-Escudero. Intelligent Data Mining Techniques and Applications. C. Lin. H. Facility location selection using fuzzy TOPSIS under group decisions. Parnell.M. [6] S. References [1] G. New York. Managing uncertainty in research and development. A new approach based on soft computing to accelerate new product introduction. Fuzzy logic to improve the robustness of decision support systems under uncertainty. Research and Technology Management 39 (1996) 50–57. Fuzzy Sets and Systems 114 (2000) 1–9. O. Kerre. [24] A. The Journal of Finance 7 (1952) 77–91. R. Naudia. Chen.L. Jovanovic´.L.L.G. 1992. Multi-attribute decision-making using a fuzzy heuristic approach.M. Pereus Publishing. International Journal of Uncertainty.L. Li.-C. A fuzzy-logic-based decision-making approach for new product development. Determinants of new product performance: a review and meta analysis. Hall. 1988. MA. Winning at New Products: Accelerating the Process from Idea to Launch. B. Decision Support Systems 27 (2000) 395–409. [29] G. Gupta. [17] J.-C. [28] S. International Thomson Publishing. New York. Fuzziness and Knowledge-Based Systems 10 (2002) 687–701. Springer-Verlag. Garavelli. Jackson.L. Eliashberg. K. Journal of Product Innovation Management 14 (1997) 65–66. Product development decisions: a review of the literature. L. Efstathiou. Calantone. O. Kahraman et al. New York. Application of sensitivity analysis in investment project evaluation under uncertainty and risk. Lehmkuhl.J. Perspective: third generation new product processes. Bu¨yu¨ko¨zkan. P. Schmidt. Cooper. Ranking fuzzy numbers with maximizing set and minimizing set. [21] C. Industrial Marketing Management 33 (2004) 307–316. Cybernetics 9 (1979) 326–333. Liou. Ivey. An R&D options selection model for investment decisions. Carbonell-Foulquie´.E. Longhurst. North-Holland. A. 1980. Calantone. K. D. P. [7] S. Hwang. [19] A. T. Pearson. J. Davila. Liang. Multiple Attribute Decision Making Methods and Applications. third ed. [8] T. A fuzzy TOPSIS method for robot selection. [22] G. Wang. 1994. Organizations and Society 25 (2000) 383–409. in: Lucas.N. Chu. Springer. Fuzzy Multiple Attribute Decision Making Methods and Applications. Burk. G. Cooper. Factors which influence decision making in new product evaluation.L. E. Bu¨yu¨ko¨zkan.S. Engelbrecht. O. North Holland. Scozzi. Journal of Product Innovation Management 11 (1994) 3–14. J. The development of a hybrid intelligent system for developing marketing strategy. 187–197. in: D. 1976. [32] M. Efstathiou.J. [18] T. IEEE Transactions on Engineering Management 37 (1990) 126–133. pp. [31] H. Management Science 47 (2001) 1–21. / Information Sciences 177 (2007) 1567–1582 1581 approach seems to be usable for the solution of this problem. [30] T. J. Portfolio selection. John Wiley & Sons. European Journal of Operational Research 112 (1999) 682–691. Coldrick. Computers and Industrial Engineering 37 (1999) 477–480. pp. [13] R. R&D concept decision analysis: using alternate futures for sensitivity analysis. Newton. Montoya-Weiss.-H. 1981. Are really new product development projects harder to shut down? Journal of Product Innovation Management 15 (1998) 111–123. [10] M. ¨ zer. Markowitz. Accounting. International Journal of Project Management 17 (1999) 217–222. Ulrich. Lee. Gorgoglione.-C. Berlin. Springer-Verlag.-S. M.C.A. R. Extensions of the TOPSIS for group decision-making under fuzzy environment. Munuera-Alema´n. International Journal of Production Economics 90 (2004) 27–45. Keeney. Chu. 2001. [5] T.J. Rajkovic. 375–394. Planning Under Uncertainty. Man. Journal of Multi-Criteria Decision Analysis 8 (1999) 119–127. Applying options thinking to R&D valuation. [14] T. Infanger. Technovation 21 (2001) 79–90. Hwang.C. The Information Systems Environment. Ruan. M. [34] G. G. Computers in Industry 54 (2004) 151–167. Y.). Feyziog˘lu. Technovation 25 (2005) 185– 193. J.

A.B. Research Management 29 (1986) 36–42. Shipley. 1995. Tourism Management 23 (2002) 107–115. R. Interfaces 16 (1996) 108.-C.-H. Multiple Attribute Decision Making: An Introduction. A. Schmidt. Calantone. de Korvin. R&D project selection models. [39] H.F. Using the analytic hierarchy process in new product screening.R. Journal of Product Innovation Management 16 (1999) 65–76. The evaluation of airline service quality by fuzzy MCDM. Kahraman et al. Information and Control 8 (1965) 338–353. T. A fuzzy logic-based decision model to satisfy goals for successful product/service introduction.F. Lu. Zhao.E. [40] S. Fusfeld. [42] L.A. . Di Benedetto. [44] G. Yen. Zadeh. [37] W. Further reading [46] R.L. Mandakovic. O. Trittle.V. Research and Technology Management March– April (2000) 47–55. Information Sciences 54 (1991) 103–130. Yoon. Zahedi.A. Hwang.-Y. Tsai. Tsaur.-W. [45] R. Scriven. Chang. E.L. C. S. An integrated group decision-making method dealing with fuzzy preferences for alternatives and individual judgments for selection criteria. Algebraic characteristics of extended fuzzy numbers. / Information Sciences 177 (2007) 1567–1582 [36] M.P. Govind. Thousand Oaks. Fuzzy sets. [41] K. Khursheed.-H. European Journal of Operational Research 135 (2001) 209–219.J. Hsiao. [38] G. J. [43] F. C. Evaluation of alternatives for product customization using fuzzy logic. C. Sage Publications. Zhang.1582 C. J. Souder. Group Decision and Negotiation 12 (2003) 501–515. The analytic hierarchy process—a survey of the method and its applications. T. Information Sciences 158 (2004) 233–262. Resolving uncertainty in R&D portfolios.