You are on page 1of 17

Expert Systems With Applications 119 (2019) 272–288

Contents lists available at ScienceDirect

Expert Systems With Applications


journal homepage: www.elsevier.com/locate/eswa

A fuzzy-based decision aid method for product deletion of fast moving


consumer goods
Pourya Pourhejazy a, Joseph Sarkis b, Qinghua Zhu a,∗
a
Antai College of Economics & Management, Shanghai Jiao Tong University, 1954 Huashan Road, Xuhui District, Shanghai 200030, China
b
Foisie Business School, Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609-2289, USA

a r t i c l e i n f o a b s t r a c t

Article history: Technological and market cycles pressure organizations to introduce new products. To make room for
Received 3 August 2018 the new products, product deletion (PD) decisions are inevitable. PD implementation, however, is beyond
Revised 4 October 2018
the reactive elimination of mature or low-profit products. As a competitive and proactive managerial
Accepted 1 November 2018
tool, PD requires incorporation of supply chain financial and non-financial attributes to not only benefit
Available online 2 November 2018
organizations, but also maintain important partnerships. To successfully apply strategic PD, consideration
Keywords: of a broader set of decision factors is necessary. To update the traditional view, this paper introduces
Product deletion supply chain and competitive factors to product deletion decision-making (PDDM). Given the complex
Supply chain multi-criteria problem, including both qualitative and quantitative factors and the relative uncertainties
Fast-moving consumer goods and interaction between the factors, a Nested-Fuzzy Inference System with Interactions (NFISI) model as
Multi-criteria decision-making part of a multi-stage, multi-method expert system is introduced to aid the decision-making process. The
Fuzzy inference system
model is verified using a practical case in a major fast-moving consumer goods (FMCG) company. The
priority of alternative products for deletion is obtained through the application of the developed model
considering the newly introduced factors. The results provide initial, albeit idiosyncratic, insights with
a discussion of practical implications. It is demonstrated that considering supply chain factors makes a
difference in PDDM outcomes. A brief overview of research opportunities is finally presented to guide
researchers in contributing to this under-studied subject.
© 2018 Published by Elsevier Ltd.

1. Introduction Consumer goods industries are a venue where the PD deci-


sion has been applied the most (Argouslidis, Baltas, & Mavromma-
Markets are increasingly dynamic as product and technologi- tis, 2015). Given low brand loyalty in fast-moving consumer goods
cal life cycles become shorter (Lin, Chiu, & Chu, 2006). In this (FMCG) (Wu, Liu, & Yang, 2018), and high consumer expectation
environment, strategic product deletion (PD) can be as critical for service quality and new products in this sector (Mbhele, 2016),
to organizational competitiveness as new product introduction pressure for product availability and innovation in FMCG is high
(Bai, Shah, Zhu, & Sarkis, 2017). A paradigm shift from prolifer- (Diehl & Spinler, 2013). Therefore, the decision on which products
ation to rationalization of products has put more emphasis on should be deleted is even more imperative in FMCG markets. FMCG
PD (Argouslidis, Baltas, & Mavrommatis, 2014). Market dynam- companies routinely trim their product lines to maintain high over-
ics and shortened product life cycles have caused greater occur- all performance and competitiveness (Chong, Ho, & Tang, 2004).
rence of PD (Wagner, Abdelkafi, & Blecker, 2017). Diverse reasons The Procter & Gamble (P&G) Corporation is a prime example of
exist for PD decisions. Product underperformance, poor product PD practices, where the company continuously tweaks its product
strategic-fit, or products damaging organizational reputation are portfolio by deleting low-profit products (Ng, 2014).
some general examples for product discontinuation in different Implementation of strategic PD practices, beyond the simple
sectors (Avlonitis, Hart, & Tzokas, 20 0 0). removal of obsolete or mature products is inevitable (Zhu &
Shah, 2018). Strategic PD can be used as a "competitive weapon" in
critical market situations (Wan, Wang, & Fung, 2005). However, PD
can cause failure if it is not applied cautiously (Shah, 2017a, 2017b).
For example, a strategic product may be discontinued merely due

Corresponding author. to its undesirable sales volume, putting important partnerships at
E-mail addresses: pourya@sjtu.edu.cn (P. Pourhejazy), jsarkis@wpi.edu (J. Sarkis), risk (Van Hoek & Pegels, 2006). On the other hand, a burdensome
qhzhu@sjtu.edu.cn (Q. Zhu).

https://doi.org/10.1016/j.eswa.2018.11.001
0957-4174/© 2018 Published by Elsevier Ltd.
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 273

product, which is adversely influencing the company’s overall per- 2. Literature review
formance, may not be discontinued on account of its high marginal
profit. Although PD has traditionally occurred as a routine and re- A review of the literature is conducted using the following key-
active decision, the proactive strategic perspective towards it is a words: "product deletion" OR "product elimination" OR "product
new phenomenon (Muir & Reynolds, 2011). discontinuation" in the Web of Science core database. Through a
The extant literature introduced financial and market consid- search of the aforementioned terms in the title, abstract, and key-
erations for product deletion decision-making (PDDM) (Avlonitis, words in the database, a total of 83 papers were initially identi-
1984, 1993; Mitchell, Taylor, & Tanyel, 1998). Given the evolving fied. After reading the title and abstract of the resulting papers,
importance of PD decisions, however, there is a need for a strategic only 18 journal and conference papers were perceived as relevant,
outlook on this managerial practice. This situation requires consid- with the rest belonging to unrelated fields such as bioengineering
eration of a broader range of operational and competitive factors and chemistry. A database was formed by recording the titles, au-
relating to PD. Additionally, the growing strategic importance of thors, and journals, as well as the year of each publication. In the
supply chains in organizational competitiveness requires that com- next step, the articles were categorized based on the methodol-
panies consider PD’s impact on the supply chain’s overall perfor- ogy. Of the existing literature on PD, the majority (8 papers) are
mance. conceptual; empirical studies and surveys followed, with 4 and 2
Due to shortened product life cycles and the need for rapid papers, respectively. Finally, a list of decision factors was identified
product development, the significance of PD and new product in- through a more comprehensive review. The identified factors were
troduction, together known as product rollover, becomes more ap- later presented to the experts of the case company to ensure full
parent (Katana, Eriksson, Hilletofth, & Eriksson, 2017). However, coverage of the relevant supply chain and the competitive factors.
research on PD, unlike other product management decision ar- On this basis, some of the factors were modified, and a new rec-
eas, is limited (Argouslidis et al., 2015; Bai et al., 2017; Shah, ommended factor, high variance in the customer’s feedback, was
Zhu, & Sarkis, 2017). Despite the interdependence among PD de- added to the list. Discussion of the results is now presented in this
cisions and supply chain operations, few studies cover both (Zhu & background review.
Shah, 2018); the organizational aspects of PD decisions have been
neglected (Argouslidis et al., 2015). No studies have jointly exam-
ined the effect of competitive forces and operational aspects on 2.1. Background on product deletion
PDDM. Moreover, the vast majority of PD studies are conceptual
in nature (Muir & Reynolds, 2011) and decision-making tools need Product elimination, deletion, and discontinuation are used in-
further development and to be tested in real-world cases. terchangeably in the academic literature to refer to the decision of
After selection of initial candidates for deletion, the final deci- stopping manufacture and sale of a specific product, brand, or SKU,
sion would benefit from a systemic ranking scheme. Given these from a company’s portfolio (Avlonitis & Argouslidis, 2012; Shah,
broad considerations for strategic PD, expert systems can support Laverie, & Davis, 2017). PD decisions can take the form of either
a systemic perspective, providing decision-makers with dependable rollovers (i.e. phasing out existing products and phasing in new
means for identifying the effective solutions to complex problems ones) or complete deletion where there is no replacement for the
(Liao, 2005). Such a decision aid method is particularly essential in deleted product (Argouslidis et al., 2015). It is a multi-functional
retail FMCG, because of the large number of stock keeping units managerial practice that can be initiated by a manufacturer, re-
(SKUs) to be managed (Joseph, Sundarakani, Hosie, & Nagarajan, tailer, or wholesaler. Marketing, accounting, and finance functions
2010), and the fact that PD requiring rapid audits and decision are typically more actively involved in the decision process at an
schemes occurs on a regular basis. To overcome the imprecision organizational level (Muir & Reynolds, 2011).
and uncertainty of this evaluation, applications of fuzzy approaches The PD process consists of a sequence of steps beginning with
are needed; vagueness has been neglected in PDDM. Moreover, for the identification of candidates for deletion, followed by an anal-
practical decision-making problems, many factors are interactive, ysis on possibilities of revitalization, and finally the evaluation of
and existing fuzzy-based modeling aggregators do not take inter- candidates for eventual PD decision-making and implementation
actions into account (Amindoust, Ahmed, Saghafinia, & Bahreinine- (Avlonitis & Argouslidis, 2012). Our study focuses on the final step
jad, 2012; Ramkumar, 2016; Ramkumar, Schoenherr, & Jenamani, of this PD process.
2016). Due to resource limitations, companies with full product lines
Given the above-mentioned concerns, this paper aims to make consider deleting existing products to free capacity for the launch
two key contributions to the under-researched area of PDDM. We of new products (Lin & Shih, 2010). PD is particularly essential to
first aim to put forward a comprehensive set of criteria by in- industries with products characterized by short life cycles and time
troducing supply chain factors and competitive forces. Secondly, a to market (Katana et al., 2017). Given the variety of products and
nested-fuzzy inference system with interactions (NFISI) model as the large assortment size in retail FMCG (Joseph et al., 2010), PD
part of a multi-stage, multi-method approach is then developed to has been applied frequently in this sector (Argouslidis et al., 2015).
aid the decision-making process. The other methods, including the PD, as a reactive managerial practice, has been predominantly
best-worst method and a multi-attribute method, “Measuring At- initiated by weak sales performance or poor market share and
tractiveness by a Categorical Based Evaluation Technique” (MAC- profitability (Vyas, 1993). Given today’s market complexities, PD
BETH), are integrated to allow for the application of NFISI. A prac- has necessarily evolved into a proactive and strategic managerial
tical application from one of the world’s top multinational compa- tool to cope with market dynamics, i.e. rationalizing the product
nies in FMCG is finally used as an illustrative case. portfolio (Leiter, 2011) or downsizing the business (Harness & Har-
The remainder of the article begins with identifying factors for ness, 2012). When analyzing PD as a strategic decision, there is
PD. The methodological tools are then introduced in Section Three. a need to thoroughly take into account the multi-faceted opera-
The illustrative case study, accompanied by a numerical analysis tional factors (Irani, Ghoneim, & Love, 2006; Sarkis & Talluri, 2002).
and a discussion on managerial implications, is subsequently pro- Beyond traditional financial factors, this study extends to include
vided to show the applicability and limitations of the proposed supply chain operational and competitive aspects. Table 1 presents
model. The paper concludes with some insights for future research a list of a comprehensive set of PD decision factors and their def-
to help address some of the limitations and challenges associated initions. Each of the major categories is summarized in the next
with PD and supply chain management. sub-section.
274 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Table 1
Summary of factors associated with product deletion decision.

Category Sym. Factor Definition Supporting References

Traditional C1 Low total profit A sharp drop in the realized sales level (turnover) and unit profit Avlonitis (1993),
Factors compared to the expected level Avlonitis (1984),
Avlonitis and
Argouslidis (2012),
Avlonitis et al. (20 0 0),
Mitchell et al. (1998)
C2 High level of allocated Resource intensive products potentially degrade the supply chain Avlonitis (1993),
supply chain performance Avlonitis (1984),
resources Avlonitis and
Argouslidis (2012),
Avlonitis et al. (20 0 0),
Mitchell et al. (1998),
Wu et al. (2018)
C3 Strategic-fit Inconsistency with the company’s business strategy i.e. cost Avlonitis (1993),
contradiction effectiveness, responsiveness, and differentiation Avlonitis (1984),
Avlonitis and
Argouslidis (2012),
Avlonitis et al. (20 0 0),
Mitchell et al., (1998)
C4 Negative customer Negative customer’s perception towards a product about its quality and Argouslidis et al. (2015),
feedback healthiness is a major reason for reformulation or discontinuation of Mitchell et al. (1998),
the product Wu et al. (2018)
C5 High variance in This factor is important in business-to-business, where business Recommended by the
customer feedback entities receive the same product, but due to different ways of interviewee
processing them (i.e. recipe for the restaurants), the feedback on the
product’s quality can be different
Supply C6 High logistic expenses It comprises the share of the product’s logistics costs comprising Wu et al. (2018)
Chain packaging, transportation, and inventory keeping expenses to the Birhanu et al. (2017),
Factors total corresponding values in the company’s supply chain Argouslidis et al. (2015)
C7 Logistic resources In case of a recent launch of a new product or a surge in demand for Zhu and Shah (2018)
reallocation another product, the released logistic resources can be reallocated
opportunities after a possible deletion
C8 Average time in the Lengthy inventory days of supply, particularly in sourcing and stocking Argouslidis et al. (2015),
supply chain processes and long return-on-capital Kocaoğlu et al. (2013)
Dollet and Díaz (2011)
C9 Low flexibility in Product assortment complexity may cause inflexibility in logistic Harness and Harness (2012)
company logistics activities and weaken the timeliness of supply
operations
C10 High percentage of Outsourcing can hinder deletion due to contractual terms (possibilities Naim et al. (2006)
insourced contents of early termination). Moreover, more resources will be released in
case of deleting a product with a high percentage of insourced
activities
C11 Low effectiveness of Sourcing effectiveness goes beyond the financial aspects and includes Fernie and Sparks (2014)
sourcing intangible dimensions like supplier relations and value procurement,
that reflects the performance of the suppliers in terms of quality and
service level
Competitive C12 Existing competitor When the product’s market is shrinking and there is no more room for Porter (2008), Wu et al. (2018)
Forces rivalry competition on price and quality
C13 New market entrant Newcomers to an industry change the balance between supply and Porter (2008)
potential demand. This can cause further shrinkage of profit margin and
increase rivalry
C14 Customer bargaining High price-sensitivity of FMCGs customers and availability of wide Nielson (1998), Porter (2008)
power range of options make it easy to switch to another product
C15 Substitute product Substitute product can be an alternative that belongs to other product Homburg et al. (2010),
availability categories/markets but offers the similar functionality Porter (2008)

2.2. Factors for product deletion siderably low contribution of a single product in FMCG (Wu et al.,
2018).
2.2.1. Traditional factors Having intensively allocated resources in the production process
Early works cited financial, resource-based, and strategic as- is another major aspect that impacts the PD decision. Considering
pects influencing the PD decision (Avlonitis, 1993; Avlonitis, 1984; the critical role of logistics in FMCG (Wu et al., 2018), the level of
Avlonitis & Argouslidis, 2012). Total profit (C1 ) is considered to rep- allocated supply chain resources (C2) is considered to represent the
resent the sales volume and unit profit of the product. The work- resource-based factors within an updated view, and in the con-
ing capital associated with the manufacture of a product is an- text of FMCG. Weak products occupy the company’s valuable re-
other major financial aspect. This factor is significantly affected by sources and increase the complexity of the operational processes,
the supply chain operational aspects, particularly inventory perfor- and discontinuing them helps to improve organizational and sup-
mance (i.e. day sales of inventory), as well as the difference be- ply chain slacks, and in some cases overall profitability (Bayus &
tween trade receivables/payables (Brandenburg & Seuring, 2011). Putsis, 1999). The increase in availability of time and capital, com-
The former component is discussed among the supply chain op- prising the human and non-human resources, are the major oppor-
erational factors, while the latter is not included due to the con- tunities of PD relating to this group of factors.
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 275

From the strategic perspective, the product’s effect on the cor- Ellram, 1993; Sellitto, Pereira, Borchardt, da Silva, & Viegas, 2015).
porate image and the company’s competitive modes are the other PD can potentially create slack capacity in logistics infrastructure,
key considerations in PD. The PD decision has a significant impact such as in warehouses and transportation vehicles, which in turn
on the corporate image (Argouslidis et al., 2015). This effect can results in wasted energy and resources (Zhu & Shah, 2018). Logis-
be positive if PD is strategically applied to proactively avoid the tics resources reallocation opportunities (C7 ) can rectify this concern
adverse effects of a product on the company’s reputation, i.e. non- and mitigate logistics costs.
sustainable supply chain operations pertaining to the product. On Time: Speed is a principal attribute in supply chains of FMCG
the other hand, PD may not be favorable if it is putting important (Dewa, Pujawan, & Vanany, 2017). Delivery time required by
partnerships at risk or causing customer dissatisfaction due to pop- business-to-business customers and timeliness of that delivery are
ularity of the product. Besides, the more an inconsistent a prod- important to supply chain performance. Besides, FMCG compa-
uct is with the company’s strategies, the more likely the decision nies strive for short-term return on capital (Dollet & Díaz, 2011).
for discontinuation. This factor is therefore referred to as strategic- The days of inventory, or average time in the supply chain (C8 )
fit contradiction (C3 ), which is a general, traditional factor and in- (Kocaoğlu, Gülsün, & Tanyaş, 2013) is chosen to represent the time
cludes the full-line strategy. attribute. This measure focuses on the time a product spends in
Besides the aforementioned factors from the early studies in the supply chain processes before being sold, and can directly in-
the field, the product’s performance is of high significance in PD fluence the inventory and transportation costs. Depending on the
(Mitchell et al., 1998). Given the high consumer expectation for type of product, the modification time can be considered a sepa-
quality in FMCG (Mbhele, 2016), perceived product effectiveness rate factor (Sellitto et al., 2015).
from the viewpoints of customers (Berry, Burton, & Howlett, 2017; Flexibility: FMCG is characterized by high product diversity and
Brown & Dacin, 1997) is imperative in a PD context. Therefore, neg- frequent deliveries of small-batches (Colicchia, Creazza, & Dal-
ative customer feedback (C4 ) on the product can be a major reason lari, 2017). These features have made the supply chain role more
for the modification or discontinuation of a product. High variance prominent in FMCG (Wu et al., 2018), so that FMCG is the largest
in customer feedback (C5 ) on product quality is the other relevant logistics outsourcer industry (Dev, Shankar, Gunasekaran, & Thakur,
factor missing in the literature, and plays a significant role, partic- 2016). This indicates the increasing importance of the flexibility
ularly in business-to-business sector of FMCG. This factor demon- of logistics operations (C9 ) in FMCG (Manders, Caniëls, & Ghijsen,
strates the need for broadening a company’s product portfolio; if 2016), which can be affected by the product portfolio variety and
this need cannot be met, PD should be applied to avoid adverse average order lot size (Sellitto et al., 2015). PD improves flexibility
effects on the company’s image. by having fewer products that require logistics activities (Harness
Owing to the fact that the contribution of a single product in & Harness, 2012). From a sourcing point of view, maintaining in-
FMCG is considerably low (Wu et al., 2018), it is imperative to house capability can help to build flexibility (Naim, Potter, Ma-
view individual products as members of a larger group. The fol- son, & Bateman, 2006). For this purpose, making excess capac-
lowing sub-sections, including supply chain operational factors and ity through PD may be useful, particularly to cope with capac-
broader dimensions of competitive forces, are elaborated upon to ity shortfalls in periods of supply/demand mismatch and improve
incorporate this viewpoint into PDDM. supply chain resilience (Pourhejazy, Kwon, Chang, & Park, 2017).
A poorly performing product with a high percentage of insourced
2.2.2. Supply chain operational factors content (C10 ) is, therefore, more likely to be deleted than a product
The greater variety and larger quantity of products that a sup- with a high percentages of outsourced parts/services. This situation
ply chain accommodates may lead to degradation of operational is due to two major reasons: (1) less complexity in governance due
performance, due to resource and capacity constraints. Product to contracts with suppliers; (2) freeing higher amounts of supply
and supply chain planning should be completed simultaneously chain resources through PD.
(Gokhana, Needy, & Normanc, 2015). Product modification, dele- Quality: The quality attribute contributes the most to FMCG
tion, and new product introduction may have potential to improve companies and their decision-making (Diehl & Spinler, 2013).
overall system performance (Garbi & Loureiro, 2014). To success- Given the low brand loyalty in FMCG (Wu et al., 2018), the
fully apply these product management practices, their potential product quality attribute is pivotal. Additionally, intense compe-
effects should be properly managed throughout the supply chain tition in FMCG obliges the companies to focus on improving
(Katana et al., 2017). Therefore, PDDM should take into considera- quality, in addition to rigorous cost control strategies (Fernie &
tion the company’s supply chain processes. Sparks, 2014). Companies with strategic sourcing emphasize qual-
To analyze the supply chain decision factors in a PD context, ity rather than price in the sourcing of parts and raw materials
the logistics elements of facilities, inventory, and transportation, (Lo & Yeung, 2004). Poor supplier performance within the supply
as well as sourcing (Chopra, 2007), are considered with respect chain dramatically influences the core competencies of the com-
to cost, quality, time (delivery), and flexibility dimensions, as sug- pany, both in terms of quality and time. From the supply chain per-
gested by De Toni and Tonchia (2001) and Sansone, Hilletofth, and spective, therefore, low effectiveness of sourcing (C11 ) can represent
Eriksson (2017). These dimensions are critical to both supply chain the quality attribute within a PD decision context. Alternatively,
and internal managers of other departments, all of whom are ex- the perceived product effectiveness from the stakeholder viewpoint
pected to be involved in PDDM. Supply chain emphasis and charac- (Kirchoff, Koch, & Satinover Nichols, 2011) can contribute to a PD
teristics differ across industries; this implies that factors and their decision.
relative importance may vary. FMCG includes a wide range of SKUs
with short life and fast, daily consumption (Pan & Choi, 2016), 2.2.3. Competitive forces
where the supply chain’s role is critical (Wu et al., 2018). A poorly Competitive forces are central to organizational and supply
performing product can be considered for deletion because of chain strategy (Porter, 2008). Given that strategic PD can be ap-
its negative impact on supply chain performance, demonstrated plied as a "competitive weapon" (Wan et al., 2005), the inclusion of
within the following attributes: competitive forces is important; this is particularly the case in mar-
Cost: Given the low profit margins in FMCG (Birhanu, Krish- kets like FMCG, which is characterized by highly intense competi-
nanand, & Rao, 2017), logistics expenses (C6 ) are particularly im- tion (Wu et al., 2018). To account for this aspect, existing competi-
perative to the FMCG sector (Wu et al., 2018). This factor includes tor rivalry (C12 ) is the first considered competitive decision factor.
transportation, facilities, and inventory cost elements (Cooper & Knowing a market’s rivalry status and predicting the competitors’
276 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

next move can significantly affect a PD decision. Besides, deletion Tatari, 2016). Expert systems, therefore, can be very helpful in ar-
of a product makes a market more attractive for aspiring entrants. riving at a reliable solution to cope with complexities in prac-
A new entrant’s identifying the potential gap in the market of that tical decision-making problems (Amirkhan, Didehkhani, Khalili-
specific product can result in market losses of other products in Damghani, & Hafezalkotob, 2018). Fuzzy expert systems use fuzzy
the product portfolio. PD may not be favorable when the new mar- set theory (Zadeh, 1965, 1973) to alleviate the aforementioned is-
ket entrant potential (C13 ) is high. sues by allowing the expert system to behave less precisely and
Customer bargaining power (C14 ) is the next relevant competi- more flexibly than conventional approaches (Liao, 2005). Various
tive force. Powerful customers pressuring lower product price or tools of this type have been developed, amongst which Mamdani’s
demanding higher quality in the product can limit product prof- FIS (Mamdani, 1974) can be used as a part of the decision sup-
itability (Porter, 2008). This situation is especially true in the FMCG port system to interpret conditional statements and to guide man-
business-to-business environment, where companies are usually agerial decision-making. FIS has been widely applied in model-
responsible for a limited number of customers (Nielson, 1998). ing decision-maker preferences in multi-criteria decision-making
Therefore, customer bargaining power may influence the PD de- (MCDM). It can help to overcome complexity, subjectivity, and un-
cision. On the supply side, product deletion can result in supplier certainty, and improve the decision-making process. However, it
dissatisfaction. In this case, a supplier may try to compensate for has certain limitations, such as ignoring interactions amongst vari-
revenue loss by charging higher prices or lowering product quality. ables. We address some of these concerns by linking FIS to inter-
The power of suppliers is especially pertinent in industries where action effects, which is supported using MACBETH.
supplier monopolies are prevalent. With multiple decision factors contributing to the decision pro-
Product substitution ease, either through similar products from cess, the factor weight assignment is central to MCDM outcomes
other markets or competitors (e.g., aluminum parts can be substi- (Ordoobadi, 2009). Practically, making direct judgments on fac-
tuted with plastic equivalents), is also a competitive concern. Avail- tor importance becomes less reliable as more factors are intro-
ability of alternatives within the company’s product portfolio im- duced. Many methods utilize interactive question approaches be-
pacts a manager’s PD decision (Homburg, Fürst, & Prigge, 2010). tween decision-makers to determine importance weights. These
Given substitute product availability (C15 ) within a company’s port- interactive, human judgment-based approaches can easily result in
folio, lost sales from PD can be compensated; otherwise, the threat weight inconsistencies (Grabisch & Labreuche, 2016). Inconsistency
of substitutes in the market may hinder the PD decision. These fac- in evaluation has remained a critical challenge in the MCDM field
tors will play a central role in the decision methodology for PD. (Herman & Koczkodaj, 1996). As an example, FIS models for rank-
The developed PD decision methodology, utilizing three analyt- ing of alternative suppliers have been applied using decision fac-
ical techniques including the best-worst method, MACBETH, and tor weights based on weak, moderate, strong, and extreme impor-
NFISI is introduced in the following section. The most relevant tance expressions, obtained through direct and interactive partici-
fuzzy inference system (FIS)-based decision models in the litera- pant measure (Amindoust et al., 2012). Relying on direct manage-
ture are reviewed in the next sub-section. rial judgment may lead to arbitrary and misleading assignment of
weights; having tools to help make this process more precise is
3. Methodology valuable (Bana E Costa, 2012).
Multiple complementary methods have been integrated with
3.1. Fuzzy inference systems and MACBETH FIS to overcome these deficiencies. The analytic hierarchy process
(AHP) has been a popular technique for factor weighting as an in-
FIS involves a series of computational processes for evaluation put to FIS (Abbasimehr & Tarokh, 2017; Singh, Olugu, Musa, & Ma-
and overall interpretation of linguistic judgments across different hat, 2018; Tavana et al., 2017). AHP has been limited in that it
decision factors so that{{input1 , ..., inputN }|rule − base −
→ out put }. cannot capture the interrelations among the factors. The analyti-
The algorithm consists of a fuzzifier that transforms crisp inputs cal network process (ANP) is used to address the interdependency
to fuzzy counterparts, using membership functions. An inference of factors as a concern (Ramkumar, 2016; Ramkumar et al., 2016).
engine processes the data inputs using a set of pre-defined set of Even with the advantages of ANP, the interactions among the fac-
conditions – the rule-base to generate fuzzy outputs. The rule-base tors and their effect on decision-making processes cannot be fully
is central to the inference system. The inference system provides a captured.
decision algorithm using inference compositional rules (Bashiri & Interactions in decision models call for inclusion of both indi-
Hosseininezhad, 2009). The defuzzifier estimates a crisp numerical vidual decision factor weights and weights of joint influences of
value to the fuzzy output. This crisp value will be utilized for the factors (Tan & Chen, 2010). Interaction effects can provide more
ultimate decision. appropriate weighting of factors when there are ‘superadditivity’,
MACBETH provides a systematic approach to obtain the deci- synergy, and ‘subadditivity’, redundancy or negative synergy, con-
sion model parameters through pairwise non-numeric judgments cerns (Grabisch, 1996).
of differences between decision-making scenarios (Bana E Costa, & Taking the PD case as a practical example, a simultaneously un-
Vansnick, 1994). Cardinal, categorical data, is utilized to help ad- satisfactory score of a pair of decision factors with a positive in-
dress some concerns of vagueness and intangibility of data and is teraction causes greater importance than the sum of single fac-
the basis for MACBETH (Craig, Busschbach, & Salomon, 2009). The tors alone, especially at the aggregate, accumulated influence, level.
technique supports pair-wise comparisons for PD likelihood. This For example, the relationship between factors "low total profit of a
likelihood is qualitatively evaluated, using linguistic categories. product" and "low logistics flexibility triggered by a product’s op-
erational needs" is a good example of a synergistic weighting sit-
3.2. Existing FIS models uation. In this case, the importance of the two factors occurring
simultaneously is greater than if each influence is considered sep-
PDDM incorporates many factors and requires a broader man- arately to make the PD decision. For example, a manager would
agerial perspective. Much of this environment, due to lack of data, see that if a product performs badly at one of these aspects sepa-
intangible managerial aspects, and the relative complexity of inter- rately, they may view it with much less concern, but together, the
factor relationships, can cause imprecision and vagueness. These concern becomes well above the concern of each handled sepa-
are major challenges when developing real-world decision mod- rately. In this case, due to the complementary nature of these two
els using managerial and expert opinion (Gumus, Kucukvar, & measures with one focused on cost measures and the other on op-
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 277

erational measures, two organizationally different metrics, a syner- factors are included (e.g., qualitative vs. quantitative), for exam-
gistic effect occurs. ple, financial factors are quantifiable measures whereas the sup-
Alternatively, a score for two decision factors with a redundant, ply chain attributes like flexibility and quality, and the competitive
subadditive, negative interaction lessens the respective importance forces are categorized as unquantifiable measures. Different factors
weight of the combination of these two factors. An example of this can be associated with different measurement units. Without some
occurs between the salient factors of "high variance in customer transformation or normalization, these differences make the aggre-
feedback" and "negative customer feedback" on products. Since gation procedure unreasonable.
there is a strong relationship and redundancy amongst these cus- In order to tackle these measure differences, transforming ver-
tomer feedback factors, a reduction of their accumulated influence bal judgment decision maker measures with respect to each crite-
on the decision outcome is warranted. This subadditive reduction rion into a relative commensurate score of alternatives is essential
is needed so as not to excessively weight the customer feedback for the MCDM process (Clivillé & Berrah, 2012). This transforma-
dimension at the expense of other less redundant, more synergistic tion is particularly important for developing general fuzzy rules for
and informative, factors. Essentially, the argument is that too much step-by-step aggregation of heterogeneous decision factors within
redundancy may double-count some factors and that this double the NFISI model. The utility function from MACBETH maps the ac-
counting needs to be discounted. tual alternative performance to satisfaction/dissatisfaction degree
The FIS models in the literature merely considered impor- with respect to each criterion and in the form of an [0,1] interval
tance weights of factors, and none of them employ factor inter- scale (see Bana E Costa (2012)).
action considerations. This is particularly important when defining The resulting vector of elementary scores for a factor i of prod-
generic fuzzy rules for a Nested-FIS. This study sees these inter- uct k– (ESik ) will be used as inputs to execute NFISI in Step 4 (and
actions as necessary to fully grasp relative factor importance and the illustrative application). TheseESik are basic product scores on
effectively model the decision-making environment. A systematic each factor without any factor interaction effect consideration, set-
mechanism linked to FIS can be used to address these factor inter- ting a baseline for each product’s value, in terms of likelihood of
action concerns more effectively. We propose using MACBETH as deletion; which could be considered how unsatisfied managers are
a complementary methodology with a novel Nested-FIS technique. with product characteristics making them more likely to delete a
The proposed method differs from existing relevant approaches for product.
multi-criteria decision-making due to the following points: first of The next major step of MACBETH takes the qualitative judgment
all, intuitionistic fuzzy set theory is used rather than fuzzy set the- about the difference between randomly generated pairs of scenar-
ory; secondly, the phenomenon of interaction among the decision ios determining PD likelihood. This step determines an estimation
factors is considered, which makes it more feasible and practi- for the direct and interaction factor utilities along with the impor-
cal than other traditional aggregation operators for real decision- tance weights. This comparison is completed by offering competing
making problems. The proposed combination makes the FIS model products with varying salient factor levels generated for compara-
more suitable for application as a strategic decision aid method in tive purposes.
PD. It can also be generalized to other complex decision domains, MACBETH traditionally uses a weighted average method (Bana E
where interactions among factors are significant. Costa et al., 1994). However, if the [un]satisfactory level of one fac-
tor influences other factor valuations, there is interaction among
3.3. The decision aid method the decision factors. In this situation, conventional aggregation
methods, such as summing simple weighted average methods, are
The methodology consists of a number of steps and stages. not suitable for modeling a decision maker’s preferences. Con-
Overall, the process includes identifying factors; determining de- ventional aggregators cannot distinguish the difference between
cision factors and alternative values; structuring and executing the the decisions of a strict and non-strict decision maker and do
NFISI network; arriving at overall alternative scores and prioritiz- not reflect exceptional circumstances (e.g., veto power). For this
ing them accordingly. Fig. 1 shows the major steps, sub-steps, and reason, MACBETH was later extended (Cliville, Berrah, & Mauris,
requirements of the decision aid method. These steps are now clar- 2007) to include the Choquet Integral (CI) integrating factor inter-
ified. action effects. It has been applied for performance measurement
in the supply chain context (Clivillé & Berrah, 2012), among other
Step 1. Identification and Filtering of the Pertinent Factors
fields of study. CI enables considering interactions among the de-
As an initiation point, the pertinent factors identified in the lit- cision factors (Grabisch & Labreuche, 2016). The interaction effect
erature research are presented to the decision maker to help iden- amongst a pair of decision factors can be defined using Eq. (1).
tify the most relevant, salient factors for further analysis. Filter-
Ii j = μ({i, j} ) − (μ({i} ) + μ({ j} ) ) + μ(Ø ) (1)
ing of salient factors through reduction from a larger set of fac-
tors may be necessary to make the MCDM model practicable (Dou This equation reflects the utility value difference between the
& Sarkis, 2010). In order to focus on the most important factors case where both of the decision factors are at a satisfactory
when dealing with managerial problems (Koch, 2011), a systematic levelμ({i, j}) and the cases where at most one factor performs
process of factors filtering is needed. For this purpose, the best- satisfactorilyμ({i}) + μ({j}). The final term in expression (1) reflects
worst method developed by (Rezaei, 2015) is preferred to direct the case where neither of the decision factors performs satisfac-
judgment of the expert over salience of the decision factors. This torily. Therefore, Iij is an average marginal interaction value be-
method requires fewer comparison data compared to the other ap- tween factors i and j. This value will be positive if the deci-
proaches and also guarantees consistency in the comparison ma- sion maker is strict, which means the decision maker demands
trix. The method is straightforward; both in terms of data collec- satisfactory performance for all the decision factors in order for
tion and calculation (see Appendix A). A list of salient factors (SFx ) the alternative to having an agreeable overall performance (that
will be the outcome of this step. is, μ({i}) = μ({j}) = 0, where a 0 value for one of the factors re-
flects overall dissatisfaction). Alternatively, the interaction is neg-
Step 2. Determine Decision Factor Weights and Interaction Values
ative if the decision maker is non-strict, loose, and a satisfactory
Having a precise numerical measure for all decision factors is level of only one factor is enough for overall satisfaction (that is,
not always possible, particularly when the quick perceptual evalu- μ({i}) = μ({j}) = 1, where a 1 value for one of the factors reflects
ation of many SKUs is the case. Usually, many types of decision the overall satisfaction). For a negligible interaction, the overall in-
278 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Fig. 1. The 4-step decision aid method.

teraction value is equal to the summation of individual contribu- Vansnick, 1997). For example, the scenarioP1(0,0,1) is a three decision
tions, that is μ({i, j}) = μ({i}) + μ({j}), which is a simple additive factor case for product 1, where the third factor is set at a high
measure (Grabisch, Kojadinovic, & Meyer, 2008). level. If the PD decision in scenario 1 is perceived as "extremely
In order to determine the importance coefficients of decision likelier" (or more preferred) compared to the second product sce-
factors and the model, the interaction amongst the decision fac- nario,P2(0,1,0) , the respective judgment can be translated to the fol-
tors the 2-additive CI model is considered. In this case, the inter- lowing equation (from expression (2)):
action between a pair of decision factors at a time will be cap-
1 1
P1(
0,0,1 )
− P2(
tured to keep a reasonable balance between the complexity of ap- 0,1,0 )
= v3 − [I13 + I23 ] − v2 + [I12 + I23 ]
plying the approach and preciseness of the outcome. In so doing, 2 2
parameters must be determined through a combination of quali- 1
= v3 − v2 − [I13 − I12 ] = 6α
tative judgments and solving the following optimization problem. 2
n
Alpha (α ) is a virtual parameter to ensure that i=1 vi = 1.
 1   
n n
PCI = vi pi − Ii j pi − p j  (2) In order to determine the model parameters as well asα value,
2 n (n+1 )
i=1 i=1 2 + 1 such scenarios comparisons are necessary to determine
Subject to: the decision parameters, where n is the number of decision criteria
(Grabisch et al., 2008). Additional comparisons in this step of MAC-
1   
n
vi − Ii j ≥ 0, ∀i ∈ [1, n], j = i (3) BETH may increase accuracy (Bana E Costa, Lourenço, Chagas, &
2 e Costa, 2008), but makes the data collection procedure lengthier.
j=1
For example, assuming 3 criteria, decision parameters, and the al-
In this set of equations,pi represents the vector of binary val-
pha values can be determined through the construction of 7 com-
ues comprising [ES1K ,…, ESi K ] for a generic product k, obtained
parative equations. This study adopted the approach developed by
through the first step of applying MACBETH; vi andIij represent the
(Clivillé & Berrah, 2012). The procedure is further explained in the
importance weight for factor i, and the interaction value between
case study illustration.
factors i and j, respectively.
In the next sub-step of Step 2, decision makers make a set of Step 3. Structure the NFISI Decision Network
comparisons among the pair of randomly generated object scenar-
ios using verbal levels of attractiveness (very weak, weak, mod- The decision network (overall module) is formed using two
erate, strong, very strong, extreme) adapted from (Bana E Costa, & types of basic “sub-modules” (see Fig. 2). We have defined these
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 279

Fig. 2. General structure of the Nested-FIS with Interactions (NFISI).

as sub-modules of type-A (FIS-A) and type-B (FIS-B). This study Step 4. Execute NFISI for All Alternatives (products considered for
employs a fuzzy rule-base, instead of simple weighted average ag- deletion)
gregators, to estimate the relativity indicator between two factors
i and j (RIi,j ), which is completed using sub-module type-A (FIS-A). To obtain the overall score for a product k (OSk ), the first group
RIi,j is a measure of relativity among a pair of factors that simul- of inputs is the relativity indicator which is required for each of
taneously accounts for the possible interaction among the criteria the aggregating sub-modules (type-B) in the NFISI structure. For
and the inherent uncertainty in the importance weight differences. this purpose, the sub-module type-A is designed to capture the ef-
Step 4 details their determination through a fuzzy rule base sys- fect of interactions among the criteria at every stage of NFISI. This
tem. relativity valuation depends on the factor importance weight (vi )
Sub-module type-B (FIS-B) is a triple-input, single-output FIS. and its interaction valuation (Ii,j ) from the MACBETH steps to de-
The triple inputs for sub-module type-B include: theRIi,j output of termine the RIi,j at layer one. The accumulated interaction among
sub-module type-A, and the competing ESik and ESjk generated in criteria pairs is considered at the other layers of the decision net-
Step 2 for the two factors that are to be aggregated for a given work, which is constructed as a hierarchical tournament compar-
product k. The aggregated score for product k from FIS sub-module ison setting, as described in the previous step. The resulting RIi,j
B-x is represented by ASxk ; it serves as the inputs for the FIS-B at along with the ESik will be used as input to the FIS-B. In order
the next layer, which aggregates scores further. to ensure a general decision network –the NFISI, the input factors
Finally, OSk is the overall aggregated score for product k and have to be sorted. In so doing, the first input is always the criterion
results from FIS B-(N-1) as the final sub-module of NFISI. The al- with larger importance. It is also important to note that inputs in
gorithm, coded in MATLAB software, includes a number of compu- this study are defined in a way that a larger value increases the
tational layers each of which comprises at least two sub-modules, likelihood of the PD decision.
and the total number of sub-module runs depends on the number The rule-base of sub-modules FIS-A and FIS-B are generic and
of decision factors. If there are an odd number of decision factors, defined based on general understanding of aggregation at the sub-
nov niv
the remaining input at the first layer can be considered as the out- module level; it requires a total of i=1 ( j=1 L j ) linguistic con-
put of this layer and be integrated for the aggregated score in the ditions; wherenovis the number of outputs defined for one set of
next layer. input variables,nivis the number of input variables, andLj represents
the number of defined linguistic terms for the input variables
280 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Table 2 Table 5
Membership functions of FIS type-A first input (the in- Fuzzy rule-base for determination of RIi,j .
teraction value Ii,j ). ∼
The difference in importance weights ( v ) Interaction (Ii,j )
i, j
Linguistic term Membership function parameters
Negative Negligible Positive
Negative (−1,−1,−0.001,0)
Negligible (−0.0 01,0,0,0.0 01) Very weakly different SL NS MS
Positive (0,0.001,1,1) Weakly different SL SS HS
Moderately different NS SS HS
Strongly different NS MS ES
Table 3
Very strongly different SS MS ES
Membership functions of FIS type-A second input (the factor

weight difference  v ).
i, j

Linguistic term Membership function parameters Table 6


Linguistic terms for the inputs of FIS-B (the elementary
Very weakly different (0, 0, 0.1, 0.2) scores ESik ).
Weakly different (1, 0.25, 0.25, 0.4)
Moderately different (0.3, 0.5, 0.5, 0.7) Elementary term∗ Membership function parameters
Strongly different (0.6, 0.75, 0.75, 0.9)
Negligible (0, 0, 1, 2)
Very strongly different (0.8, 0.9, 1, 1)
Low (1, 2.5, 2.5, 4)
Moderate (3, 5, 5, 7)
Table 4 High (6, 7.5, 7.5, 9)
Membership functions of FIS type-A output (the relativity index Extreme (8, 9, 10, 10)
RIi,j ). ∗
Smaller-is-better
Linguistic term Membership function parameters

Slightly loose (SL) (0,0,1,2) Table 7


Non-strict (NS) (1,2.5,2.5,4) Linguistic terms for the output of FIS-B (the aggregated score ASxk ).
Slightly strict (SS) (3,4.5,4.5,6)
Aggregation term Membership function parameters
Moderately strict (MS) (4,5.5,5.5,7)
Highly strict (HS) (6,7.5,7.5,9) Negligible (N) (0,0,1,2)
Extremely strict (ES) (8,9,10,10) Very Low (VL) (1,2,2,3)
Low (L) (2,3,3,4)
Low Moderate (LM) (3,4,4,5)
Moderate (M) (4,5,5,6)
(Bashiri & Hosseininezhad, 2009). For example, with one output
High Moderate (HM) (5,6,6,7)
variable (nov = 1), two input variables (niv = 2), and three linguis- High (H) (6,7,7,8)
tic levels for each of the input variables (L1 = L2 = 5), the number Very High (VH) (7,8,8,9)
of fuzzy rules is (5 × 5) = 25. This is the situation for each of the Extreme (E) (8,9,10,10)
FIS-A sub-modules.

Usingvi from Step 2 we first determine a normalized weight vfor
i
each factor at the sub-module level. The normalization expression ∼
The normalized importance weight differences ( v ) are repre-
is shown in expression (4). i, j
∼ vi ∼ vj sented by the rows in Table 5 and is the second input to FIS-A
v= ,v = (4) ∼
i vi + v j j vi + v j sub-module of the NFISI structure.  v can be calculated using ex-
i, j
∼ pression 4, and ranges from very weakly different to very strongly
 v values are then calculated based on the normalized im-
i, j different. Thus, using Table 5 rule base as an example, ifIi,j is “Negli-
∼ ∼ ∼
portance weights for factors i and j, ( v , v ) within each aggregation gible” (2nd Column) and  v is “Moderately Different” (3rd Row),
i j i, j
sub-module using the expression (5). the output RIi,j from these two is “slightly strict” (SS).
∼ ∼ 
∼   Finally, defuzzification should take place to convert the output
 v = vi − v j  (5) from its current format (fuzzy membership functions) to a crisp
i, j

value. This process differs in the way the crisp value is estimated
The fuzzy membership functions for input terms (Ii,j , v ) and from the union of different membership functions which result
i, j
output term (RIi,j ) and the generic rule-base of the FIS-A sub- from the rule-base (Ross, 2010). A handful of defuzzification meth-
module forRIij are defined in Tables 2–5, respectively. The linear ods are developed in the literature among which centroids and
triangular and trapezoidal membership functions, the most fre- maxima methods have usually been used as standard approaches
quently used approaches in the literature, are applied for fuzzifica- (Talon & Curt, 2017). To this end, Centroid of Area (COA) along with
tion of input data (Chen, 2009). Using fuzzified input variables and the maxima methods, Smallest of Maximum (SOM), Mean of Maxi-
the rule-base, RIi,j is estimated to demonstrate the extent a factor mum (MOM), and Largest of Maximum (LOM) are employed in this
is contributing to the output value at each sub-module level. study to determine this defuzzified value.
In Table 4, the linguistic output terms range from slightly loose The step-wise aggregation sub-module (FIS-B) then usesRIi,j
to extremely strict. The slightly loose result occurs when the dif- along with ESik and ESjk as inputs to generate the aggregated score
∼ (ASxk ). In executing FIS-B, the inference engine processes the data
ference in importance weights ( v ) is small and the interaction
i, j considering two groups of inputs from Steps 2 and 4 and the gen-
is negative. Extremely strict occurs when there is a significant fac- eral rule-base and generates the corresponding fuzzy output. For
tor importance weight difference and a positive interaction value this purpose, the fuzzy membership functions for the input and
among the factors. Extremely strict refers to the case where one output variables, and the generic rule-base of sub-module FIS-B
factor contributes greatly to the aggregation procedure. are elaborated in Tables 6–8, respectively; the specifications are
In Table 5, the columns represent whether the interaction term similar to FIS-A. Referring to Table 8 as the rule base of FIS-B,
(Ii,j ) calculated by expression (1) is negative, negligible or positive. and considering an example whereRIi,j is “Slightly Loose” (1st sub-
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 281
Table 8
Fuzzy rules for FIS-B, with respect to relativity indicators.

RIi,j : Slightly loose ESik

Negligible Low Moderate High Extreme

ESjk Negligible N L L M M
Low L L M M H
Moderate L M M H H
High M M H H E
Extreme M H H E E
RIi,j : Non-strict ESik

Negligible Low Moderate High Extreme

ESjk Negligible N VL L LM M
Low VL L LM M HM
Moderate L LM M HM H
High LM M HM H VH
Extreme M HM H VH E
RIi,j : Slightly Strict ESik

Negligible Low Moderate High Extreme

ESjk Negligible N N L L M
Low L L L M M
Moderate L M M M H
High M M H H H
Extreme M H H E E
RIi,j : Moderately Strict ESik

Negligible Low Moderate High Extreme

ESjk Negligible N N N L L
Low L L L L M
Moderate M M M M M
High M H H H H
Extreme H H E E E
RIi,j : Highly Strict ESik

Negligible Low Moderate High Extreme

ESjk Negligible N N N N L
Low L L L L L
Moderate M M M M M
High H H H H H
Extreme H E E E E
RIi,j : Extremely Strict ESik

Negligible Low Moderate High Extreme

ESjk Negligible N N N N N
Low L L L L L
Moderate M M M M M
High H H H H H
Extreme E E E E E

table), ESik and ESjk are “Moderate” (3rd Column) and “Extreme” rounds of interviews were used to help gather the input data for
(5th Row), the resulting output (ASxk ) is “High”. this study.
According to the interviewee, the company is planning to in-
4. An illustrative application crease the speed of launching new products. The long-term plan of
the company is shifting the business model from a manufacturer to
4.1. Case company background a platform where most of the production is outsourced. PD is ex-
pected to become more important in the future of the company’s
The case company is one of the oldest multinational consumer operations. Given the outsourcing strategy, supply chain issues are
goods producers. It is ranked among the world’s five largest com- also likely to become even more prevalent. Currently, the PD deci-
panies in FMCG based on net sales. The company operates glob- sions occur in the case company either as a short-term action or a
ally, and provides a range of products, from food and refreshments planned one in forms of a mid-/long- term practice.
to home and personal care, with over 400 established brands.
This study uses food and refreshment products, which includes 4.2. Application of the developed model
the retail business (business-to-consumer) and solution business
(business-to-business). Products of the studied company can be 4.2.1. Data collection and preparation
categorized into "local products" which are sourced and produced Many of the model parameters used are based on two rounds of
in China, and "imported products" which are made elsewhere. interviews: an introductory interview about the on-going PD prac-
Considering the company’s reputation in the FMCG sector, the tice with the determination of salient factors through a system-
identified factors and the developed decision model have been atic filtering and screening process (see Appendix A); and a struc-
confirmed by a business development manager of the company, tured interview where a number of pairwise random scenarios for
who previously served as sales manager for nearly six years. Two PD were evaluated by the interviewee. A hypothetical case is con-
282 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Table 9 final prioritization can only be completed after having an aggre-


Salient factors resulting from the filtering process.
gated measure that includes all decision factors and their interac-
Factors Symbol tions as defined by the NFISI methodology.
High variance in customer feedback SF1
In this step, the decision model parameters, the importance
Strategic-fit contradiction SF2 weights, and interaction parameters are determined. Given the to-
Low total profit SF3 tal number of decision factors and the alpha value, 37 pairwise
Negative customer feedback SF4 product scenario comparisons should take place in order to deter-
Low flexibility in the company’s logistics SF5
mine the 8-factor importance weights and 28 interaction param-
High level of allocated supply chain resources SF6
High logistics expenses SF7 eters. To this end, the managerial pairwise qualitative preferences
Average time in the supply chain SF8 of product scenarios are presented. The responding FMCG manager
was asked the following question: "How much more likely is the
decision on deletion of a certain product in the first scenario com-
pared to the other?" The linguistic comparisons are then translated
into linear algebraic equations (Table B.1 of Appendix B). The solu-
tion to this set of equations, the importance weights, and interac-
tion indexes are summarized in Table 11.
Results from this step indicate that five interaction terms (I1,3 ,
I1,4 , I1,5 , I1,7 , I3,7 ) have a negative value. In other words, the cor-
responding pairs of factors are perceived as redundant, and the
overall importance weight of each pair is less than the case where
there is negligible interaction. In this case, a more flexible scheme
Fig. 3. Decision maker’s judgment on "Strategic-fit contradiction". compared to the other cases should be applied in the inference
system of the respective aggregation sub-modules at NFISI. This
scheme can be non-strict (optimistic aggregation) when the dif-
sidered where several SKUs of the company should go through a ference of importance among two decision factors is not large. For
quick audit in order to be prioritized for possible deletion. example, aggregation of SF1 and SF7 should take place based on a
The fifteen factors listed in Table 1 were deemed relevant to non-strict scheme, because this pair of decision factors represents
the particular case of FMCG and chosen (by the interviewee) as PD a negligible difference in the importance value. A small score for
decision criteria. For the sake of simplicity, the first eight decision the first criterion is favorable, and can compensate for the sec-
factors are considered as the salient factors (SFx ). The factors listed ond criterion’s high corresponding value, even if it is relatively
in Table 9 showed the highest prominence in the filtering process more important in the overall situation. When the difference be-
(Appendix A), and together account for about 80 percent of con- tween importance weights of the two decision factors is consid-
tribution to the PD decision. The rest of the factors, i.e. the com- erably large, the more important factor dominates the aggregation
petitive factors, may be recognized as salient PD factors in other despite the weak value of the other factor (e.g. I3,7 in the illustra-
industries, like apparel and sports goods. tive case).
The FMCG product manager is given five products within a hy-
pothetical case. Two products are exemplary: extreme cases, in or- 4.2.2. Executing NFISI for all alternatives
der to keep the comparison simple and consistent, while asking for The schematic structure of the developed NFISI for this illus-
fewer comparative judgments (Kirkwood, 1997). The undesirable trative problem situation is in Fig. 4, where eight decision factors
exemplary product represents an undesirable SKU (SKU-U) which are included in the PD evaluation. The fuzzy membership functions
is poorly performing and very likely to be deleted, and a desirable and rules were described in the previous section. The final re-
exemplary product (SKU-D) which is likely to survive the deletion sults for prioritizing the candidate products for deletion are shown
evaluation. The other three products are assumed to go through here. Moreover, and following other FIS applications (Amindoust
an initial screening of the SKUs; most show some unsatisfactory et al., 2012; Ordoobadi, 2009), five different methods of defuzzi-
performance. For example, SKU-A shows the weakest sales per- fication are applied to show the robustness of the Overall Aggre-
formance compared to the other two SKUs. SKU-B is performing gate Score from the NFISI method for a product. The Overall Ag-
better financially, but is a poorer relative strategic-fit. SKU-C is a gregated Scores, under different defuzzification methods, appear in
burdensome product in terms of supply chain operational metrics, Table 12. In this analysis, altering the defuzzification method does
while representing an average performance in terms of financial, not lead to a major change in the decision-making outcome. De-
market, and strategic considerations. spite the difference in the overall value for the candidate SKUs, the
M-Macbeth software determines the relative commensurate prioritization of the products remains the same when using differ-
scores on product performance for each criterion. The software ent methods, which helps validate the accuracy of the developed
progressively checks if the three measurement conditions (see model over different phases of defuzzification.
Bana E Costa (2012)) are true with the entered judgment and con- Apart from the exemplary undesirable product which received
firms whether there is any inconsistency with the comparison ma- the highest overall dissatisfaction score, SKU-B is unanimously
trix of decision factors before generating the final scores. The lin- given the highest priority for PD. In other words, this SKU has the
guistic inputs used in this scoring system identifies the extent a highest unsatisfactory score when considering its status with re-
row item is worse than a column item; this is the qualitative pair- spect to all the salient decision factors. SKU-C is next to SKU-B in
wise judgment of “how much worse is the item in the row com- Table 12, while representing a significantly better (smaller) score
pared to the column item”. For example, SKU-B is strongly worse compared to SKU-B. SKU-A is likely to survive this evaluation pro-
than SKU-C and SKU-A in terms of "strategic-fit contradiction", as cess, and remains in the organization’s product portfolio. Expect-
shown in Fig. 3. However, SKU-B is weakly better than the exem- edly, the lowest priority for PD is for the exemplary desirable can-
plary undesirable product. Accordingly, the SKUs can be prioritized didate.
from worst to best (B, C, and A) considering this specific decision After executing NFISI, the salient supply chain factors are ex-
factor. Following the same procedure, Table 10 provides the order cluded from the analysis to determine the robustness of the so-
of likelihood for deletion considering each criterion. However, the lution, and whether supply chain operational factors can play a
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 283

Table 10
Elementary scores (ESik ) of products on each criterion.

Products Salient factors

SF1 SF2 SF3 SF4 SF5 SF6 SF7 SF8

SKU-U 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0


Priority 1 1 1 1 1 1 1 1
SKU-A 2.508 1.828 5.714 6.363 2.143 3.334 3.371 3.089
Priority 4 4 2 2 3 3 3 4
SKU-B 7.530 8.182 2.143 5.450 1.786 0.833 1.315 4.612
Priority 2 2 4 3 4 4 4 3
SKU-C 3.350 3.637 4.286 1.809 8.362 5.00 7.312 7.631
Priority 3 3 3 4 2 2 2 2
SKU-D 0 0 0 0 0 0 0 0
Priority 5 5 5 5 5 5 5 5

Table 11
Model parameters: importance weights and interaction indexes.

V1 V2 V3 V4 V5 V6 V7 V8 I1,2 I1,3 I1,4 I1,5


0.0243 0.1897 0.2246 0.0941 0.1419 0.1229 0.0577 0.1449 0.0501 −0.0379 −0.0364 −0.0273
I1,6 I1,7 I1,8 I2,3 I2,4 I2,5 I2,6 I2,7 I2,8 I3,4 I3,5 I3,6
0.0243 −0.0137 0.0501 0.1745 0.0364 0.1442 0.0592 0.0971 0.0819 0.1533 0.2367 0.1442
I3,7 I3,8 I4,5 I4,6 I4,7 I4,8 I5,6 I5,7 I5,8 I6,7 I6,8 I7,8
−0.0258 0.132 0.0728 0.0653 0.0395 0.0364 0.0637 0.0865 0.0744 0.0349 0.0243 0.1002

Fig. 4. Nested-Fuzzy Inference System with Interactions.


284 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Table 12 is shown that high variance in customer feedback is as crucial as


Overall dissatisfaction scores and priority from the set of
the negative feedback to a PD decision. On the supply side, low
candidate SKUs for deletion.
flexibility in the operational capabilities is introduced as a factor
Products Nested-FIS with Interactions which may encourage PD implementation. The results also show
COAa BOMb MOMc SOMd LOMe that the longer average time a product spends in the supply chain
SKU-U 9.24 9.3 9.5 8.5 10
can contribute greatly to a PD decision, given the costs and barriers
Priority 1 1 1 1 1 it causes. The importance of supply chain activities in FMCG makes
SKU-A 3.0 3.0 3.0 2.9 3.6 the aforementioned factors even more important for PDDM in this
Priority 4 4 3 3 4 sector. Simply considering total profit, which is the on-going prac-
SKU-B 7.0 7.0 7.0 4.3 7.1
tice in the case study company, cannot guarantee the company’s
Priority 2 2 2 2 2
SKU-C 5.0 3.4 3.0 2.9 5.6 long-term financial performance. The balance between costs and
Priority 3 3 3 3 3 benefits of keeping or deleting a product in the long term may not
SKU-D 0.75 0.7 0.5 0 1.5 match the short-term profit perspective currently in place.
Priority 5 5 5 5 5 A second evident implication from the analysis shows that
a
Center of area. considering interaction amongst factors significantly influences
b
Bisector of area. decision-making outcomes. When there is a synergy amongst two
c
Mean of maximum. factors (positive interaction), it becomes essential to include this
d
Smallest of maximum.
e
Largest of maximum.
factor interaction when evaluating alternatives. Positive interaction
amongst the salient supply chain factors and the financial factor
role in the ultimate deletion decision. These results are added to of low total profit is the best example of this circumstance. This
Table 13 (Set 1 –Traditional factors), which also includes all factors example factor interaction confirms the necessity of putting em-
from Table 12’s final column ranking. phasis on the aforementioned factors when approaching the PD
The first row of Table 13 shows the prioritization based only on decision. Alternatively, the factors with negative interaction share,
a profit ranking. The candidate PD priorities, except for the exem- to some extent, a similar objection and their combination should
plary SKUs, have changed. The overall score and priority of SKU-B not be overemphasized in the evaluation process. Application of
remain about the same when comparing sets 1 and 2. However, it other techniques, such as AHP and ANP, may lead to major changes
is given the first priority for deletion in both cases (after the ex- in the decision-making outcomes due to synergy or redundancy
emplary SKU). This result is in contrast with the on-going PD prac- amongst the factors that may not be captured, since zero or negli-
tice of the company, according to which the SKU-B would survive gible interaction is assumed.
this evaluation and remain in the company’s portfolio due to its This study’s proposed technique provides a ranking of the alter-
profitability. In other words, SKU-B as an un-fit product might not natives for eventual PD implementation. The decision maker can
have been deleted if only financial performance is considered, even arrive at a final decision based on the priorities given to the al-
though it is adversely impacting overall profitability. SKU-C has ternatives, and decide to delete one or more SKUs based on the
a weaker overall score after taking supply chain decision factors company’s situation. A measure for the overall score of each dele-
into consideration, although its rank remains the same over the tion alternative is also provided. This measure aids management
three scenarios. SKU-A could have been dropped from the com- to effectively analyze and compare the gaps amongst alternatives.
pany’s portfolio in case of applying the on-going practice. When Management can also set a threshold for the overall score, a sat-
compared to the situation where only low sales volume and unit isficing approach, and delete all the alternatives which receive a
profit is considered, the below-average sales performance of SKU-A score worse than the threshold, while considering the situation.
is compensated through the inclusion of supply chain operational The developed approach can also be applied for the selection of
factors, making it the last priority of deletion amongst the other alternative products for launch from the "innovation gate" of the
alternatives. SKU-B is performing better financially, but it does not company. In this particular application area, the fuzziness of the
conform to strategic-fit of the company, and the positive inter- approach can be particularly useful since there is no real data of
action value among the sales performance and the other factors the possible performance of each alternative after the launch.
averts it from a positive compensation in the overall score. Although the developed model is suitable for application in
similar industries, such as sports goods and apparel industries,
4.3. Discussion there are limitations to our study. The factors, importance weights,
and interactions will vary depending on the context. The first limi-
To successfully apply strategic PD, a decision aid method, as tation of this study is the consideration of a single illustrative com-
part of an expert system, considering a broader set of decision pany and industry situation. Future studies with other contexts
factors is necessary. The major issue in current PD practice is a may require different or additional PD decision factors in accor-
disregard of operational and competitive factors; especially along dance with industrial and company characteristics. Other factor di-
the supply chain. Additionally, taking interactions among the fac- mensions such as feasibility, sustainability, and risks, may be uti-
tors into account is of significant importance to MCDM outcomes. lized in addition to the studied operational and competitive fac-
The following managerial implications can be extrapolated from tors. This approach would likely entail involvement by additional
the study. organizational functions such as environmental management, en-
The first evident implication can be drawn from the results of gineering, risk management tools, and life cycle thinking for a PD
the first two steps of the multi-stage, multi-method expert system. decision. For example, the studied FMCG case company has a very
The initial results confirm that the operational and competitive fac- strict assessment of various sustainability factors, and a product
tors listed in Table 1, in addition to the traditional factors intro- that does not satisfy the company’s standards cannot be launched
duced by earlier studies (Avlonitis, 1984, 1993; Avlonitis & Argous- in the first place. Environmental and risk factors can be particularly
lidis, 2012; Avlonitis et al., 20 0 0; Mitchell et al., 1998), are all rec- influential to PD decisions, especially in industries facing interna-
ognized by the experts as relevant. We have noticed that, given the tional regulations and regulatory compliance. Another main limita-
examples, some of the supply chain operational factors are more tion of this study is the number of respondents and the fact that
significant for PD decisions; inclusion of these factors makes a dif- it did not contain the managers across other disciplines involved
ference in the decision-making outcomes. On the demand side, it in PDDM. Overcoming this limitation, particularly through apply-
P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 285

Table 13
Comparison results (financial factor vs comprehensive sets).

Products Low total profit Set 1 –Traditional factors (SF1 : SF4 ) Set 2 –All factors (SF1 : SF8 )

SKU-U 10 10 10
Priority 1 1 1
SKU-A 5.7 3.6 3.6
Priority 2 3 4
SKU-B 2.1 7.5 7.1
Priority 4 2 2
SKU-C 4.3 3.6 5.6
Priority 3 3 3
SKU-D 0 1.5 1.5
Priority 5 5 5

ing the developed model in other industries, is another potential Acknowledgments


research opportunity that will contribute to the understudied topic
of PD. This work is supported by National Natural Science Foundation
of China (71632007, 71690241, and 71472021).

5. Conclusions
Appendix A
Acknowledging the role of supply chains, business activities go
After determining the most important decision factor and its
beyond just traditional financial thinking. A number of organiza-
preference over the others (Tables A.1), and the least important de-
tional strategic decisions need to incorporate supply chain opera-
cision factor and preferences of other decision factors to the least
tional aspects, because competencies of a company rest upon its
important decision factor, (Tables A.2), optimum value of weight
supply chain performance (Pourhejazy & Kwon, 2016). Critical and
and ε can be obtained by solving the following optimization prob-
strategic decisions that used to be made based on short-term fi-
lem:
nancial indicators can and should be revised using systematic ap-
proaches that take into account strategic and non-financial aspects. Min ε (A1)
This study introduced a broader set of decision factors to the PD
decision process including consideration of competitive and sup- Subject to:
 
ply chain factors. An expert system is developed which includes  Wbest 
 
multiple criteria and fuzzy approaches to filter, generate data to
 W j − am, j  ≤ ε (A2)
form the knowledge base, and aggregate data in a series of stages.
This paper proposes using and extending NFISI, as the inference  
 Wj 
engine of the expert system, and conducting a simultaneous anal-  
ysis of traditional and supply chain operational factors that interact  Wworst − a j,l  ≤ ε (A3)

in the evaluation. Given the comparatively more frequent nature of



PD practices in FMCG, an illustrative case from a well-known com- Wj = 1 (A4)
pany from this sector is put forward to verify the applicability of j
the model Results of the case are discussed from the perspective
of traditional factors and broader evaluations, including integrat-
ing the supply chain operational factors into the traditional fac- Table A.2
tors. Our initial finding from this study is that the inclusion of the BWM data from the interviewee: the least important factor com-
introduced salient factors, even for a small problem, can radically pared to others.
change the final decision. Thus, careful examination and data eval- Least Important Criterion: C12
uation are needed, and various complexities in relationships and
The other decision factors C1 9
interactions can cause diversions away from simpler evaluations. C2 7
Overall, this study provides an opportunity for an under- C3 7
researched topic to blossom in many directions including empir- C4 5
ical, case, and analytical studies. Application of operations research C5 7
C6 5
approaches is a possible research direction to contribute to the PD C7 4
decision-making topic. PD also brings further attention to the need C8 5
for strategic consideration of PD decisions, similar to the strate- C9 7
gic consideration of new product development and introduction. C10 3
C11 4
When additional data on applying strategic PD is available, one
C13 2
can assess the extent to which the competitive and supply chain C14 2
factors can actually benefit from the decision. Finally, PD can be C15 4
studied as a tool for improving the sustainability and resilience of
supply chains i.e. through optimization approaches.

Table A.1
BWM data from the interviewee: the most important factor compared to others.

The other decision factors

Most Important Criterion: C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15


3 3 5 3 5 6 5 3 7 6 9 8 8 6
286 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Table A.3
Factor weights from BWM.

Factor C1 ∗ C2 ∗ C3 ∗ C4 ∗ C5 ∗ C6 ∗ C7 C8 ∗

Importance 0.2587 0.0862 0.0862 0.0517 0.0862 0.0517 0.0431 0.0517


Rank 1 2 2 6 2 6 9 6

Factor C9 ∗ C10 C11 C12 C13 C14 C15 ε


Importance 0.0862 0.0370 0.0431 0.0103 0.0323 0.0323 0.0431 0.1655
Rank 2 12 9 15 13 13 9 -

Chosen as a salient factor (SFx ).

The objective function ensures maximum reliability of the re- Appendix B


sults (higherε is equivalent to less consistency among the compar-
isons). The parametersam,j andaj,l demonstrate the pairwise compar- The equational counterpart to the cardinal preference of the in-
ison of the most and the least significant criterion with the rest terviewee over the randomly generated pair of scenarios is listed
of decision factors, respectively. Wbest ,Wworst , andWj are the deci- in Table B.1, where the alpha value demonstrates the cardinal dif-
sion variables representing reference and non-reference variables (C ,C ,C ,C ,C ,C ,C ,C )
ference and Px 1 2 3 4 5 6 7 8 represents the scenario x status
(Table A.3). in terms of decision factors one to eight.

Table B.1
Equational transformation of the cardinal measures.

No. Cardinal measure Weight Interaction Alpha Const.


(0,0,0,0,1,0,0,1)
P1 − + 12 [I1,7 + I2,7 + I3,7 + I4,7 − I5,7 + I6,7 ]−
1 − v7 + v8 − 5α 0
P2(0,0,0,0,1,0,1,0) =
1
[I + I2,8 + I3,8 + I4,8 − I5,8 + I6,8 ] =
2 1,8
P1(1,0,0,0,0,0,0,1) − + 12 [−I1,6 + I2,6 + I3,6 + I4,6 + I5,6 + I6,7 ]−
2 − v6 + v8 2α 0
P2(1,0,0,0,0,1,0,0) =
1
2
[−I1,8 + I2,8 + I3,8 + I4,8 + I5,8 + I7,8 ] =
P1(0,0,0,0,1,0,1,0) − + 12 [I1,6 + I2,6 + I3,6 + I4,6 − I5,6 + I6,8 ]−
3 − v6 + v7 2α 0
P2(0,0,0,0,1,1,0,0) =
1
[I + I2,7 + I3,7 + I4,7 − I5,7 + I7,8 ] =
2 1,7
P1(0,0,1,0,0,0,1,1) − + 12 [I1,5 + I2,5 − I3,5 + I4,5 + I5,6 − I5,7 ]−
4 − v5 + v8 − 4α 0
P2(0,0,1,0,1,0,1,0) =
1
[I + I2,8 − I3,8 + I4,8 + I6,8 − I7,8 ] =
2 1,8
P1(0,1,0,0,0,1,1,0) − + 12 [I1,5 − I2,5 + I3,5 + I4,5 − I5,6 + I5,8 ]−
5 − v5 + v7 2α 0
P2(0,1,0,0,1,1,0,0) =
1
[I − I2,7 + I3,7 + I4,7 − I6,7 + I7,8 ] =
2 1,7
P1(0,0,0,0,0,1,1,1) − + 12 [I1,5 + I2,5 + I3,5 + I4,5 − I5,7 − I5,8 ]−
6 − v5 + v6 −α 0
P2(0,0,0,0,1,0,1,1) =
1
[I + I2,6 + I3,6 + I4,6 − I6,7 − I6,8 ] =
2 1,6
P1(0,0,0,0,0,0,0,1) − + 12 [I1,4 + I2,4 + I3,4 + I4,5 + I4,6 + I4,7 ]−
7 − v4 + v8 − 5α 0
P2(0,0,0,1,0,0,0,0) =
1
[I + I2,8 + I3,8 + I5,8 + I6,8 + I7,8 ] =
2 1,8
P1(0,0,0,0,0,0,1,0) − + 12 [I1,4 + I2,4 + I3,4 + I4,5 + I4,6 + I4,8 ]−
8 − v4 + v7 − 4α 0
P2(0,0,0,1,0,0,0,0) =
1
[I + I2,7 + I3,7 + I5,7 + I6,7 + I7,8 ] =
2 1,7
P1(0,0,0,0,0,1,1,1) − + 12 [I1,4 + I2,4 + I3,4 + I4,5 − I4,7 − I4,8 ]−
9 − v4 + v6 − 4α 0
P2(0,0,0,1,0,0,1,1) =
1
[I + I2,6 + I3,6 + I5,6 − I6,7 − I6,8 ] =
2 1,6
P1(0,0,1,0,1,1,0,0) − + 12 [I1,4 + I2,4 − I3,4 − I4,6 + I4,7 + I4,8 ]−
10 − v4 + v5 − 4α 0
P2(0,0,1,1,0,1,0,0) =
1
[I + I2,5 − I3,5 − I5,6 + I5,7 + I5,8 ] =
2 1,5
P1(1,0,0,0,0,1,0,1) − + 12 [−I1,3 + I2,3 + I3,4 + I3,5 − I3,6 + I3,7 ]−
11 − v3 + v8 − 5α 0
P2(1,0,1,0,0,1,0,0) =
1
2
[−I1,8 + I2,8 + I4,8 + I5,8 − I6,8 + I7,8 ] =
P1(0,1,0,0,0,0,1,0) − + 12 [I1,3 − I2,3 + I3,4 + I3,5 + I3,6 + I3,8 ]−
12 − v3 + v7 − 5α 0
P2(0,1,1,0,0,0,0,0) =
1
[I − I2,7 + I4,7 + I5,7 + I6,7 + I7,8 ] =
2 1,7
P1(0,0,0,1,0,1,0,0) − + 12 [I1,3 + I2,3 − I3,4 + I3,5 + I3,6 + I3,8 ]−
13 − v3 + v6 − 3α 0
P2(0,0,1,1,0,0,0,0) =
1
[I + I2,6 − I4,6 + I5,6 + I6,7 + I6,8 ] =
2 1,6
P1(1,0,0,0,1,0,0,0) − + 12 [−I1,3 + I2,3 + I3,4 + I3,5 + I3,6 + I3,8 ]−
14 − v3 + v5 − 3α 0
P2(1,0,1,0,0,0,0,0) =
1
2
[−I1,5 + I2,5 + I4,5 + I5,6 + I5,7 + I5,8 ] =
P1(1,0,0,1,0,1,0,0) − + 12 [−I1,3 + I2,3 + I3,5 − I3,6 + I3,7 + I3,8 ]−
15 − v3 + v4 −α 0
P2(1,0,1,0,0,1,0,0) =
1
2
[−I1,4 + I2,4 + I4,5 − I4,6 + I4,7 + I4,8 ] =
P1(0,0,1,0,0,0,0,1) − + 12 [I1,2 − I2,3 + I2,4 + I2,5 + I2,6 + I2,7 ]−
16 − v2 + v8 − 5α 0
P2(0,1,1,0,0,0,0,0) =
1
[I − I3,8 + I4,8 + I5,8 + I6,8 + I7,8 ] =
2 1,8
P1(1,0,0,0,0,0,1,1) − + 12 [−I1,2 + I2,3 + I2,4 + I2,5 + I2,6 − I2,8 ]−
17 − v2 + v7 − 5α 0
P2(1,1,0,0,0,0,0,1) =
1
2
[−I1,7 + I3,7 + I4,7 + I5,7 + I6,7 − I7,8 ] =
P1(0,0,0,0,0,1,1,0) − + 12 [I1,2 + I2,3 + I2,4 + I2,5 − I2,7 + I2,8 ]−
18 − v2 + v6 − 5α 0
P2(0,1,0,0,0,0,1,0) =
1
[I + I3,6 + I4,6 + I5,6 − I6,7 + I6,8 ] =
2 1,6
P1(0,0,0,1,1,0,0,0) − + 12 [I1,2 + I2,3 − I2,4 + I2,6 + I2,7 + I2,8 ]−
19 − v2 + v5 − 5α 0
P2(0,1,0,1,0,0,0,0) =
1
[I + I3,5 − I4,5 + I5,6 + I5,7 + I5,8 ] =
2 1,5
P1(0,0,0,1,0,0,1,0) − + 12 [I1,2 + I2,3 + I2,5 + I2,6 − I2,7 + I2,8 ]−
20 − v2 + v4 − 5α 0
P2(0,1,0,0,0,0,1,0) =
1
[I + I3,4 + I4,5 + I4,6 − I4,7 + I4,8 ] =
2 1,4
P1(1,0,1,1,0,0,0,0) − + 12 [−I1,2 − I2,4 + I2,5 + I2,6 + I2,7 + I2,8 ]−
21 − v2 + v3 −α 0
P2(1,1,0,1,0,0,0,0) =
1
2
[−I1,3 − I3,4 + I3,5 + I3,6 + I3,7 + I3,8 ] =
P1(0,1,0,0,1,0,0,1) − + 12 [−I1,2 + I1,3 + I1,4 − I1,5 + I1,6 + I1,7 ]−
22 − v1 + v8 3α 0
P2(1,1,0,0,1,0,0,0) =
1
2
[−I2,8 + I3,8 + I4,8 − I5,8 + I6,8 + I7,8 ] =
P1(0,0,0,0,1,1,1,0) − + 12 [I1,2 + I1,3 + I1,4 − I1,5 − I1,6 + I1,8 ]−
23 − v1 + v7 α 0
P2(1,0,0,0,1,1,0,0) =
1
[I + I3,7 + I4,7 − I5,7 − I6,7 + I7,8 ] =
2 2,7

(continued on next page)


P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288 287

Table B.1 (continued)

No. Cardinal measure Weight Interaction Alpha Const.


(0,0,0,1,0,1,0,0)
P1 − + 12 [−I1,2 + I1,3 − I1,4 + I1,5 + I1,7 + I1,8 ]−
24 − v1 + v6 2α 0
P2(1,0,0,1,0,0,0,0) =
1
2
[−I2,6 + I3,6 − I4,6 + I5,6 + I6,7 + I6,8 ] =
P1(0,0,0,1,1,0,1,0) − + 12 [I1,2 + I1,3 − I1,4 + I1,6 − I1,7 + I1,8 ]−
25 − v1 + v5 2α 0
P2(1,0,0,1,0,0,1,0) =
1
[I + I3,5 − I4,5 + I5,6 − I5,7 + I5,8 ] =
2 2,5
P1(0,0,0,1,1,0,0,0) − + 12 [I1,2 + I1,3 − I1,5 + I1,6 + I1,7 + I1,8 ]−
26 − v1 + v4 − 3α 0
P2(1,0,0,0,1,0,0,0) =
1
[I + I3,4 − I4,5 + I4,6 + I4,7 + I4,8 ] =
2 2,4
P1(0,0,1,1,0,0,0,0) − + 12 [I1,2 − I1,4 + I1,5 + I1,6 + I1,7 + I1,8 ]−
27 − v1 + v3 2α 0
P2(1,0,0,1,0,0,0,0) =
1
[I − I3,4 + I3,5 + I3,6 + I3,7 + I3,8 ] =
2 2,3
P1(0,1,0,0,0,0,1,1) − + 12 [I1,3 + I1,4 + I1,5 + I1,6 − I1,7 − I1,8 ]−
28 − v1 + v2 − 3α 0
P2(1,0,0,0,0,0,1,1) =
1
[I + I2,4 + I2,5 + I2,6 − I2,7 − I2,8 ] =
2 2,3
v1 + v2 +
v3 + v4 +
29 1
v5 + v6 +
v7 + v8 =
P1(0,1,0,0,0,1,1,0) −
30 v6 − 12 [I1,6 − I2,6 + I3,6 + I4,6 + I5,6 − I6,7 + I6,8 ] = 3α 0
P2(0,1,0,0,0,0,1,0) =
P1(1,0,1,1,0,1,1,1) −
31 v1 − 12 [I1,2 − I1,3 − I1,4 + I1,5 − I1,6 − I1,7 − I1,8 ] = 2α 0
P2(0,0,1,1,0,1,1,1) =
P1(0,0,0,1,0,1,1,0) −
32 v4 − 12 [I1,4 + I2,4 + I3,4 + I4,5 − I4,6 − I4,7 + I4,8 ] = 5α 0
P2(0,0,0,0,0,1,1,0) =
P1(1,0,0,0,0,0,1,0) −
33 v1 − 12 [I1,2 + I1,3 + I1,4 + I1,5 + I1,6 − I1,7 + I1,8 ] = 2α 0
P2(0,0,0,0,0,0,1,0) =
P1(0,0,1,1,0,0,1,1) −
34 v7 − 12 [I1,7 + I2,7 − I3,7 − I4,7 + I5,7 + I6,7 − I7,8 ] = 4α 0
P2(0,0,1,1,0,0,0,1) =
P1(0,0,0,1,0,1,0,1) −
35 v4 − 12 [I1,4 + I2,4 + I3,4 + I4,5 − I4,6 + I4,7 − I4,8 ] = 4α 0
P2(0,0,0,0,0,1,0,1) =
P1(1,0,0,1,1,0,0,1) −
36 v1 − 12 [I1,2 + I1,3 − I1,4 − I1,5 + I1,6 + I1,7 − I1,8 ] = 2α 0
P2(0,0,0,1,1,0,0,1) =
P1(0,1,0,1,0,1,0,0) −
37 v4 − 12 [I1,4 − I2,4 + I3,4 + I4,5 − I4,6 + I4,7 + I4,8 ] = 4α 0
P2(0,1,0,0,0,1,0,0) =

References Brandenburg, M., & Seuring, S. (2011). Impacts of supply chain management on
company value: Benchmarking companies from the fast moving consumer
Abbasimehr, H., & Tarokh, M. (2017). A combined approach based on fuzzy AHP and goods industry. Logistics Research, 3(4), 233–248.
fuzzy inference system to rank reviewers in online communities. Turkish Journal Brown, T. J., & Dacin, P. A. (1997). The company and the product: Corporate associ-
of Electrical Engineering & Computer Sciences, 25(2), 862–876. ations and consumer product responses. The Journal of Marketing, 61(1), 68–84.
Amindoust, A., Ahmed, S., Saghafinia, A., & Bahreininejad, A. (2012). Sustainable Chen, C.-M. (2009). A fuzzy-based decision-support model for rebuy procurement.
supplier selection: A ranking model based on fuzzy inference system. Applied International Journal of Production Economics, 122(2), 714–724.
Soft Computing Journal, 12(6), 1668–1677. doi:10.1016/j.asoc.2012.01.023. Chong, J.-K., Ho, T.-H., & Tang, C. S. (2004). Demand modeling in product line
Amirkhan, M., Didehkhani, H., Khalili-Damghani, K., & Hafezalkotob, A. (2018). trimming: Substitutability and variability. In A. K. Chakravarty, & J. Eliash-
Mixed uncertainties in data envelopment analysis: A fuzzy-robust approach. Ex- berg (Eds.), Managing business interfaces: marketing and engineering issues in
pert Systems with Applications, 103, 218–237. the supply chain and internet domains (pp. 39–62). Boston, MA: Springer US.
Argouslidis, P., Baltas, G., & Mavrommatis, A. (2014). Outcomes of decision speed: doi:10.1007/0- 387- 25002- 6_2.
An empirical study in product elimination decision-making processes. European Chopra, S. (2007). Supply Chain Management, 3/e. Pearson Education Retrieved
Journal of Marketing, 48(5/6), 982–1008. fromhttps://books.google.co.kr/books?id=rtjvMOO_1X4C.
Argouslidis, P. C., Baltas, G., & Mavrommatis, A. (2015). An empirical investigation Clivillé, V., & Berrah, L. (2012). Overall performance measurement in a supply chain:
into the determinants of decision speed in product elimination decision pro- Towards a supplier-prime manufacturer based model. Journal of Intelligent Man-
cesses. European Management Journal, 33(4), 268–286. ufacturing, 23, 2459–2469. doi:10.1007/s10845- 011- 0512- x.
Avlonitis, G. J. (1993). Project dropstrat: What factors do managers consider in de- Cliville, V., Berrah, L., & Mauris, G. (2007). Quantitative expression and aggregation
ciding whether to drop a product. European Journal of Marketing, 27(4), 35–57. of performance measurements based on the MACBETH multi-criteria method.
Avlonitis, G. J., & Argouslidis, P. C. (2012). Tracking the evolution of theory on prod- International Journal of Production Economics, 105(1), 171–189.
uct elimination: Past, present, and future. The Marketing Review, 12(4), 345–379. Colicchia, C., Creazza, A., & Dallari, F. (2017). The management of operations Lean
Avlonitis, G. J., Hart, S. J., & Tzokas, N. X. (20 0 0). An analysis of product deletion and green supply chain management through intermodal transport : Insights
scenarios. Journal of Product Innovation Management, 17(1), 41–56. from the fast moving consumer goods industry. Production Planning & Control,
Avlonitis, G. J. (1984). Industrial product elimination: Major factors to consider. In- 28(4), 319–332. doi:10.1080/09537287.2017.1282642.
dustrial Marketing Management, 13(2), 77–85. Cooper, M. C., & Ellram, L. M. (1993). Characteristics of supply chain management
Bai, C., Shah, P., Zhu, Q., & Sarkis, J. (2017). Green product deletion decisions: An and the implications for purchasing and logistics strategy. The International Jour-
integrated sustainable production and consumption approach. Industrial Man- nal of Logistics Management, 4(2), 13–24.
agement & Data Systems, 118(2), 349–389. doi:10.1108/IMDS- 05- 2017- 0175. Bana E Costa, Carlos A. (2012). MACBETH. International Journal of Information Tech-
Bana E Costa, Carlos A., & Vansnick, J.-C. (1997). Applications of the MACBETH ap- nology & Decision Making, 11(2), 359–387. doi:10.1142/S021962201240 0 068.
proach in the framework of an additive aggregation model. Journal of Multi-Cri- Bana E Costa, Carlos A., & Vansnick, J.-C. (1994). MACBETH—An interactive path to-
teria Decision Analysis, 6(2), 107–114. wards the construction of cardinal value functions. International Transactions in
Bashiri, M., & Hosseininezhad, S. J. (2009). A fuzzy group decision support system Operational Research, 1(4), 489–500.
for multifacility location problems. International Journal of Advanced Manufactur- Craig, B. M., Busschbach, J. J. V., & Salomon, J. A. (2009). Keep it simple: Ranking
ing Technology, 42, 533–543. doi:10.10 07/s0 0170-0 08-1621-3. health states yields values similar to cardinal measurement approaches. Journal
Bayus, B. L., & Putsis, W. P., Jr. (1999). Product proliferation: An empirical analy- of Clinical Epidemiology, 62(3), 296–305.
sis of product line determinants and market outcomes. Marketing Science, 18(2), De Toni, A., & Tonchia, S. (2001). Performance measurement systems-models, char-
137–153. acteristics and measures. International Journal of Operations & Production Man-
Berry, C., Burton, S., & Howlett, E. (2017). It’s only natural: The mediating im- agement, 21(1/2), 46–71.
pact of consumers’ attribute inferences on the relationships between product Dev, N. K., Shankar, R., Gunasekaran, A., & Thakur, L. S. (2016). A hybrid adaptive
claims, perceived product healthfulness, and purchase intentions. Journal of the decision system for supply chain reconfiguration. International Journal of Pro-
Academy of Marketing Science, 45(5), 698–719. duction Research, 54(23), 7100–7114.
Birhanu, D., Krishnanand, L., & Rao, A. N. (2017). Comparison of select financial pa- Dewa, P. K., Pujawan, I. N., & Vanany, I. (2017). Human errors in warehouse op-
rameters of Ethiopian consumer goods supply chains. Benchmarking: An Interna- erations: An improvement model. International Journal of Logistics Systems and
tional Journal, 24(1), 102–117. doi:10.1108/BIJ- 04- 2015- 0032. Management, 27(3), 298–317.
288 P. Pourhejazy, J. Sarkis and Q. Zhu / Expert Systems With Applications 119 (2019) 272–288

Diehl, D., & Spinler, S. (2013). Defining a common ground for supply chain risk Naim, M. M., Potter, A. T., Mason, R. J., & Bateman, N. (2006). The role of trans-
management–A case study in the fast-moving consumer goods industry. Inter- port flexibility in logistics provision. The International Journal of Logistics Man-
national Journal of Logistics Research and Applications, 16(4), 311–327. agement, 17(3), 297–311.
Dollet, J. N., & Díaz, A. (2011). Supply chain orchestration for the luxury alco- Ng, S. (2014). P&G to shed more than half its brands: Company to focus on
holic beverage sector. IUP Journal of Supply Chain Management, 8(3), 42–65. Re- 70 to 80 products, including Tide and Pampers. The Wall Street Journal,
trieved from http://search.ebscohost.com/login.aspx?direct=true&db=bth&AN= 1st August. Online article to be accessed at https://www.wsj.com/articles/
71490553&site=ehost-live . procter- gamble- posts- higher- profit- on- cost- cutting- 1406892304.
Dou, Y., & Sarkis, J. (2010). A joint location and outsourcing sustainability analysis Nielson, C. C. (1998). An empirical examination of the role of “closeness” in indus-
for a strategic offshoring decision. International Journal of Production Research, trial buyer-seller relationships. European Journal of Marketing, 32(5/6), 441–463.
48(2), 567–592. doi:10.1080/00207540903175145. Ordoobadi, S. M. (2009). Development of a supplier selection model using fuzzy
Bana E Costa, Carlos A., Lourenço, J. C., Chagas, M. P., & e Costa, J. C. (2008). Devel- logic. Supply Chain Management: An International Journal, 14(4), 314–327.
opment of reusable bid evaluation models for the Portuguese Electric Transmis- Pan, A., & Choi, T.-M. (2016). An agent-based negotiation model on price and de-
sion Company. Decision Analysis, 5(1), 22–42. livery date in a fashion supply chain. Annals of Operations Research, 242(2),
Fernie, J., & Sparks, L. (2014). Logistics and retail management: Emerging issues and 529–557.
new challenges in the retail supply chain. Kogan Page Publishers.. Porter, M. E. (2008). The five competitive forces that shape strategy. Harvard Busi-
Garbi, G. P., & Loureiro, G. (2014). Shared management of product portfolio. In ISPE ness Review, 86(1), 25–40.
CE: 55 (pp. 537–546). doi:10.3233/978- 1- 61499- 440- 4- 537. Pourhejazy, P., & Kwon, O. K. (2016). The new generation of operations research
Gokhana, N. M., Needy, K. L., & Normanc, B. A. (2015). Development of a simultane- methods in supply chain optimization: A review. Sustainability, 8(10), 1033.
ous design for supply chain process for the optimization of the product design doi:10.3390/su8101033.
and supply chain configuration problem. Engineering Management Journal, 22(4), Pourhejazy, P., Kwon, O. K., Chang, Y.-T., & Park, H. (2017). Evaluating resiliency
20–30. doi:10.1080/10429247.2010.11431876. of supply chain network: A data envelopment analysis approach. Sustainability,
Grabisch, M. (1996). The application of fuzzy integrals in multicriteria decision mak- 9(2), 255. doi:10.3390/su9020255.
ing. European Journal of Operational Research, 89(3), 445–456. Ramkumar, M. (2016). A modified ANP and fuzzy inference system based approach
Grabisch, M., Kojadinovic, I., & Meyer, P. (2008). A review of methods for capac- for risk assessment of in-house and third party e-procurement systems. Strategic
ity identification in Choquet integral based multi-attribute utility theory: Ap- Outsourcing: An International Journal, 9(2), 159–188.
plications of the Kappalab R package. European Journal of Operational Research, Ramkumar, M., Schoenherr, T., & Jenamani, M. (2016). Risk assessment of outsourc-
186(2), 766–785. ing e-procurement services: Integrating SWOT analysis with a modified ANP-
Grabisch, M., & Labreuche, C. (2016). Fuzzy measures and integrals in MCDA. In Mul- based fuzzy inference system. Production Planning & Control, 27(14), 1171–1190.
tiple Criteria Decision Analysis (pp. 553–603). Springer. doi:10.1080/09537287.2016.1190877.
Gumus, S., Kucukvar, M., & Tatari, O. (2016). Intuitionistic fuzzy multi-criteria Rezaei, J. (2015). Best-worst multi-criteria decision-making method. Omega, 53, 49–
decision-making framework based on life cycle environmental, economic and 57. doi:10.1016/j.omega.2014.11.009.
social impacts: The case of U.S. wind energy. Sustainable Production and Con- Ross, T. J. (2010). Fuzzy Logic with Engineering Applications. John Wiley & Sons, Ltd.
sumption, 8(June), 78–92. doi:10.1016/j.spc.2016.06.006. doi:10.1002/9781119994374.
Harness, D. R., & Harness, T. (2012). Can product elimination support post-downsiz- Sarkis, J., & Talluri, S. (2002). A model for strategic supplier selection. Journal of
ing success aspirations. Journal of General Management, 38(2), 39–60. Supply Chain Management, 38(4), 18–28.
Herman, M. W., & Koczkodaj, W. W. (1996). A Monte Carlo study of pairwise com- Sansone, C., Hilletofth, P., & Eriksson, D. (2017). Critical operations capabilities for
parison. Information Processing Letters, 57(1), 25–29. competitive manufacturing: A systematic review. Industrial Management & Data
Homburg, C., Fürst, A., & Prigge, J. K. (2010). A customer perspective on prod- Systems, 117(5), 801–837.
uct eliminations: How the removal of products affects customers and busi- Sellitto, M. A., Pereira, G. M., Borchardt, M., da Silva, R. I., & Viegas, C. V. (2015).
ness relationships. Journal of the Academy of Marketing Science, 38(5), 531–549. A SCOR-based model for supply chain performance measurement: Application
doi:10.10 07/s11747-0 09-0174-9. in the footwear industry. International Journal of Production Research, 53(16),
Irani, Z., Ghoneim, A., & Love, P. E. D. (2006). Evaluating cost taxonomies for infor- 4917–4926.
mation systems management. European Journal of Operational Research, 173(3), Shah, P. (2017a). Culling the brand portfolio: Brand deletion outcomes and success
1103–1122. factors. Management Research Review, 40(4), 370–377.
Joseph, J. F., Sundarakani, B., Hosie, P., & Nagarajan, S. (2010). Analysis of vendor Shah, P. (2017b). Why do firms delete brands? Insights from a qualitative study.
managed inventory practices for greater supply chain performance. International Journal of Marketing Management, 33(5–6), 446–463.
Journal of Logistics Economics and Globalisation, 2(4), 297–315. Shah, P., Laverie, D. A., & Davis, D. F. (2017). Brand deletion. Journal of Brand Strategy,
Katana, T., Eriksson, A., Hilletofth, P., & Eriksson, D. (2017). Decision model for prod- 5(4), 434–452.
uct rollover in manufacturing operations. Production Planning & Control, 28(15), Shah, P., Zhu, Q., & Sarkis, J. (2017). Product deletion and the supply chain: A green-
1264–1277. doi:10.1080/09537287.2017.1367859. ing perspective. In Proceedings of the Technology & Engineering Management Con-
Kirchoff, J. F., Koch, C., & Satinover Nichols, B. (2011). Stakeholder perceptions of ference (TEMSCON), 2017 IEEE (pp. 324–328).
green marketing: The effect of demand and supply integration. International Singh, S., Olugu, E. U., Musa, S. N., & Mahat, A. B. (2018). Fuzzy-based sustainabil-
Journal of Physical Distribution & Logistics Management, 41(7), 684–696. ity evaluation method for manufacturing SMEs using balanced scorecard frame-
Kirkwood, C. W. (1997). Strategic decision making. Wadsworth: Duxbury Press. work. Journal of Intelligent Manufacturing, 29(1), 1–18.
Kocaoğlu, B., Gülsün, B., & Tanyaş, M. (2013). A SCOR based approach for measuring Talon, A., & Curt, C. (2017). Selection of appropriate defuzzification methods: Appli-
a benchmarkable supply chain performance. Journal of Intelligent Manufacturing, cation to the assessment of dam performance. Expert Systems with Applications,
24(1), 113–132. doi:10.1007/s10845- 011- 0547- z. 70, 160–174. doi:10.1016/j.eswa.2016.09.004.
Koch, R. (2011). The 80/20 principle: The secret to achieving more with less. Crown Tan, C., & Chen, X. (2010). Intuitionistic fuzzy Choquet integral operator for multi-
Business. -criteria decision making. Expert Systems with Applications, 37(1), 149–157.
Leiter, K. M. (2011). Assessing and reducing product portfolio complexity in the phar- Tavana, M., Arteaga, F. J. S., Mohammadi, S., & Alimohammadi, M. (2017). A fuzzy
maceutical industry. Massachusetts Institute of Technology. multi-criteria spatial decision support system for solar farm location planning.
Liao, S.-H. (2005). Expert system methodologies and applications—A decade review Energy Strategy Reviews, 18, 93–105.
from 1995 to 2004. Expert Systems with Applications, 28(1), 93–103. Van Hoek, R., & Pegels, K. (2006). Growing by cutting SKUs at Clorox. Harvard Busi-
Lin, C.-T., Chiu, H., & Chu, P.-Y. (2006). Agility index in the supply chain. International ness Review, 84(4), 22.
Journal of Production Economics, 100(2), 285–299. Vyas, N. M. (1993). Industrial product elimination decisions: Some complex issues.
Lin, K., & Shih, L. (2010). An optimization model of product line rollover: A case European Journal of Marketing, 27(4), 58–76.
study of the notebook computer industry in Taiwan, 4(11), 2258–2268. Wagner, R., Abdelkafi, N., & Blecker, T. (2017). Exploration of the product phase-out
Lo, V. H. Y., & Yeung, A. H. W. (2004). Practical framework for strategic alliance in process in manufacturing firms: A human factor perspective. Business Process
Pearl River Delta manufacturing supply chain: A total quality approach. Interna- Management Journal, 23(5), 10 0 0–1017.
tional Journal of Production Economics, 87(3), 231–240. Wan, F., Wang, D., & Fung, R. Y. K. (2005). New product introduction planning using
Mamdani, E. H. (1974). Application of fuzzy algorithms for control of simple a 0–1 semi-infinite programming model. Production Planning & Control, 16(1),
dynamic plant. In Proceedings of the institution of electrical engineers: 121 12–20.
(pp. 1585–1588). Wu, M., Liu, K., & Yang, H. (2018). Supply chain production and delivery scheduling
Manders, J. H. M., Caniëls, M. C. J., & Ghijsen, P. W. T. (2016). Exploring supply chain based on data mining. Cluster Computing, 1–12. https://link.springer.com/article/
flexibility in a FMCG food supply chain. Journal of Purchasing and Supply Man- 10.1007%2Fs10586- 018- 1894- 8#citeas.
agement, 22(3), 181–195. doi:10.1016/j.pursup.2016.06.001. Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8, 338–353.
Mbhele, T. P. (2016). Decoupling paradigm of push-pull theory of oscillation in the Zadeh, L. A. (1973). Outline of a new approach to the analysis of complex sys-
FMCG industry. South African Journal of Business Management, 47(2), 53–66. tems and decision processes. IEEE Transactions on Systems, Man, and Cybernetics,
Mitchell, M. A., Taylor, R. D., & Tanyel, F. (1998). Product elimination decisions: A SMC-3(1), 28–44.
comparison of American and British manufacturing firms. International Journal Zhu, Q., & Shah, P. (2018). Product deletion and its impact on supply chain environ-
of Commerce and Management, 8(1), 8–27. mental sustainability. Resources, Conservation and Recycling, 132(January), 1–2.
Muir, J., & Reynolds, N. (2011). Product deletion: A critical overview and em- doi:10.1016/j.resconrec.2018.01.010.
pirical insight into this process. Journal of General Management, 37(1), 5–30.
doi:10.1177/030630701103700102.

You might also like