Int. J. Production Economics 90 (2004) 27–45

A fuzzy-logic-based decision-making approach for new product development
Gulcin Buyukozkan*, Orhan Feyz’o* lu . . . . ı g
Industrial Engineering Department, Galatasaray University, 80840 Ortakoy, Istanbul, Turkey . ’ Received 15 April 2002; accepted 16 April 2002

Abstract The managers spend most of their time to take critical decisions in each level of the organizational hierarchy. Basically, the decision process is to weight diverse alternatives each having the purpose to attain some of the desired objectives and to figure out the best solution in the complete challenging goals set. In today’s rapidly changing and highly uncertain environment, the strategic decisions have an extremely complex and fuzzy nature. In the meantime, the enterprises have tendency to appreciate the new product development (NPD) activities so as to deal with the innovation in the new economy and to fulfill the customer demands adequately. The aim of this study is then to improve the accuracy of decision-making in NPD under uncertainty. We first identify the decision points in the NPD process and the uncertainty factors affecting those points. Next, we determine the necessary decision models and techniques to help the decision makers to reduce their risks. Finally, we propose an integrated approach based on fuzzy logic to shape the decisions and illustrate with an application in software development. r 2003 Elsevier Science B.V. All rights reserved.
Keywords: New product selection; Uncertainty; Fuzzy logic; Multi-criteria decision making; Software development

1. Introduction The new product development (NPD) and innovation are often recognized as the key processes of competition in a variety of markets (Brown and Eisenhardt, 1995; Drucker, 1999; Hamel and Prahalad, 1994; Jones, 1997; McQuater et al., 1998). Today, markets are generally perceived to be demanding higher quality and higher performing products, in shorter and more predictable development cycle-times and at lower
*Corresponding author. E-mail addresses: (G. Buyukozkan), (O. Feyz’o* lu). . . . ı g

cost (Maffin and Braiden, 2001). In order to obtain best performance from NPD, the efficient and effective management of the product development process is vital. However, new product failure rates are substantial and the cost of failure is large, and this makes successful NPD rather a complicating task to be exercised with caution. The NPD process, that its objective is to translate 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. The decision maker must take into account the customers’ needs, the company’s strategies as well as technological opportunities and the company’s

0925-5273/03/$ - see front matter r 2003 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 5 - 5 2 7 3 ( 0 2 ) 0 0 3 3 0 - 4

28 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

resources, and, deduce the goals based on these factors for a successful NPD. With the support of a successful management system, an enterprise must be able to determine the right products or features to be developed, the right time to develop and launch, the right amount of development investments, its effective implementation, etc. As it can be easily understood, no NPD operation can be accomplished without effective and timely decision-making. An important cornerstone of the new product management is the idea selection and new product project launch decision. Several researchers have suggested that it is difficult for managers to end NPD projects once they are begun (Cooper, 1994; Schmidt and Calantone, 1998). For this reason, here we focus especially on increasing the accuracy of the necessary decisions before a new product project launch. Similar to all decision problems, NPD decisions are affected by many uncertainty-causing elements that confuse the decision maker to reach targeted performance. Uncertainty is an information defect (Spender, 1993), which may be defined as the difference between the amount of information required to perform a particular task and the amount of information already possessed (Galbraith, 1973). It arises from a multiplicity of sources including technical, management and commercial issues, both internal and external to the project. It is also widely recognized and accepted that successful management of uncertainty is intimately associated with project success, as the proactive project manager constantly seeks to steer the project towards achievement of desired objectives (Hillson, 2002). Thus, it is critical to use a structured approach that can minimize the uncertainty at NPD projects. The rest of the paper is organized as follows. In the next section, we present briefly NPD process and its decision points. In Section 3, we give in a general setting the uncertainty factors affecting those decision points and the available riskreducing techniques for the decision makers. Section 4 describes our proposed decision-making methodology to select and launch NPD projects. The details of the methodology are explained through a case study in Section 5. Finally, the last

section contains some concluding remarks and perspectives.

2. New product development process and its decision points NPD is defined as the transformation of a market opportunity and a set of assumptions about product technology into a product available for sale (Krishnan and Ulrich, 2001). NPD is an interdisciplinary activity (Davila, 2000) including marketing management, organizations, engineering design, operations management and requires contributions from nearly all the functions of an enterprise, whether it is an upgrade (an improvement of an existing product) or a new concept either to the company or to the market (Haque et al., 2000). With the NPD activities, it is aimed to create value for enterprises while renewing and developing (Matheson and Matheson, 1998). As we have pointed out earlier, NPD has a vast working area and it address different strategic, tactic and operational managerial abstraction levels in the organization. This is why methodologies, assumptions, goals and realization stages vary among companies. Although different organizations can make different choices and may use different methods, all of them make decisions about a collection of issues such as the product concept, architecture, configuration, procurement and distribution arrangements, project schedule, etc. Consequently, NPD can be defined as a process including many ‘‘generic decision’’ points, likewise ‘‘decision perspective’’ of Krishnan and Ulrich (2001). In their related work, Urban and Hauser (1993) recommend a five-step decision process for NPD: opportunity identification, design, testing, introduction and life cycle management. These phases are briefly illustrated in Fig. 1. To conclude, NPD process may be accepted as a dynamic decision process where each decision point must be evaluated, selected, and prioritized. All the stages of the process are affected by uncertain, changing information and dynamic opportunities, which will now be summarized.

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . 29

Fig. 1. NPD process (Matheson and Matheson, 1998).

3. Uncertainty in NPD process and decisionmaking methods NPD, by its very nature, is characterized by uncertainty (Davila, 2000; Lysonski et al., 1995; Mullins et al., 1999; Smith, 1999). Uncertainty is an information defect (Spender, 1993), which may be defined as the difference between the amount of information required to perform a particular task and the amount of information already possessed

(Galbraith, 1973). Uncertainty management is an integral part of NPD projects and so it can be observed that different approaches exist in the literature to define and analyze uncertainty in NPD. Fox et al. (1998) combine three dimensions of uncertainty as technical, market and process. They rate and categorize uncertainty along each dimension as being either low or high. For technical uncertainty, when uncertainty is low, the technologies used in the development of the project are well known to the organization and relatively stable. When technical uncertainty is high, technologies used in the development of the project are neither existent nor proven at the start of the project, and/or are rapidly changing over time. For market uncertainty, when uncertainty is low the organization has good market data on both customers and competitors, and product is being sold through familiar channels of distribution. When market uncertainty is high, the organization has little information regarding who the customer is, how the market is segmented and what are the needed channels of distribution. For process uncertainty, when uncertainty is low the engineering, marketing, and communications (both internal and external) processes used in this project are well tested, stable, and embedded in the organization. When process uncertainty is high, a significant portion of any or all of the engineering, marketing, and communications processes are relatively new, unstable, or evolving. Similarly, Mullins and Sutherland (1998) identify three levels of uncertainty that confront companies operating in rapidly changing markets. First, potential customers cannot easily articulate needs that a new technology may fulfill. Consequently, NPD managers are uncertain about the market opportunities that a new technology offers. Second, NPD managers are uncertain about how to turn the new technologies into new products that meet customer needs. This uncertainty arises, not only from customers’ inability to articulate their needs, but also from managers’ difficulties in translating technological advancements into product features and benefits. Finally, senior management faces uncertainty about how much capital to invest in pursuit of rapidly changing markets as well as when to invest.

30 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

Miller and Lessard (2001) identify three main risk categories for engineering projects: ‘‘completion risks’’ group formed by technical, construction and operational risks, ‘‘market related risks’’ group formed by demand, financial and supply risks and finally, ‘‘institutional risks’’ group formed by social acceptability and sovereign risks. We refer also to the recent work of Riek (2001) where NPD risks from uncertainty are organized into three general categories such as technical risks, commercial risks and NPD personnel. If we analyze NPD from different perspectives, we can precise risk structure in a more detailed manner. As an example, we can allocate product positioning, pricing and customer uncertainties to marketing; organizational alignment and team characteristics uncertainties to organizations; concept, configuration and performance uncertainties to engineering design; supplier, material, design of production sequence and project management uncertainties to operations management. As it can be observed, uncertainty factors highly depend on the way of how to focus and investigate the theme. However, we can briefly state that, all kinds of uncertainties for NPD can be classified generally in two main categories: uncertainty caused by external factors and uncertainty caused by internal factors. External factors can be further subdivided into two groups: market factors regarding to competitors, customers and suppliers, and technological factors. By the same reasoning, internal factors can be subdivided into to personnel and project management factors. While considering the decision points in whole NPD process, we require to minimize the side effects of uncertainties described previously and to increase the effectiveness of the decisions. Different decision methods have been developed to overcome the uncertainty related problems. Some of the methods that can be used in NPD process are summarized below (Davila, 2000; Doctor et al., 2001; Infanger, 1994; Li, 2000; Trittle et al., 2000). Probabilistic models: These include Monte Carlo Simulation and decision trees (Souder and Mandakovic, 1986). Monte Carlo analysis uses the process of simulation to achieve a range of solutions to a problem. Decision tree is a diagram that provides a structured approach to decision-

making that incorporates uncertainty of outcome and expected revenues. Options pricing theory (OPT): It is being proposed as a means of understanding what level of research investment is justified for a particular project. It treats each stage of the new product project much like purchasing an option on a future investment (Faulkner, 1996). Scoring models and checklists: Here, projects are rated and scored on a variety of qualitative questions (in some cases, the project score becomes the criterion for project prioritization) (Hall and Naudia, 1990). The questions or items often capture proven drivers of new product success such as product advantage, market attractiveness, and synergy with the base business (leverages core competencies), familiarity, etc. (Montoya-Weiss and Calantone, 1994). Behavioral approaches: These are tools designed to bring managers to a consensus in terms of which projects to undertake, and include methods such as the Delphi method that is a qualitative forecasting method which uses a panel of experts (Souder and Mandakovic, 1986). They are particularly useful for the early stages, where only qualitative information is available. Analytical hierarchy process (AHP): These are decision tools based on paired comparisons of both projects and criteria (Saaty, 1980). Software tools such as expert choice enable a team of managers to arrive at the preferred set of projects in a portfolio (Zahedi, 1996), with relative ease. Fuzzy logic: It deals with problems in which a source of vagueness is involved (Zadeh, 1965). In general, the probability concept is related to the frequency of occurrence of events, captured by repeated experiments whose outcomes are recorded, while the fuzzy sets provide the appropriate framework to evaluate the possibility of events rather than their probability (Garavelli et al., 1999). Sensitivity analysis: It examines how the optimal solution and the optimal objective value are affected from the changes of the uncertainty parameters (values and probabilities) that are considered to be important (Parnell et al., 1999). Scenario analysis: This technique has been widely preferred and used by many decision

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . 31

makers. Here, a combination of possible values of the uncertainty parameters are assumed regarding to different point of views (e.g. pessimistic, neutral and optimistic), and the resulting scenario is solved. By solving the problem repeatedly for different scenarios and studying the solutions obtained, the decision maker observes sensitivities and heuristically decides on an appropriate solution. These techniques can be used exclusively or in a hybrid way. In this study, we have used fuzzylogic-based methods, which will be described in details in the next sections.

4. Proposed fuzzy-logic-based decision-making approach ‘‘Project selection is pivotal to effective risk reduction in new product development’’ (Cooper, 1983). Knowing that uncertainty highly disturb the accuracy of the decisions in NPD process, it is important to balance strategic management decisions (develop the right product while allocating resources) with tactical management decisions (execution of projects by minimizing its risks factors). The strategic aspect of NPD alone requires the resolution of some very important questions, namely:



Are we allocating our budget to the right new product proposal? Do we have the right balance; of risk and return, of long and short term projects? Are we working on the right projects and programs with the appropriate effort?

It is clear that for success in NPD, it is critical to determine what is ‘‘right’’ for the particular company. After defining the business objectives for the NPD program and the overall strategic framework that will govern the development plan, it is possible to move on to what is probably one of the most problematic parts of NPD management: the selection of NPD project. There is a comprehensive literature of potential methods, which can be used for this selection phase (Cooper

et al., 1999; Martino, 1995; Shipley et al., 2001; Subramanian et al., 2001). A feasible starting point for evaluating any investment is to consider the investment’s revenues and risks (Markowitz, 1952). However, NPD investments have some special characteristics that make it very complicated to evaluate their costs, and especially their benefits. First, they are intangible in nature. In many cases, monetary measures cannot be used directly, but subjective arguments have to be applied also. Second, the benefits of NPD investments are typically realized over a long period of time. Thus, using only traditional investment evaluation techniques for evaluating these investments is not sufficient. Third, the success factors and the benefits are recognized differently by various interest groups in NPD investments. Fourth, these NPD investments are irreversible in nature. Therefore, an appropriate decision criteria collection and evaluation/ discrimination methodology has to be identified while respecting these special attributes of NPD process and the requirements of the enterprise. Here we suggest using an integrated decisionmaking approach for NPD under uncertainty. Focused on the fundamental stage of the NPD process, it enables the identification of the best new product to be developed together with its bestsuited development strategy while trying to minimize the associated risk and uncertainty. The first step of the method is to support toplevel managers to evaluate multiple new product ideas at a strategic level. At that level, the needed information for the evaluation is numerous and highly uncertain. Moreover, decision criteria do not have an apparent superiority on each other and estimated performances of ideas are imprecise. As this strategic decision will have a great impact on the company’s future, we propose not to figure out the best new product idea at this early stage, but rather choosing a nondominated subset of them. This set must consists of ideas proven to be superior from the remaining ones in some extent, but cannot be differentiated from each other. In order to realize such a classification, responsible managers have to weight the proposed new product ideas by taking into account the apparent benefits and risks. The assessments are based on

32 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

the hierarchical structure shown in Fig. 2. Here, it is more convenient to use subjective judgments on ideas in linguistic terms like ‘‘fairly good’’ or ‘‘poor’’. Then, individual opinions are pooled to arrive at a final decision. This step make use of the pseudo-order fuzzy preference model (Roy and Vincke, 1984; Wang, 1997) to discriminate the set of ideas into dominance and nondominance sets without the information about the relative importance of each evaluation criterion. The first level’s winning ideas are further evaluated in the second step in order to choose the best one. Here, more product specific features are highlighted and investigated, and detailed NPD projects (namely also proposals) are prepared. The decision criteria are weighted according to their relative importance for the company. Again, we do not require exact judgments values neither for criteria nor for projects. Criterion importance is expressed with terms like ‘‘low’’ or ‘‘high’’ similar to the desirability of the proposals for these criteria. In this step the fuzzy weighted average (FWA) method (Vanegas and Labib, 2001) is used to find the ranking of projects. As projects features are more involved at this decision point,

we are confronted more or less with a tactical decision rather than a strategic decision. Evaluators’ judgments are once more aggregated to obtain a refined result. The conclusion is not only the recognition of the best proposal but also the right order of the new product projects to be developed subsequently. In the third and last step, we try to best figure out the necessary development strategy for the preferred NPD project. Depending on the business area and industry specific production means, these strategies are supposed and exposed by the project responsive team. This is tactical decision where selected strategies are criticized for different hierarchically leveled criteria by means of pairwise comparisons and the best is chosen accordingly. One of the methods that is intended for this type of evaluation is the fuzzy analytic hierarchy process (FAHP) (Triantaphyllou, 2000) and is applied here. The method itself possesses the capability to join evaluators’ ideas without pressurizing on them to make strict judgments on alternatives. Our proposed three-step decision-making approach to evaluate and launch NPD projects is summarized in Table 1.

Fig. 2. NPD strategic hierarchic decision criteria.

Table 1 Proposed methodology for the fundamental evaluation in NPD process Fundamental evaluation in NPD process Selection of new product ideas Major concerns Concerning organizational level Strategic Proposed methods Motivation

Selection of the most appropriate ideas in terms of company’s benefits and risks

Pseudo-order fuzzy preference model (Roy and Vincke, 1984)

When the estimated performances for all alternatives are imprecise, evaluating and ranking all of them in detail is not practical. The pseudo-order preference model is used to discriminate the set of alternatives into dominance and nondominance sets without the information about the relative importance of each criterion (Wang, 1997) The fuzzy weighted average method used in this study is better than the conventional fuzzy weighted average method, because it produce less imprecision and the balance point of the fuzzy number is more realistic, and this point dmay be better basis for making decisions (Vanegas and Labib, 2001) Among the different fuzzy MCDM methods, AHP is the most widely used and easily understandable one. Fuzzy AHP is a natural extension of this traditional method where decision makers do not require to express their assessments through crisp values but rather they use fuzzy numbers which is more suitable when uncertainty is high

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .


Selection of the best new product project

Evaluation of the new product proposals according to selected decision criteria and identification of the new product project launch order consistent with their desirability

Strategic– Tactical

Fuzzy weighted average method (Vanegas and Labib, 2001)

Selection of the best development strategy

Selection of development strategies which minimize project uncertainties


Fuzzy AHP method (Triantaphyllou, 2000)


34 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

The next section is devoted to a case study for the new software development process in order to clarify the subject. A brief introduction to fuzzy sets theory is given in Appendix A for the unfamiliar readers about the fuzzy logic.

5.1. Strategic decision stage A fuzzy preference relation R on a set A is a fuzzy set on the product A  A; such that ma : A  A-½0; 1Š: Let Pða; bÞAR be the fuzzy preference relation between a and b; where a; bAA: Then Pða; bÞ þ Pðb; aÞ ¼ 1: Note that the higher value of Pða; bÞ means a stronger intensity. The fuzzy preference relation between two ideas a and b for criterion i is obtained by a pairwise comparison of gi ðaÞ and gi ðbÞ which show the linguistic performance of ideas a and b; respectively. gi ðaÞ and gi ðbÞ are represented by fuzzy numbers. This paper utilizes Tseng and Klein’s (1989) approach based on Hamming distance to derive the preference relation between two proposals. If a and b are two fuzzy numbers, the fuzzy preference relationship is given as Dða; bÞ þ Dða-b; 0Þ ; ð1Þ Pða; bÞ ¼ Dða; 0Þ þ Dðb; 0Þ where Dða; bÞ is the area where a dominates b (none in Fig. 3), Dða; 0Þ the area of a (areas 1 and 2 in Fig. 3), Dðb; 0Þ the area of b (areas 2 and 3 in Fig. 3), Dða-b; 0Þ intersection areas of a and b (area 2 in Fig. 3). As easily seen from this equation, preference relations are obtained by using related areas under fuzzy membership functions (Tseng and Klein, 1989). For the example given in Fig. 3, we have Pða; bÞ ¼ 0:18 and Pðb; aÞ ¼ 0:82: When the relative importance of criterion is not really known, the pseudo-order preference model can be used to set the alternatives into nondominance and dominance set. Three types of preference relations are defined in terms of the fuzzy preference relations between two alternatives: 8a; bAA and iAC: aPi b3Pðgi ðaÞ; gi ðbÞÞ À Pðgi ðbÞ; gi ðaÞÞ > pi ; aQi b3Pðgi ðaÞ; gi ðbÞÞ À Pðgi ðbÞ; gi ðaÞÞppi ; aIi b3jPðgi ðaÞ; gi ðbÞÞ À Pðgi ðbÞ; gi ðaÞÞjpqi ; where Pi depicts strict preference, Qi depicts weak preference and Ii depicts indifference. The preference threshold pi and indifference threshold qi are used to discriminate between indifference,

5. A case study in new software development process As Ulrish (2001) indicates, design and NPD problems are common across the domains of physical goods, software and services. Even though our approach can be adapted to any development process, we decided to apply it to software development since effective deployment of computer software has emerged as one of the most important determinants of success in the business world over the past decades. As information technology infiltrates and plays a critical role in all aspects of the value chain, firms are continuing to investigate heavily in software (Harter et al., 2000). Software development is a special case for NPD and there exists differences that are worthwhile to notice between software production and traditional production: it is not repetitive, individual attributes of the personnel have an extreme influence on the quality of the product, the failure probability in one production step is highly dependent to the previous steps (i.e. if conception phase is not adequately ended, then it is very likely that the implementation phase will fail). This implies that uncertainty factors are dominants in whole production process. With this case study, we give the details of the fuzzy-logic-based three-step evaluation methodology starting from the selection of the most appropriate new software idea to the selection of the least risky and most value added software project development strategy. The necessary data sets of this study are obtained from a small– medium-sized Turkish software development company. However, we do not supply more projects’ details since the subject company has reserved the information as confidential.

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . 35

Fig. 3. Example of the fuzzy preference relation between a and b:

strict preference, and weak preference of two alternatives for criterion i: According to the three preference relations outranking is as follows (Roy and Vincke, 1984): 8a; bAA and iAC: ðOutrankingÞ aSb3fbPi a is false; 8ig and jV j þ jW jXjY j ðIncomparabilityÞ aRb3 otherwise; ð2Þ

decision stage, we choose seven levels in linguistic terms: ‘‘very poor (VP)’’, ‘‘poor (P)’’, ‘‘fairly poor (FP)’’, ‘‘medium (M)’’, ‘‘fairly good (FG)’’, ‘‘good (G)’’, ‘‘very good (VG)’’. The linguistic terms are transformed into fuzzy numbers with Fig. 4 and are listed in Table 2. The strategic managerial team consisting of two company owners, four branch managers and one sales and marketing manager has joined together to rate new product ideas based on their expertise and initiatives. Note that all shown values in Table 2 are the aggregated ones to represent the whole decision range of the decision makers. The decision range is determined by taking the worst and the best assessments among evaluators for each comparison. According to Eq. (1), the fuzzy preference relations among eight project ideas obtained and displayed in Table 3. A node of the outranking graph represents an alternative and an arc represents the outranking relation between two alternatives. If idea a outranks b; then there is an arc from node a to b: When the indifference threshold qi and the preference threshold pi for all i are accepted, the outranking graph can obtained according to Eq. (2). By choosing qi ¼ 0:25 and pi ¼ 0:85 for all i like in Wang (1997), the graph illustrated in Fig. 5 is obtained. It can be easily deduced that the nondominance sets (SND) and dominance sets (SD) are SND ¼ f1; 3; 6g and SD ¼ f2; 4; 5; 7; 8g

where V ¼ fi j aPi bg is the set of a’s strictly preferred to b; W ¼ fi j aQi bg is the set of a’s weakly preferred to b; Y ¼ fi j bQi ag is the set of b’s weakly preferred to a; and jxj represents the cardinality of the set x: The above implies that a project a outranks b if no criterion considers that concept b is strictly preferred to a; and the number of criteria which consider a is preferred to b is more than the number of criteria which consider that b is weakly preferred to a: Otherwise, a is incomparable to b: In our case, we have eight new software idea to be evaluated according to eight criteria given in Fig. 2. The universe of discourse is a finite set of fuzzy numbers used to express an imprecise level of performance of each criterion. In this strategic

and that ideas 1, 3 and 6 outrank ideas 2, 4, 5, 7 and 8. Therefore, these nondominated ideas are kept for further development. 5.2. Strategic–tactic decision stage Bearing in mind that there exists n proposals to be evaluated versus m criteria, let us denote Wi as the fuzzy number representing the importance of the criterion Ci ; and Dij as the desirability of a jth proposal with respect to that criterion, for all iAI ¼ f1; 2; y; mg and for all jAJ ¼ f1; 2; y; ng: Let also Dj as the overall desirability of the proposal j: The a-cut of the over all desirability Dj of a proposal j; calculated through fuzzy weighted

36 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

Fig. 4. The outranking graph of pseudo-order preference structure, when qi ¼ 0:25 and pi ¼ 0:85: Table 2 Evaluation results for each project proposal in strategic decision stage Project proposals 1 Profitability Efficiency Strategic value Business impact Financial Technical Managerial Personel VG–G FG–M VG–FG VG–G G–M M–FP VG–G FG–FP 2 M–FP M–P FG–FP FG–FP FG–M FP–P FG–P M–FP 3 VG–M VG–G FG–M VG–M G–FG FG–M FG–M FG–M 4 FG–M M–FP G–M FG–M FG–M FG–FP FG–M G–M 5 G–FP M–P P–VP G–M M–FP FP–P M–FP FP–P 6 FG–M FG–M VG–FG FG–M M–FP G–M G–M VG–G 7 FP–P FP–P FG–M FP–P G–FG FP–P M–FP FG–M 8 M–P M–FP G–FP P–VP FG–M FG–FP FP–P FP–P

average (FWA) (Vanegas and Labib, 2001) for m desirability levels represented by the fuzzy numbers D1j ; D2j ; y; Dmj ; with weights (fuzzy numbers) W1 ; W2 ; y; Wm ; is given by Da ¼ ½Da ; Da Š; j j;a j;b where Da j;a P a a iAI D w P ij;aa i ¼ min iAI wi ð3Þ

and Da j;b 

P a a iAI D w P ij;ba i ¼ min iAI wi 8iAI and 8aAð0; 1Š:


for all jAJ: Here,
a a wa A½Wi;a ; Wi;b Š i



As it can easily be understood, Da and Da rej;a j;b present, respectively, the lower and upper limits of the a-cut Da ; Da and Da represent, respectively, j ij;a ij;b

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . Table 3 The fuzzy preference relations between proposals for each criterion in strategic decision stage 1 2 3 0.68 0.13 0.50 0.37 0.35 0.37 0.03 0.12 4 0.95 0.18 0.63 0.50 0.44 0.50 0.05 0.16 5 0.87 0.34 0.65 0.56 0.50 0.56 0.25 0.31 6 0.95 0.18 0.63 0.50 0.44 0.50 0.05 0.16 7 1.00 0.64 0.97 0.95 0.75 0.95 0.50 0.56 8 1.00 0.56 0.88 0.84 0.69 0.84 0.44 0.50 1 Efficiency 1 0.50 2 0.16 3 0.95 4 0.18 5 0.16 6 0.50 7 0.05 8 0.18 2 0.84 0.50 1.00 0.56 0.50 0.84 0.44 0.56 3 0.05 0.00 0.50 0.00 0.00 0.05 0.00 0.00 4 0.82 0.44 1.00 0.50 0.44 0.82 0.36 0.50 5 0.84 0.50 1.00 0.56 0.50 0.84 0.44 0.56 6 0.50 0.16 0.95 0.18 0.16 0.50 0.05 0.18 7 0.95 0.56 1.00 0.64 0.56 0.95 0.50 0.64 8 0.82 0.44 1.00 0.50 0.44 0.82 0.36 0.50 37

Profitability 1 0.50 1.00 2 0.00 0.50 3 0.32 0.87 4 0.05 0.82 5 0.13 0.66 6 0.05 0.82 7 0.00 0.36 8 0.00 0.44 Strategic value 1 0.50 0.76 2 0.24 0.50 3 0.30 0.62 4 0.37 0.66 5 0.00 0.04 6 0.50 0.76 7 0.30 0.62 8 0.30 0.54 Financial 1 0.50 2 0.44 3 0.56 4 0.44 5 0.16 6 0.16 7 0.56 8 0.44 Managerial 1 0.50 2 0.03 3 0.05 4 0.05 5 0.00 6 0.17 7 0.00 8 0.00

0.70 0.38 0.50 0.56 0.00 0.70 0.50 0.44

0.63 0.34 0.44 0.50 0.00 0.63 0.44 0.40

1.00 0.96 1.00 1.00 0.50 1.00 1.00 0.97

0.50 0.24 0.30 0.37 0.00 0.50 0.30 0.30

0.70 0.38 0.50 0.56 0.00 0.70 0.50 0.44

0.70 0.46 0.56 0.60 0.03 0.70 0.56 0.50

Business impact 1 0.50 0.96 2 0.04 0.50 3 0.32 0.70 4 0.05 0.62 5 0.17 0.66 6 0.05 0.72 7 0.00 0.28 8 0.00 0.04 Technical 1 0.50 2 0.36 3 0.82 4 0.62 5 0.36 6 0.84 7 0.36 8 0.62 Personnel 1 0.50 2 0.38 3 0.62 4 0.66 5 0.28 6 0.96 7 0.62 8 0.28

0.68 0.30 0.50 0.37 0.42 0.37 0.03 0.00

0.95 0.38 0.63 0.50 0.56 0.50 0.05 0.00

0.83 0.34 0.58 0.44 0.50 0.44 0.04 0.00

0.95 0.28 0.63 0.50 0.56 0.50 0.05 0.00

1.00 0.72 0.97 0.95 0.96 0.95 0.50 0.20

1.00 0.96 1.00 1.00 1.00 1.00 0.80 0.50

0.56 0.50 0.64 0.50 0.18 0.18 0.64 0.50

0.44 0.36 0.50 0.36 0.05 0.05 0.50 0.36

0.56 0.50 0.64 0.50 0.18 0.18 0.64 0.50

0.84 0.82 0.95 0.82 0.50 0.50 0.95 0.82

0.84 0.82 0.95 0.82 0.50 0.50 0.95 0.82

0.44 0.36 0.50 0.36 0.05 0.05 0.50 0.36

0.56 0.50 0.64 0.50 0.18 0.18 0.64 0.50

0.64 0.50 0.95 0.72 0.50 0.96 0.50 0.72

0.18 0.05 0.50 0.38 0.05 0.56 0.05 0.38

0.38 0.28 0.62 0.50 0.28 0.66 0.28 0.50

0.64 0.50 0.95 0.72 0.50 0.96 0.50 0.72

0.16 0.04 0.44 0.34 0.04 0.50 0.04 0.34

0.64 0.50 0.95 0.72 0.50 0.96 0.50 0.72

0.38 0.28 0.62 0.50 0.28 0.66 0.28 0.50

0.97 0.50 0.66 0.66 0.44 0.69 0.44 0.34

0.95 0.34 0.50 0.50 0.18 0.56 0.18 0.05

0.95 0.34 0.50 0.50 0.18 0.56 0.18 0.05

1.00 0.56 0.82 0.82 0.50 0.84 0.50 0.36

0.83 0.31 0.44 0.44 0.16 0.50 0.16 0.04

1.00 0.56 0.82 0.82 0.50 0.84 0.50 0.36

1.00 0.66 0.95 0.95 0.64 0.96 0.64 0.50

0.62 0.50 0.82 0.84 0.36 1.00 0.82 0.36

0.38 0.18 0.50 0.56 0.05 0.95 0.50 0.05

0.34 0.16 0.44 0.50 0.04 0.83 0.44 0.04

0.72 0.64 0.95 0.96 0.50 1.00 0.95 0.50

0.04 0.00 0.05 0.17 0.00 0.50 0.05 0.00

0.38 0.18 0.50 0.56 0.05 0.95 0.50 0.05

0.72 0.64 0.95 0.96 0.50 1.00 0.95 0.50

a the lower and upper limits of the a-cut Da ; and Wi;a ij a and Wi;b represent, respectively, the lower and upper limits of the a-cut Wia : The ‘‘min’’ and ‘‘max’’ operators take the minimum and maximum values, respectively, that can be calculated through the combination of the wi in all the possible ways. The minimum and maximum values in Eqs. (4) and (5) are always obtained by taking (real) values a of weight wi equal to the extreme values, Wi;a or

a Wi;b ; of the a-cut Wia for all iAI and all aAð0; 1Š: Then condition in Eq. (6) can be relaxed to a a wa AfWi;a ; Wi;b g i

8iAI and 8aAð0; 1Š


(see proof in Vanegas (1999) and Vanegas and Labib (2001)). A simple ranking method, which seems to be adequate for applications in which fuzzy numbers represent possibility distributions like here, is the

38 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

Fig. 5. The outranking graph of pseudo-order preference structure, when qi ¼ 0:25 and pi ¼ 0:85:

increasing fidelity, these submodels are called the applications composition, early design, and postarchitecture models. However, our work differs from the COCOMO II model in that we use fuzzy logic to evaluate the performance of proposals versus criteria. An illustrative evaluation is given in Table 4 where the examined criteria, their rated importance for the firm and the evaluations of the proposals are shown. We use appropriate triangular fuzzy numbers as indicated in Fig. 6 to capture the linguistic terms of ‘‘very low’’ (VL), ‘‘low’’ (L), ‘‘medium low’’ (ML), ‘‘medium’’ (M), ‘‘medium high’’ (MH), ‘‘high’’ (H) and ‘‘very high’’ (VH). The individual assessments of the decision makers are aggregated such that Dagg ¼ ð1=nÞ#ðD1 "D2 "?"Dk Þ; j j j j ð9Þ

calculation of balance points. A balance point of a fuzzy number is the x coordinate of the centroid of the area under the membership function curve and calculated such that 0Z b Z b x¼ mðxÞx dx mðxÞ dx; ð8Þ %
a a

where x is the balance point. Balance point % calculation is also known as a defuzzification method. In order to rate the proposals, we benefit from the main criteria considered in COCOMO II model [Web Address:]. This model allows one to estimate the cost, effort, and schedule when planning a new software development activity. It consists of three submodels, each one offering increased fidelity the further along one is in the project planning and design process. Listed in

where Dk stands for decision maker k judgment on j the desirability of proposal j calculated through FWA and Dagg for the aggregate assessment. j The development team of the company has prepared detailed projects for the first stage prosperous ideas and presented to the top managers. Then, managers’ individual judgments on projects and on the importance of the decision criteria are collected. After calculating the average assessment by means of Eq. (9), the overall desirability levels of proposals 1, 3 and 6 are about ‘‘medium’’, ‘‘medium towards medium low’’ and ‘‘high towards medium high’’ (see Fig. 6). Here, the balance points of proposals 1, 3 and 6 are calculated as 0.51, 0.46 and 0.81, respectively. Therefore, the best one is project 6 that has the highest balance point.

Table 4 Linguistic description of desirability and importance levels Importance Proposal desirability level Proposal 1 Scale drivers Product attributes Platform attributes Personnel attributes Project attributes Medium high Medium low Medium High Medium Medium Medium low High Medium Medium low Proposal 3 Low Medium high Medium low Medium high Medium Proposal 6 Medium high High High Very high Medium high

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . 39

Fig. 6. Overall desirability levels of the proposals calculated through the NFWA.

5.3. Tactic decision stage The analytic hierarchy process (AHP) (Saaty, 1977, 1980) uses the principle of comparative judgments to construct pairwise comparisons of the relative importance of elements at some given level of a criteria hierarchy with respect to shared criteria or property at the level above, giving rise to a weight matrix. Priorities are synthesized from the second level down from multiplying local priorities by the priority of their corresponding criterion in the level above and adding for each element in a level according to the criterion it effects. A traditional AHP-based approach requires construction of comparison matrices in which the relative importance among attributes is expressed as precise numbers on a standard scale (usually from 1 to 10). AHP technique requires decision makers to express their preferences for attributes using crisp numbers, and calculates a weight vector that quantifies the level of importance of attributes. However, precise numbers fail to contain the subjectivity and vagueness in decision-making in NPD. We overcome this difficulty of subjectivity and vagueness in the strategy selection problem by using fuzzy numbers as a

superior means of representing pairwise comparisons in the AHP judgment matrix. A comparison * matrix R is constructed for the n-strategy selection problem, in which pairwise comparisons are triangular fuzzy numbers aij * for all i; jAf1; 2; y; ng; such that * R¼
2 ð1; 1; 1Þ 6 * 6 1=a12 6 6 ^ 6 6 * 4 1=a1ðnÀ1Þ 1=a1n * a12 * ð1; 1; 1Þ ^ 1=a2ðnÀ1Þ * 1=a2n * a13 * a23 * ^ 1=a3ðnÀ1Þ * 1=a3n * ? ? ^ ? ? a1ðnÀ1Þ * a2ðnÀ1Þ * ^ ð1; 1; 1Þ 1=aðnÀ1Þn * a1n * a2n * ^ aðnÀ1Þn * ð1; 1; 1Þ 3 7 7 7 7: 7 7 5

The triangular fuzzy number aij ¼ ðlij ; mij ; uij Þ is * obtained for each criteria and alternative strategy by the integration of different members of the project team such that !1=K K Y mijk ; lij ¼ minðlijk Þ; mijk ¼

uij ¼ maxðuijk Þ;


where lijk ; mijk and uijk are the lower width, mean and upper width, respectively, for the decision maker k; and the project team has K members. lijk ; mijk and uijk can take any values between 1 and 1 10 and also their reciprocals (i.e. from 10 to 1) as

40 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

Fig. 7. Hierarchy of decision criteria in software development strategy selection.

original AHP provided that lijk pmijk puijk : Next, * the fuzzy eigenvector of the matrix R is estimated. According to Saaty (1980), the right principal eigenvector of the matrix expresses the importance of the alternatives. The combination of Saaty’s eigenvector approximation with the fuzzy arithmetic operations leads to the following equation: * li ¼ ðai1 #ai2 #?#ain Þ1=n ; iAf1; 2; y; ng; ð11Þ * * * À Á * * * * where l ¼ l1 ; l2 ; y; ln is the right principal eigenvector. Then we normalize the relative performance measures of alternatives for a given criterion by dividing the values with the sum of all. * * * * Therefore, if we denote l0 ¼ ðl01 ; l02 ; y; l0n Þ as the adjusted performance measures of alternatives, then we can construct the following fuzzy decision matrix for each subcriteria: C1 * 11 l 6* 6 l21 6 6 ^ 4 * ln1 2 C2 ? Cm 3 * * 12 ? l1m l * 7 * l22 ? l2m 7 7: ^ ^ ^ 7 5 * * n2 ? lnm l

Note that l*ij is the performance measure of the alternative i for the criteria j: Finally, the weighted average of each alternative is obtained by multiplying each criteria weight wj (given again as a * fuzzy triangular number) by the related alternative’s performance. In other words, we require * * * pi ¼ ðw1 #li1 Þ"ðw2 #li2 Þ"?"ðwm #lim Þ: ð12Þ * * * * This calculation process continues level by level in the hierarchical structure until the finite performance of alternatives is computed. The ranking of alternatives can be obtained by defuzzifying the final scores pi ¼ ðlpi ; mpi ; upi Þ with the following * method: rpi ¼ ½ðupi À lpi Þ þ ðmpi À lpi ފ=3 þ lpi 8i: ð13Þ Obviously, maxi rpi will be the best alternative. The hierarchical structure of decision criteria in software development strategy selection is shown in Fig. 7. More insight about the significance of these criteria can be found in Web Address: http:// In our stated SME, we have four general project teams. The managers of these teams are decided to evaluate three development strategies versus 5 main criteria and 21 subcriteria. In order to reduce space, we give in Table 5 the aggregated assessments only for

A1 * ¼ A2 Z ^ An

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . Table 5 Pairwise comparison matrices of development strategies for project attributes’ subcriteria Strategy 1 Criteria: TOOL Strategy 1 Strategy 2 Strategy 3 Criteria: SCED Strategy 1 Strategy 2 Strategy 3 Criteria: SITE Strategy 1 Strategy 2 Strategy 3 (1.00, 1.00, 1.00) (0.25, 0.49, 3.00) (0.25, 0.56, 3.00) Strategy 2 (0.33, 2.05, 4.00) (1.00, 1.00, 1.00) (0.25, 1.52, 4.00) Strategy 3 (0.33, 1.78, 4.00) (0.25, 0.66, 4.00) (1.00, 1.00, 1.00) 41

(1.00, 1.00, 1.00) (0.25, 0.76, 3.00) (0.20, 0.70, 3.00)

(0.33, 1.32, 4.00) (1.00, 1.00, 1.00) (0.25, 1.52, 4.00)

(0.33, 1.43, 5.00) (0.25, 0.66, 4.00) (1.00, 1.00, 1.00)

(1.00, 1.00, 1.00) (0.20, 0.49, 2.00) (0.20, 0.61, 2.00)

(0.50, 2.05, 5.00) (1.00, 1.00, 1.00) (0.20, 0.42, 2.00)

(0.50, 1.64, 5.00) (0.50, 2.41, 5.00) (1.00, 1.00, 1.00)

‘‘Project attributes’’ and its subcriteria: use of software tools (TOOL), multisite development (SITE) and required development schedule (SCED). The pairwise comparison matrices shown in Table 5 are constructed first by collecting individual judgments and then by applying Eq. (10) to combine all of them. When this step is repeated for all subcriteria, we use Eq. (11) to calculate all performances of alternative strategies and prepare the fuzzy decision matrices shown in Table 6. It can be seen from the table that each criterion’s weight is shown beneath criterion abbreviation. The weight represents the combined opinions of decision makers about the importance of that criterion and it is calculated by using Eq. (10). Note that the mean values of subcriteria for a given upper decision criterion must sum up to 1 for the sake of the robustness of the method. Therefore, we normalize the criteria weights as we explained previously for the relative performance measures of alternatives. Next, Eq. (12) is applied to obtain the performance of strategies against main criteria and results are shown in Table 7. Again by using Eq. (12), the final ranking is obtained such that p1 ¼ ð0:03; 0:44; 4:85Þ; p2 ¼ * * ð0:03; 0:28; 4:03Þ and p3 ¼ ð0:02; 0:28; 3:75Þ: Thus, * after applying Eq. (13) to defuzzify these numbers and normalizing the crisp values, we get rp1 ¼ 0:39; rp2 ¼ 0:32 and rp3 ¼ 0:29: This implies that the development strategy 1 has to be selected for the chosen project.

6. Concluding remarks and perspectives The managers spend most of their time to take critical decisions in each level of the organizational hierarchy. NPD is one of the most important activities covering crucial decisions for the survival of a company. This implies that the decisionmaking process must be timed and effective. For this reason, this work suggests a basic integrated decision-making approach for NPD under uncertainty. Focused on the fundamental stages of the NPD process, it enables the identification of the best new product to be developed together with its best-suited development strategy while trying to minimize the associated risks. Simplicity does not imply direct applicability and other traditional evaluation methods may be incorporated in each selection phase to our methodology. However, we claim that the application of our method will be a good practice in terms of the aggregation and purification of the subjective judgments and to clarify the big picture, which is covered by uncertainties. Besides, the company’s authorities participated in this case study have also approved that our approach has a significant support to improve the quality of the decision-making. Our proposed methodology requires tedious calculations which may mean some practitioners avoid using it. Actually, the necessary computations concerning the assessments have been carried

42 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . .

Table 6 Fuzzy decision matrices for all subcriteria Scale drivers [0.15, 0.24, 0.28] Strategies Strategy 1 Strategy 2 Strategy 3 PREC (0.14, 0.17, 0.42) (0.05, 0.29, 2.56) (0.05, 0.21, 2.00) (0.05, 0.50, 2.38) Product attributes [0.05, 0.12, 0.25] Strategies Strategy 1 Strategy 2 Strategy 3 RELY (0.17, 0.23, 0.36) (0.08, 0.37, 1.65) (0.06, 0.27, 1.65) (0.07, 0.36, 1.45) DATA (0.17, 0.18, 0.23) (0.11, 0.47, 1.48) (0.08, 0.34, 1.60) (0.06, 0.19, 0.94) CPLX (0.31, 0.34, 0.35) (0.07, 0.43, 2.01) (0.07, 0.35, 1.70) (0.05, 0.21, 1.35) RUSE (0.09, 0.13, 0.17) (0.05, 0.25, 2.40) (0.05, 0.51, 2.03) (0.05, 0.25, 1.71) DOCU (0.08, 0.12, 0.14) (0.08, 0.42, 2.06) (0.06, 0.21, 1.52) (0.05, 0.37, 1.74) FLEX (0.16, 0.30, 0.34) (0.08, 0.47, 2.01) (0.06, 0.30, 1.69) (0.06, 0.22, 1.34) RESL (0.21, 0.31, 0.41) (0.06, 0.46, 2.40) (0.05, 0.21, 1.88) (0.05, 0.34, 1.88) TEAM (0.18, 0.22, 0.34) (0.08, 0.48, 1.87) (0.06, 0.20, 1.49) (0.06, 0.32, 1.70)

Platform attributes [0.09, 0.18, 0.33] Strategies Strategy 1 Strategy 2 Strategy 3 TIME (0.31, 0.36, 0.57) (0.12, 0.53, 1.91) (0.06, 0.19, 1.31) (0.06, 0.28, 1.04) Project attributes [0.12, 0.18, 0.34] Strategies Strategy 1 Strategy 2 Strategy 3 TOOL (0.16, 0.30, 0.37) (0.07, 0.49, 1.98) (0.06, 0.22, 1.80) (0.06, 0.30, 1.80) Personnel attributes [0.18, 0.28, 0.36] Strategies Strategy 1 Strategy 2 Strategy 3 ACAP (0.21, 0.21, 0.34) (0.06, 0.50, 2.51) (0.04, 0.20, 2.12) (0.04, 0.30, 2.12) PCAP (0.15, 0.18, 0.28) (0.10, 0.52, 1.77) (0.08, 0.28, 1.18) (0.07, 0.20, 1.03) PCON (0.24, 0.26, 0.41) (0.10, 0.48, 1.77) (0.06, 0.25, 1.40) (0.06, 0.27, 1.40) PEXP (0.09, 0.11, 0.18) (0.07, 0.34, 1.85) (0.07, 0.44, 1.72) (0.05, 0.22, 1.56) AEXP (0.04, 0.08, 0.17) (0.06, 0.29, 1.93) (0.06, 0.50, 2.29) (0.06, 0.21, 1.43) LTEX (0.13, 0.16, 0.29) (0.09, 0.43, 1.56) (0.07, 0.24, 1.24) (0.08, 0.33, 1.42) SCED (0.22, 0.45, 0.53) (0.07, 0.41, 2.18) (0.05, 0.26, 1.84) (0.05, 0.33, 1.84) SITE (0.13, 0.25, 0.34) (0.09, 0.47, 2.04) (0.07, 0.33, 1.50) (0.05, 0.20, 1.11) STOR (0.32, 0.40, 0.50) (0.08, 0.48, 1.86) (0.06, 0.29, 1.59) (0.05, 0.23, 1.53) PVOL (0.21, 0.24, 0.25) (0.07, 0.28, 1.84) (0.07, 0.49, 1.98) (0.05, 0.23, 1.16)

Table 7 Fuzzy decision matrix for main criteria Strategies Main criteria Scale drivers (0.14, 0.24, 0.42) Strategy 1 Strategy 2 Strategy 3 (0.05, 0.44, 3.38) (0.04, 0.23, 2.69) (0.04, 0.33, 2.80) Product attr. (0.05, 0.12, 0.25) (0.06, 0.40, 2.34) (0.05, 0.34, 2.11) (0.04, 0.26, 1.74) Platform attr. (0.09, 0.18, 0.33) (0.08, 0.45, 2.48) (0.05, 0.30, 2.04) (0.05, 0.25, 1.65) Project attr. (0.12, 0.18, 0.34) (0.04, 0.45, 2.58) (0.03, 0.26, 2.15) (0.03, 0.29, 2.01) Personnel attr. (0.18, 0.28, 0.36) (0.07, 0.45, 3.19) (0.05, 0.28, 2.69) (0.05, 0.26, 2.52)

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . 43

out by the authors and only the final results have been presented to the participants. The following points will be the natural extension of our work: As we have tested the validity of our methodology with the given application, we are now developing a decision support tool that will provide user-friendly graphical interfaces and hide the necessary computations. Thus, a broader industrial user group will have the opportunity to apply the proposed approach without expert intervention. The financial dimension has not been investigated. In future research, this subject will be primarily considered for different selection phases while keeping in mind the fuzziness of the cash flows. Similar approaches have to be provided to increase the efficiency of other critical decisions in NPD process. We have employed different ranking, selection and multi-criteria decision-making techniques in our work, which we consider the most appropriate. The influence of other methods on the final quality and accuracy of decisions has to be evaluated.

standing to an expression like ‘‘performance close to 90’’, which is vague (fuzzy). A.2. Fuzzy sets and fuzzy numbers Fuzzy sets were first introduced by Zadeh (1965) and are a generalization of crisp sets. The degree of membership of an individual in a fuzzy set expresses the degree of compatibility of the individual with the concept represented by the fuzzy set. Each fuzzy set, A; is defined in terms of a relevant universal set, X ; by a function called a membership function which assigns to each element x of X a number AðxÞ in the closed interval ½0; 1Š: This characterizes the degree of membership of x in A: Therefore, membership functions are of the form mA ðxÞ : X -½0; 1Š: ðA:1Þ

Acknowledgements We thank the anonymous referees for their valuable comments and suggestions, which have improved the final quality of the paper.

Appendix A A.1. Fuzzy set theory Fuzzy analysis is an important tool to represent vagueness and a kind of imprecision and uncertainty. The term fuzzy is meant to represent expressions and judgments that have no clear (crisp) value or boundary. For example, when we express that the performance of the personnel is outstanding, this linguistic expression is fuzzy since it cannot be precisely associated with a real number. However, we can always associate out-

Larger values of mA ðxÞ imply higher degrees of set membership. The fuzzy set A is normal, if supx mA ðxÞ ¼ 1: The fuzzy set A is defined as convex if and only if mA ðlx1 þ ð1 À lÞx2 ÞXminðmA ðx1 Þ; mA ðx1 ÞÞ for all x1 ; x2 ARn and for all lA½0; 1Š: A normal and convex fuzzy set defined on Rn whose membership function is piecewise continuous is called a fuzzy number. In other words, a fuzzy number represents the conception of a set of real numbers close to x; % where x is the number being fuzzified. As an % example, a triangular fuzzy number is expressed as 8 xoa or x > c; > 0; < mA ðxÞ ¼ ðx À aÞ=ðb À aÞ; apxpb; ðA:2Þ > : ðx À cÞ=ðb À cÞ; bpxpc: a-cut of a set. The concept of an a-cut of a fuzzy set is especially useful for the arithmetic operations on fuzzy numbers which will be given next. a-cut of a fuzzy set A is the set Aa of elements x; such that their degree of membership in the set A is at least equal to a (Klir and Yuan, 1995). Therefore, the a-cut is then expressed by Aa ¼ fxAX j mA ðxÞXag: ðA:3Þ

Any fuzzy number can be completely defined by its family of a-cuts.

44 G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . Cooper, R.G., 1983. A process model for industrial new product development. IEEE Transactions on Engineering Management 13, 2–11. Cooper, R.G., 1994. Perspective: Third generation new product processes. Journal of Product Innovation Management 11, 3–14. Cooper, R.G., Edgett, S.J., Kleinschmidt, E.J., 1999. New product portfolio management: Practices and performance. Journal of Product Innovation Management 16 (4), 333–351. Davila, T., 2000. An empirical study on the drivers of management control systems’ design in new product development. Accounting, Organizations and Society 25, 383–409. Doctor, R.N., Newton, D.P., Pearson, A., 2001. Managing uncertainty in research and development. Technovation 21 (2), 79–90. Drucker, P.F., 1999. Management Challenges for the 21st Century. Harper Business, New York. Faulkner, T., 1996. Applying options thinking to R&D valuation. Research and Technology Management 39, 50–57. Fox, J., Gann, R., Shur, A., Glahn, L., Zaas, B., 1998. Process uncertainty: A new dimension for new product development. Engineering Management Journal 10, 19–27. Galbraith, J., 1973. Designing Complex Organizations. Addison-Wesley, Reading, MA. Garavelli, A.C., Gorgoglione, M., Scozzi, B., 1999. Fuzzy logic to improve the robustness of decision support systems under uncertainty. Computers and Industrial Engineering 37, 477–480. Hall, D.L., Naudia, A., 1990. An interactive approach for selecting R&D project. IEEE Transactions on Engineering Management 37, 126–133. Hamel, G., Prahalad, C.K., 1994. Competing for the Future. Harvard Business School Press, Boston, MA. Haque, B., Pawar, K.S., Barson, R.J., 2000. Analysing organisational issues in concurrent new product development. International Journal of Production Economics 67 (2), 169–182. Harter, D.E., Krishnan, M.S., Slaughter, S.A., 2000. Effects of process maturity on quality, cycle time, and effort in software product development. Management Science 46 (4), 451–466. Hillson, D., 2002. Extending the risk process to manage opportunities. International Journal of Project Management 20 (3), 235–240. Infanger, G., 1994. Planning Under Uncertainty. International Thomson Publishing, MA. Jones, T., 1997. New Product Development: An Introduction to a Multifunctional Process. Butterworth, Heinemann, Oxford. Klir, G.J., Yuan, B., 1995. Fuzzy Sets and Fuzzy Logic: Theory and Applications. Prentice-Hall, Englewood Cliffs, NJ. Krishnan, V., Ulrich, K.T., 2001. Product development decisions: A review of the literature. Management Science 47 (1), 1–21.

A.3. Extended algebraic operations Extended operations make use of the extension principle (Zadeh, 1965) and are defined by Klir and Yuan (1995), based on arithmetic on intervals and assuming and assuming that fuzzy numbers are represented by continuous membership functions. The fuzzy set obtained by an arithmetic operation on the fuzzy numbers A1 and A2 ; on R; is defined by its a-cut such that Addition: ðA1 "A2 Þa ¼ Aa þ Aa ; 1 2 ðA1 ~A2 Þa ¼ Aa À Aa ; 1 2


Multiplication: ðA1 #A2 Þa ¼ Aa :Aa ; 1 2 Division: ðA1 +A2 Þa ¼ Aa =Aa 1 2 ðprovided that 0eAa 8aÞ: 2 Here Aa and Aa represent the a-cuts of the fuzzy 1 2 numbers A1 and A2 ; respectively. Then, the family of a-cuts ðA1 }A2 Þ ð} is any of the four arithmetic operations) that is obtained through an arithmetic operation defines the new fuzzy set, which also classifies as a fuzzy number. An arithmetic operation on the fuzzy numbers A1 and A2 is, therefore, reduced to operations on intervals Aa and Aa : The four arithmetic opera1 2 tions on closed intervals are defined as follows (Klir and Yuan, 1995): ½a; bŠ þ ½c; dŠ ¼ ½a þ c; b þ dŠ; ½a; bŠ À ½c; dŠ ¼ ½a À d; b À cŠ; ½a; bŠ:½c; dŠ ¼ ½minfac; ad; bc; bdg; maxfac; ad; bc; bdgŠ; ½a; bŠ=½d; eŠ ¼ ½minfa=c; a=d; b=c; b=dg; maxfa=c; a=d; b=c; b=dgŠ ðprovided that 0e½c; dŠÞ:

Brown, S.L., Eisenhardt, K.M., 1995. Product development: Past research, present findings and future directions. Academy of Management Review 20, 343–378.

G. Buyukozkan, O. Feyz’oglu / Int. J. Production Economics 90 (2004) 27–45 ı * . . . Li, S., 2000. The development of a hybrid intelligent system for developing marketing strategy. Decision Support Systems 27, 395–409. Lysonski, S., Levas, M., Lavenka, N., 1995. Environmental uncertainty and organizational structure: A product management perspective. Journal of Product and Brand Management 4 (3), 7–18. Maffin, D., Braiden, P., 2001. Manufacturing and supplier roles in product development. International Journal of Production Economics 69, 205–213. Markowitz, H., 1952. Portfolio selection. The Journal of Finance 7 (1), 77–91. Martino, J.P., 1995. Research and Development Project Selection. Wiley, New York. Matheson, D., Matheson, J., 1998. The Smart Organization Creating Value Through Strategic R&D. Harvard Business School Press, Boston, MA. McQuater, R.E., Peters, A.J., Dale, B.G., Spring, M., Rogerson, J.H., Rooney, E.M., 1998. The management and organizational context of new product development: Diagnosis and self-assessment. International Journal of Production Economics 55, 121–131. Miller, R., Lessard, D., 2001. Understanding and managing risks in large engineering projects. International Journal of Project Management 19, 437–443. Montoya-Weiss, M.M., Calantone, R.J., 1994. Determinants of new product performance: A review and meta analysis. Journal of Product Innovation Management 11, 397–417. Mullins, J.W., Sutherland, D.J., 1998. New product development in rapidly changing markets: An exploratory study. Journal of Product Innovation Management 15, 224–236. Mullins, J.W., Forlani, D., Walker, O.C., 1999. Effects of organizational and decision-maker factors on new product risk taking. Journal of Product Innovation Management 16, 282–294. Parnell, G.S., Jackson, J.A., Burk, R.C., Lehmkuhl, L.J., Engelbrecht, J.A., 1999. R&D concept decision analysis: Using alternate futures for sensitivity analysis. Journal of Multi-Criteria Decision Analysis 8, 119–127. Riek, R.F., 2001. From experience: Capturing hard-won NPD lessons in checklists. Journal of Product Innovation Management 18, 301–313. Roy, B., Vincke, P.H., 1984. Relational systems of preference with one or more pseudo-criteria: Some new concepts and results. Management Science 30, 1323–1335. Saaty, T.L., 1977. A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology 15, 234–281. 45

Saaty, T.L., 1980. The Analytic Hierarchic Process. McGrawHill International, New York. Schmidt, J.B., Calantone, R.J., 1998. Are really new product development projects harder to shut down? Journal of Product Innovation Management 15 (2), 111–123. Shipley, M.F., de Korvin, A., Khursheed, O., 2001. A fuzzylogic-based decision model to satisfy goals for successful product/service introduction. European Journal of Operational Research 135 (1), 209–219. Smith, P.G., 1999. Managing risk as product development schedules shrink. Research and Technology Management 42, 25–32. Souder, W.E., Mandakovic, T., 1986. R&D project selection models. Research Management 29, 36–42. Spender, J., 1993. Some frontier activities around strategy theorizing. Journal of Management Studies 30 (1), 11–30. Subramanian, D., Pekny, J.F., Reklaitis, G.V., 2001. A simulation-optimization framework for research and development pipeline management. A.I.Ch.E. Journal 47 (10), 2226–2242. Triantaphyllou, E., 2000. Multi-criteria decision making methods: A comparative study. Kluwer Academic Publishers, London. Trittle, G.L., Scriven, E.F.V., Fusfeld, A.R., 2000. Resolving uncertainty in R&D portfolios. Research and Technology Management 43, 47–55. Tseng, T.Y., Klein, C.M., 1989. New algorithm for the ranking procedure in fuzzy decision making. IEEE Transactions on Systems, Man and Cybernetics 19 (5), 1289–1296. Ulrish, K.T., 2001. Introduction to the special issue on design and development. Management Science 47 (1), v–vi. Urban, G.L., Hauser, J.R., 1993. Design and Marketing of New Product, 2nd Edition. Prentice-Hall Inc., Englewood Cliffs, NJ. Vanegas, L.V., 1999. Application of fuzzy set theory in engineering design. Dissertation, University of Manchester Institute of Science and Technology. Vanegas, L.V., Labib, A.W., 2001. Application of new fuzzyweighted average method to engineering design evaluation. International Journal of Production Research 39 (6), 1147–1162. Wang, J., 1997. A fuzzy outranking method for conceptual design evaluation. International Journal of Production Research 35 (4), 995–1010. Zadeh, L.A., 1965. Fuzzy sets. Information and control 8, 338–353. Zahedi, F., 1996. The analytic hierarchy process—a survey of the method and its applications. Interfaces 16, 108.

Sign up to vote on this title
UsefulNot useful