Computers & Industrial Engineering 87 (2015) 328–342

Contents lists available at ScienceDirect

Computers & Industrial Engineering

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

Modeling and multi-objective optimization of closed loop supply chains:

A case study
Eleonora Bottani ⇑, Roberto Montanari, Marta Rinaldi, Giuseppe Vignali
Department of Industrial Engineering, University of Parma, viale G.P.Usberti 181/A, 43124 Parma, Italy

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

Article history: This study investigates the issue of optimizing the asset management process in a real closed-loop supply
Received 7 January 2015 chain (CLSC), consisting of a pallet provider, a manufacturer and 7 retailers. A detailed simulation model,
Received in revised form 5 May 2015 based on an adapted economic order quantity (EOQ) policy is developed under Microsoft Excel™ to
Accepted 6 May 2015
reproduce the reorder process of assets by the manufacturer and the corresponding flow of returnable
Available online 27 May 2015
transport items (RTIs) in the CLSC. A multi-objective optimization, including both economic and strategic
key performance indicators of the system, is then carried out exploiting the commercial software
Keywords:
ModeFRONTIER™. The optimization investigates three scenarios, which refer to as many operating con-
Closed-loop supply chain (CLSC)
Returnable transport items (RTIs)
ditions of the manufacturer. Scenario 1 basically reproduces the current operating conditions of the man-
Simulation model ufacturer, while scenarios 2 and 3 are both hypothetical, and describe situations where the manufacturer
Multi-objective optimization would like to minimize the purchase of new assets and the pick-up of assets from its customers, respec-
Case study tively. For each scenario, the optimal configuration (i.e., the setting of the asset management process that
performs best in the multi-objective optimization) is identified. Scenarios 1 and 3 are found to generate
the most interesting performance of the assets management process, from both the economic and strate-
gic perspectives. Because the present paper is grounded on a real CLSC, the results are expected to be use-
ful to logistics and supply chain managers, to support the evaluation of the performance of CLSCs.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction semi-finished products and for the distribution of finished prod-

ucts. According to the European Commission (2007), RTIs are
Closed-loop supply chains (CLSCs) focus on managing the ‘means to assemble goods for transportation, storage, handling
returns of items (i.e., product and assets) from customers and and product protection in the supply chain which are returned
recovering added value by reusing them entirely and/or in some for further usage’. Among RTIs, pallets as well as all forms of reusa-
of their modules, components and parts (Guide & Van ble crates, totes, trays, boxes, roll pallets, roll cages, barrels, trol-
Wassenhove, 2009). Returns of items in a supply chain can occur leys, pallet collars, racks, lids and refillable liquid or gas
for a number of reasons. Following a product life-cycle perspective, containers can be mentioned (ISO/IEC, 2007).
Dekker, Inderfurth, Van Wassenhove, and Fleischmann (2004) and CLSCs have received increased attention in supply chain and
Flapper, van Nunen, and Van Wassenhove (2005) suggest the operations management literature. The main reasons for the
returns to be classified into production-, distribution-, use- and increased interest include the tightening of environmental regula-
end-of-life-related. Looking at the production and distribution per- tions and the business opportunities related to the residual value
spectives, commercial returns involve products that are returned of end-of-life products (Guide, Jayaraman, & Linton, 2003).
by consumers to the vendor, within some days after the purchase Indeed, with the awareness of the environmental protection
(Tibben-Lembke, 2004). End-of-use returns occur when a func- increasing, reducing the use of materials, by reusing RTIs and
tional product is replaced by a technological upgrade; hence, they remanufacturing the used products, is currently a critical issue
are particularly frequent when the product becomes technically for enterprises (Wang & Hsu, 2010). Moreover, besides the opera-
obsolete or no longer contains any utility for the current user tional and ecological benefits, RTIs are recognized as means to help
(Guide & Van Wassenhove, 2009). Additionally, in production- comply with waste regulation (Karkkainen, Ala-Risku, & Herold,
and distribution-related returns, returnable transport items 2004). To achieve these benefits, however, there is the need for
(RTIs) are used for internal transport of materials, components, procedures that ensure efficient and loss-free flows of RTIs in the
CLSC (Martınez-Sala, Egea-Lo pez, Garcıa-Sa nchez, & Garcıa-Haro,
⇑ Corresponding author. Tel.: +39 0521 905872; fax: +39 0521 905705.
2009).
E-mail address: [email protected] (E. Bottani).

http://dx.doi.org/10.1016/j.cie.2015.05.009
0360-8352/Ó 2015 Elsevier Ltd. All rights reserved.
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 329

To this latter extent, there are basically two procedures to han- containers and formulated a model for purchasing quantities of
dle the flow of RTIs within a CLSC, namely: (i) direct or deferred new containers in a returnable network. Choong, Cole, and
exchange between supply chain partners and (ii) asset pooling. In Kutanoglu (2002) and Di Francesco, Crainic, and Zuddas (2009)
(i), supply chain players own some RTIs, which are exchanged proposed models for repositioning of empty containers.
between all the actors of the chain. The exchange can be ‘direct’, Hellström and Johansson (2010) proposed a simulation model to
meaning that assets are returned immediately and in the same analyse the impact of the control strategy on the investment and
quantity of those shipped, or ‘deferred’, if the return is completed operating cost of RTIs.
at later time. Conversely, in (ii), a pool operator owns the RTIs Given the limited number of studies related to the optimization
and manages assets’ deliveries and returns (Johansson & of RTIs flows in the supply chain, and in real contexts in particular,
Hellstrom, 2007). The direct/deferred exchange is widely adopted this study tries to contribute to the literature by proposing a model
by many companies, owing to its simplicity and ease of implemen- to optimize the cost of RTIs management in a real CLSC. A detailed
tation; however, such procedure often suffers from several weak- model is developed to reproduce the CLSC, which consists of a pal-
nesses, namely the difficulty of checking the quality of the assets let provider, a manufacturer and 7 retailers, and to evaluate the
exchanged or the need for numerous paper documents to keep flows of RTIs in the system. The model is then reproduced on a
track of the inbound/outbound flow of RTIs. On the other hand, Microsoft Excel™ simulator and exploited for multi-objective opti-
one of the main advantages of pooling systems is the possibility mization purpose, supported by the commercial software
to outsource the RTI management, thus allowing producers to ModeFRONTIER™, to identify the optimal setting of the system
focus on their core activities, e.g. production processes (Bottani, under different operating conditions of the manufacturer.
Rizzi, & Vignali, in press). Moreover, pooled RTIs can be reused The remainder of the paper is organized as follows. Section 2
and do not add to the amount of items to be recycled or destroyed. describes the context where the present study was developed. In
Nonetheless, to work correctly, pooling systems require fast and Section 3, we detail the simulation model developed to reproduce
synchronized communications between partners (European the flow of RTIs in the targeted CLSC. The multi-objective optimiza-
Commission, 2007). tion procedure and the scenarios examined are illustrated in
Many cost components are associated with the use of RTIs in a Section 4. The main results from the simulation runs are discussed
CLSC. Rosenau, Twede, Mazzeo, and Singh (1996) provide a com- in Sections 5 and 6. Section 7 summarizes the main findings of the
prehensive list of those costs, which include, among others, pack- study, discusses implications and limitations and outlines future
aging material, damages, inbound and outbound transportation, research directions.
sorting, solid waste, tracking, labor, maintenance, ergonomics
and safety. Moreover, a RTI fleet often represents a significant cap-
ital investment for a company. Estimates by Aberdeen Group 2. The context
(2004) indicate that companies can lose more than 10% of their
RTIs fleet annually; this can generate a significant cost, considering 2.1. The CLSC analyzed
the unitary cost of RTI (approx. 10€ for pallets up to thousands of €
for containers). The considerations above suggest that RTIs are The core company of the CLSC examined in this study is a man-
often of high value, vulnerable to thefts or misplacements, critical ufacturer of fast moving consumer goods operating in the North of
to production and distribution activities, and have a relevant Italy; it will be referred to as Company A for confidentiality.
impact on the environmental performance of a company; there- Company A owns a stock of proprietary pallets, used for its ship-
fore, their management is particularly crucial (McKerrow, 1996; ments to 7 customers (delivery points), i.e. distribution centers of
Twede, 1999; Witt, 2000). fast moving consumer goods. Moreover, the CLSC includes a pallet
There is a large body of literature related to the design of CLSCs, provider that supplies new pallets to Company A when necessary.
either by means of simulation or by other optimization techniques. Shipments from/to the customers can be made either by Company
However, most of those studies focus on optimally balancing man- A or exploiting third party logistics (3PL) service providers. A
ufacturing and remanufacturing activities of items (see, e.g., scheme of the CLSC is proposed in Fig. 1.
Alinovi, Bottani, & Montanari, 2012; Georgiadis & Athanasiou,
2013; Zhang, Zhao, & Zhang, 2014, among recent works), including 2.2. The pallet management process
end-of-life considerations (Kim, Glock, & Kwon, 2014; Özkır &
Baslıgil, 2013; Tanimizu & Shimizu, 2014). The focus of this paper, The analysis carried out in this paper focuses expressively on
conversely, is on the optimal management of RTIs in CLSC. Company A, because of its key role in the asset management pro-
Literature related to this topic covers two main areas of interest. cess in the CLSC examined. The way pallets are currently managed
A first group of papers proposes the use of advanced ICT tools to in the CLSC examined is as follows. Company A receives orders (of
track the flows of RTIs in the CLSC. For instance, Thoroe, Melski, finished products) from its final customers. Fullfilling those orders
and Schumann (2009), Bottani and Bertolini (2009) and requires a certain amount of pallets, which are used by Company A
Martınez-Sala et al. (2009) discussed the use of radio-frequency to prepare the stock keeping units (SKUs) for shipment. More pre-
identification (RFID) technology to track different kinds of RTIs, cisely, pallets are used after picking operations, when products are
such as pallets or containers. A second group of studies proposes placed on the asset and packed. Once the order preparation is com-
decision support models for the cost-effective and efficient design plete, pallets are loaded on trucks to be transported to the different
of RTIs system. Since the present study focuses on optimizing the delivery points and shipped to the company’s customers. Here, the
cost of assets management in a CLSC, we have paid particular packaged SKUs (including pallets) will be unloaded and stored in
attention to this second group of works. Among them, the customer’s warehouse.
Mollenkopf, Closs, Twede, Lee, and Burgess (2005) proposed a In the current scenario, Company A adopts the deferred
model to compare the relative cost of an expendable container sys- exchange of pallets with its customers. Specifically, customers
tem to the cost of a reusable one, with the purpose of assessing the are not able to return the amount of pallets received to Company
viability of implementing a reusable container system in a refer- A immediately after receiving. More often, the pallets received will
ence scenario. Kroon and Vrijens (1995) presented a model for be used by the customers for a given time, to store products in
the placement and set-up of a logistics depot system. Kelle and their warehouse. At the same time, customers will return some
Silver (1989) considered the forecasting of returns of reusable empty pallets, currently available in their warehouse, but not
330 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

Pallet provider

Delivery point 1 Delivery point 7

Delivery point 2 Delivery point 6

Company A

Delivery point 3 Delivery point 5

Delivery point 4

Pallets damaged or lost Shipment to the customers

(outbound flow) (outbound flow)

Flow of new pallets Returned pallets

(indound flow) (inbound flow)

Fig. 1. The CLSC considered.

corresponding to the whole amount of pallets shipped. Hence, be too low to justify the retrieving. The order lot size for new
Company A will have to collect the remaining pallets from the cus- assets is fixed and corresponds, approximately, to a full truck
tomers at a later date. To this purpose, the manufacturer should load shipment of empty pallets;
keep track of its pallet flows, meaning that it should know the 3. purchase new pallets with urgency (urgent order). Such solu-
exact inbound/outbound flows of assets and the corresponding tion is exploited by Company A when an out-of-stock situation
debit of each customer. Moreover, to retrieve the remaining pal- is observed. In this case, the amount of pallets purchased covers
lets, Company A needs to organize dedicated trips, since it is not the exact need of the company and the purchasing cost is
always possible to exploit the return flows of subsequent ship- higher, because of the urgent delivery.
ments to pick pallets from the customers. Retrieving pallets from
the delivery points, however, is a very critical process. First of all, Obviously, both the retrieving operations and the purchase of
some pallets are inevitably lost during this process. The company new assets have a fixed lead time, meaning that the pallets will
has low control on the flow of assets damaged or lost during the be available at Company A after some days. The lead time of urgent
shipment; nonetheless, by assessing the difference between the orders, instead, is significantly lower. Overall, the pallet manage-
assets shipped and those returned, Company A estimates to lose ment process generates the following flows at Company A:
approx. 2.5% of pallets per cycle. This means that, for each ship-
ment to a customer, this percentage of pallets will not be returned – outbound flow of pallets used for shipments to the customers;
by that customer, neither during the interchange, neither in subse- – outbound flow of pallets lost or damaged during the shipment;
quent retrieving. Moreover, additional losses are generated by – inbound flow of new pallets, purchased from the pallet provider
damages of the pallet: according to the company’s estimates, this (either as regular or urgent orders);
accounts for an additional 1% of the total amount of pallets handled – inbound flow of returned pallets, i.e. pallets got back from cus-
per year. Overall, the flows of returned pallets are somehow tomers as a result of the deferred interchange or retrieving
affected by stochasticity, both in time and quantity, meaning that operations.
Company A does not exactly know how many pallets will be
returned and when the pallets will be back to the company. 2.3. Criticalities of the current process
At the same time, because the assets are used to ship products
to its customers, Company A should avoid out-of-stock situations. The current assets management process of Company A, as
Indeed, lack of pallets in stock means that Company A will not be described above, is characterized by some inefficiencies and could
able to ship products to its customers, resulting in a loss of sales. be improved in several ways. The main inefficiencies are listed
Hence, in the case the amount of assets currently in stock is not below:
sufficient to fullfil the orders of those customers, Company A will
either (in order of priority):  Despite the fact that Company A should properly balance its
inbound and outbound flows of assets, to avoid out-of-stock sit-
1. try to retrieve pallets from the customers (retrieving process). uations, the current inventory management process of the com-
On the basis of the flow of pallets shipped and got back during pany appears as non-optimized. Indeed, because the flow of
the exchange, Company A can estimate the amount of assets returned pallets is unknown and difficult to predict, the com-
available at each customer’s site. The customer that owns the pany is unable to apply any specific inventory management pol-
highest amount of assets could be selected by Company A to icy, which prevents the optimization of the asset inventory.
retrieve the pallets; Moreover, because of the presence of return flows, traditional
2. purchase new pallets from the pallet provider (regular order). inventory management policies (e.g., EOQ or EOI) could not
This alternative solution will be exploited whenever the even be directly applied; rather, those policies would need to
amount of assets available at the customers’ sites is found to be adapted to the case of a CLSC;
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 331

 we have mentioned that Company A should keep track of the Start of the process. At time t, Company A receives orders from
exact inbound/outbound flows of assets of each customer, but its delivery points. The company checks whether the whole
that the company has low control on the flow of assets damaged amount of orders received can be fullfilled exploiting its ‘physical’
or lost during the shipment and, therefore, cannot precisely stock of assets. The initial (t = 0) stock of assets is generated as a
P
know the exact amount of assets available at each customer. random number ranging from 0 to OA . If IPA ðt  1Þ P 7i¼1 ODP;i ðtÞ,
The logical consequence of this considerations is that Company A will follow the order fullfilment process (Branch 1);
Company A is unable to optimize the process of retrieving otherwise, it will incur in an out-of-stock situation (Branch 2).
assets from its final customers;
 as a final point, the Company would like to limit the use of P
Branch 1 (order fullfilment). Whenever IPA ðt  1Þ P 7i¼1 ODP;i ðtÞ,
urgent orders, because of the higher purchasing cost they
Company A will prepare the order and ship the pallets to its
generate.
customers. The stock of pallets available at Company A will be
subject to a first update, as follows:
Moving from the criticalities listed above, the analysis carried
out in this paper is intended to optimize the current asset manage- X
7

ment process of Company A, from the economic and strategic IPA;I ðtÞ ¼ IPA ðt  1Þ  SA;i ðtÞ ð1Þ
i¼1
perspectives.
where SA;i ðtÞ ¼ ODP;i ðtÞ. At the beginning of the process ðt ¼ 0Þ, the
3. Modeling framework ‘theoretical’ inventory position of Company A does not differ from
the physical one, i.e. IPA;I ðtÞ ¼ ITA;I ðtÞ. Once pallets are shipped, the
3.1. Decision process of Company A inventory of each delivery point will be checked, to assess whether
some assets can be returned through the deferred interchange. A
To model the reorder process of assets at Company A, we specific decision process, called ‘pallet exchange logic’ (cf.
started from the traditional EOQ policy (Harris, 1913), which was Section 3.1.1) is applied to this extent. The deferred interchange
modified and adapted to the presence of return flows. A scheme requires LT r days and leads to DIDP;i ðtÞ assets returned to
of the decision process of Company A resulting with the adapted Company A. The inventory of the delivery point is thus updated
EOQ policy is shown in Fig. 2. The notation in Table 1 is used to as follows (first update):
describe the process.
IPDP;i;I ðtÞ ¼ IPDP;i ðt  1Þ þ SA;i ðt  LT d Þ  DIDP;i ðt  LT r Þ
Because of the presence of several inbound/outbound flows, the
decision process is quite articulated, consisting, overall, of 6
branches and including several circumstances where the stock of ITDP;i;I ðtÞ ¼ ITDP;i ðt  1Þ þ SA;i ðtÞ  DIDP;i ðtÞ ð2Þ
assets of both Company A and its customers should be updated.
At the same time, pallets exchanged will be received at Company A,
The decision process is characterized by two operating leverages,
causing a second update of the company’s assets inventory, as
namely OP and MPQ. OP is the traditional order point of the EOQ
follows:
policy and is used to decide whether the stock of assets should
be replenished. MPQ, instead, denotes a minimum amount of pal- IPA;II ðtÞ ¼ IPA;I ðtÞ þ DIDP;i ðt  LT r Þ
lets that should be available for retrieving and is used to drive
the retrieving operations at the delivery points. The decision pro- ITA;II ðtÞ ¼ ITA;I ðtÞ þ DIDP;i ðtÞ ð3Þ
cess is detailed below with respect to the branches of the decision
tree.

Delivery point (final customer)

Start of the process

Second update of First update of the Applicaon of the pallet

Order Order received
the pallet stock pallet stock exchange logic

Company A Can the order be fulfilled with Y

Order preparaon
the current stock of assets? Branch 1 Pallet voucher
(order fulfillment)
N Branch 2
(out-of-stock) Pallet shipment
Out-of-stock Urgent order

N First update of the

End of the process Pallet stock ≤ order point?
Branch 1.1 pallet stock
(no replenishment)
Branch 1.2 Y
(asset replenishment)

Can assets be picked from the Y

stock of one customer?
Branch 1.2.1
(retrieving)
Branch 1.2.2 N
(reorder process)

Applicaon of the
Fih update of the Fourth update of Third update of the Second update of retrieving logic
pallet stock the pallet stock pallet stock the pallet stock
Physical flow (pallet)
Pallet retrieving from
Pallets received Pallets received Pallets received Pallets received Regular order Informaon flow
customers
(urgent order) (regular order) (retrieving) (exchange)

Order shipment

Regular order fulfilment Urgent order fulfilment

Pallet provider

Fig. 2. The decision process of Company A.

332 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

Table 1
Notation used for the model.

Subscripts
i Delivery point number (i = 1, . . . , 7) –
DP Delivery point –
A Company A –
I, II, III, IV, V Step of inventory update
Superscripts
P ‘Physical’ –
T ‘Theoretical’ –
Simulation parameters
t Simulation day (t = 0, . . . , N days ) –
N days Simulation duration [days]

Delivery point parameters

ODP;i ðtÞ Order issued [pallets]
IPDP;i ðtÞ; ITDP;i ðtÞ Physical and theoretical stock of assets [pallets]
DIDP;i ðtÞ; DIDP;i ðtÞ Theoretical and real amount of assets returned by deferred interchange [pallets]
RDP;i ðtÞ Amount of assets retrieved [pallets]
LDP;i ðtÞ; DDP;i ðtÞ Amounts of assets lost and damaged [pallets]
%L ; %D Percentage of assets lost and damaged [%]
cr;i Unitary retrieving cost [€/pallet]
dA;i Distance to Company A [km]
T stock;DP Time assets are in stock at the delivery point warehouse [days]

Company A parameters
SA;i ðtÞ Shipment to the i-th delivery point [pallets]
IPA ðtÞ; ITA ðtÞ Physical and theoretical stock of assets [pallets]
UOA ðtÞ; OA ðtÞ Amount of assets purchased through a urgent order or a regular order [pallets]
P A ðtÞ Amount of proprietary assets [pallets]
T stock;A Time assets are in stock at company’s warehouse [days]
OP Order point [pallets]
MPQ Minimum picked quantity, i.e. minimum amount of assets to be [pallets]
collected through retrieving operations
Economic parameters
ckm Unitary cost of transport of a truck [€/km/truck]
casset;r ; casset;u Cost of assets for regular and urgent orders [€/pallet]
cI Unitary cost of holding stocks [€/pallet/day]
Other parameters
npallets=truck Amount of pallets that can be loaded on a truck [pallets/truck]
ntrucks Number of trucks required for retrieving [trucks]
LT r Retrieving lead time [days]
LT d Delivery lead time [days]
LT o Lead time for regular orders [days]
LT u Lead time for urgent orders [days]

Once the inventory is updated, Company A will check whether the can be retrieved from the delivery points, so as to avoid purchasing
stock of assets is lower than OP. In line with the EOQ policy, the new assets. In general, delivery points always have assets than can
check is made on the theoretical inventory position (i.e., be retrieved; therefore, the check reduces to identify one delivery
 
ITA;II ðtÞ 6 OP), to take into account also those assets that have been point whose inventory exceeds the MPQ, i.e. 9iIDP;i ðtÞ P MPQ ; i
ordered, or exchanged, but are ‘in transit’ and have not yet been ¼ 1; . . . 7:g. The rationale behind the introduction of MPQ, which
received by Company A. If ITA;II ðtÞ > OP, Company A will not need does not formally exist in the current decision process of Company
to replenish its stock of assets (Branch 1.1), otherwise it will follow A, is essentially economic. Indeed, to retrieve pallets from the deliv-
the replenishment procedure (Branch 1.2). ery points, it is likely that Company A should organize a dedicated
trip, with related cost; this will be done only if that trip allows
retrieving at least a minimum amount of pallets. For a similar reason,
Branch 1.1 (no replenishment). If replenishment is not required, the possibility of organizing a trip to visit more than one delivery
the process ends. The ‘theoretical’ and ‘physical’ inventory posi- point to retrieve pallets is not considered, since, because of the dis-
tions of Company A and its customers are updated as follows: tance between the delivery points, the resulting cost would be
excessively high. This means that MPQ assets should be available
IPA ðtÞ ¼ IPA;II ðtÞ at one single delivery point. In the case one (or more) customers
own more than MPQ assets, the Company will start the retrieving
ITA ðtÞ ¼ ITA;II ðtÞ ð4Þ process (Branch 1.2.1); otherwise, it will follow the reorder process
(Branch 1.2.2).
IPDP;i ðtÞ ¼ IPDP;i;I ðtÞ
Branch 1.2.1 (retrieving). Company A will apply the ‘retrieving

logic’ to choose the i -th customer from which pallets should be
ITDP;i ðtÞ ¼ ITDP;i;I ðtÞ ð5Þ
retrieved (cf. Section 3.1.2). We recall that retrieving pallets from
the delivery points requires LT r days and the amount of pallets
Branch 1.2 (asset replenishment). In line with the decision process retrieved is RDP;i ðtÞ. The retrieving process causes a second update
described in Section 2.2, the company will first check whether assets of the inventory of the delivery points, as follows:
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 333

IPDP;i;II ðtÞ ¼ IPDP;i;I ðtÞ  RDP;i ðt  LT r Þ Branch 2 (out-of-stock). In the case the physical inventory of
assets at the company’s site does not allows fullfilling the orders
P
received from its customers, i.e. IPA ðt  1Þ < 7i¼1 ODP;i ðtÞ, Company

IPDP;i;II ðtÞ ¼ IPDP;i;I ðtÞ  RDP;i ðt  LT r Þ if i ¼ i A will incur in an out-of-stock situation. Under that circumstance,
the company will place an urgent order to the pallet provider.
Urgent orders are characterized by LT u < LT o , so that the new
IPDP;i;II ðtÞ ¼ IPDP;i;I ðtÞ assets are available in a shorter time compared to regular orders.
On the other hand, the cost of urgent orders is significantly higher
than that of regular orders. Because of this higher cost, the amount
IPDP;i;II ðtÞ ¼ IPDP;i;I ðtÞ otherwise ð6Þ of assets purchased urgently is limited to those strictly required to
fullfil the order, i.e.

The pallets retrieved from the i -th customer will cause a third
update of the inventory of Company A, as follows: X
7
UOA ðtÞ ¼ ODP;i ðtÞ  IPA ðt  1Þ ð12Þ
i¼1

IPA;III ðtÞ ¼ IPA;II ðtÞ þ RDP;i ðt  LT r Þ UOA ðtÞ also denotes the amount of out-of-stock experienced. The
urgent order generates a physical flow of pallets, from the pallet
provider to Company A, and therefore will cause the (fifth) update
IPA;III ðtÞ ¼ IPA;II ðtÞ þ RDP;i ðtÞ ð7Þ of the inventory position, as follows:
The update ends Branch 1.2.1 of the decision process. The IPA;V ðtÞ ¼ IPA ðt  1Þ þ UOA ðt  LT u Þ
inventory positions of Company A and its customers are finally
set at:
ITA;V ðtÞ ¼ ITA ðt  1Þ þ UOA ðtÞ ð13Þ

IPA ðtÞ ¼ IPA;III ðtÞ Therefore, following this branch of the decision process, the inven-
tory positions of Company A account for:

ITA ðtÞ ¼ ITA;III ðtÞ ð8Þ IPA ðtÞ ¼ IPA;V ðtÞ

ITA ðtÞ ¼ ITA;V ðtÞ ð14Þ

IPDP;i ðtÞ ¼ IPDP;i;II ðtÞ

3.1.1. The ‘pallet exchange logic’

ITDP;i ðtÞ ¼ ITDP;i;II ðtÞ ð9Þ As mentioned, anytime Company A ships pallets to a delivery
point, this latter can return some assets. However, the amount of
assets returned does not reflect the original shipment, because of
Branch 1.2.2 (reorder process). If none of the delivery points owns
   the pallets lost and damaged, as well as because the delivery point
a sufficient stock of assets (i.e. :9iIDP;i ðtÞ P MPQ ; i ¼ 1; . . . ; 7 ,
may not have the empty pallets available for being returned imme-
Company A will place a regular order for new assets to a pallet pro- diately. The amounts of pallets of Company A that are lost or dam-
vider. In line with the EOQ policy, the amount of pallets purchased aged at the i-th delivery point are estimated as follows:
through regular orders is fixed (i.e., OA ðtÞ ¼ OA ¼ cost;8t) and the
lot size should be preliminary determined by Company A. The pal- LDP;i ðtÞ ¼ SA;i ðt  LT s Þ  %L ð15Þ
let provider is assumed to have infinite availability of assets, mean-
ing that out-of-stock situations cannot occur for this player (this DDP;i ðtÞ ¼ SA;i ðt  LT s Þ  %D ð16Þ
also applies to urgent orders). Regular orders are available after
The amount of pallets the i-th delivery point can, theoretically,
LT o days and generate a physical flow of new pallets from the pallet
return by deferred exchange at time t accounts for:
provider to Company A. Hence, the inventory position of Company
A should be, once again, updated, according to the following for- DIDP;i ðtÞ ¼ IPDP;i ðtÞ  DDP;i ðtÞ  LDP;i ðtÞ ð17Þ
mula (fourth update):
However, to take into account the fact that some pallets can be
unavailable, the real amount of assets returned at time t is
IPA;IV ðtÞ ¼ IPA;II ðtÞ þ OA ðt  LT o Þ obtained as a random number, ranging from 0 and DIDP;i ðtÞ, i.e.:
h i
DIDP;i ðtÞ ¼ rnd 0; DIDP;i ðtÞ ð18Þ
ITA;IV ðtÞ ¼ ITA;II ðtÞ þ OA ðtÞ ð10Þ

This ends Branch 1.2.2 of the decision process. Therefore, the 3.1.2. The ‘retrieving logic’
inventory positions of Company A are finally set at: The ‘retrieving logic’ is as follows. Whenever only one delivery
point owns more than MPQ assets, Company A will retrieve pallets
IPA ðtÞ ¼ IPA;IV ðtÞ from that customer. Conversely, in the case more than one delivery
point owns a stock of assets higher than MPQ, Company A will have
to choose the customer from which pallets should be retrieved. The
ITA ðtÞ ¼ ITA;IV ðtÞ ð11Þ choice grounds mainly on economic considerations; specifically,
the logic is expected to identify the delivery point which will gen-
The inventory of the delivery points was not subject to updates erate the lowest retrieving cost, thus minimizing the total cost of
in this branch; therefore it is the same as described in Eq. (5). the CLSC.
334 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

To this extent, it is first necessary to estimate the unitary

PNdays  
t¼0 cI  IPA ðtÞ
retrieving cost [€/pallet] for each delivery point whose stock of CI ¼ ½€=day ð25Þ
N days
assets is higher than MPQ. The computation is as follows:
1 5. Opportunity cost ðC opp Þ, which reflects the immobilization of
cr;i ¼ dA;i  ckm  ntrucks  8ijIPDP;i ðtÞ P MPQ ð19Þ capital, used to purchase pallets, with the sacrifice of alternative
IPDP;i ðtÞ
investments. It is quantified starting from the economic value of
In turn, ntrucks [trucks] is computed by rounding to the next the assets Company A owns at a given time and the weighted
whole number the following expression: average capital cost (WACC). The result is averaged over the
time horizon considered. The economic value of assets CP A ðtÞ
IPDP;i ðtÞ is computed as follows:
ntrucks ¼ ð20Þ
npallets=truck
CP A ð0Þ ¼ IPA ð0Þ  casset;r
where npallets=truck is a fixed quantity describing the amount of pallets
that can be loaded on a truck of a given capacity. Once cr;i has been
CP A ðtÞ ¼ CP A ðt  1Þ þ UOA ðtÞ  casset;u þ OA ðtÞ  casset;r ½€
computed for all the relevant delivery points, Company A will select

the i -th delivery point that minimizes the retrieving cost, i.e. ð26Þ
n o

i jcr;i ¼ min cr;i and IPDP;i ðtÞ P MPQ . The amount of assets C opp thus results from the following formula:

retrieved from the i -th delivery point will account for the entire PNdays
t¼0 CPA ðtÞ  WACC
stock of pallets available at that customer, i.e.: C opp ¼ ½€=day ð27Þ
Ndays
RDP;i ðtÞ ¼ IPDP;i ðtÞ ð21Þ
6. Total cost ðC tot Þ, which reflects the total cost Company A incurs
Combining the above formula with Eq. (6)., it is easy to deduce in. It is obtained adding up all the contributions above, accord-

that, after retrieving, the inventory of the i -th delivery point will ing to the following formula:
score 0. C tot ¼ C opp þ C I þ C p;u þ C p;r þ C r ½€=day ð28Þ

3.2. Key performance indicators

3.2.2. Strategic KPIs
The key performance indicators (KPIs) used to assess the perfor- With respect to the strategic aspects of the pallet management
mance of Company A against the pallet management process cover process, the following KPIs are considered in this study:
both economic aspects and strategic ones.
1. Amount of proprietary assets ðP A Þ, which reflects the average
3.2.1. Economic KPIs number of pallets owned by Company A and is expressed in
Given the characteristics of the CLSC analyzed and the decision [pallets]. As mentioned, PA is computed starting from the phys-
process described above, the following economic KPIs are ical inventory of Company A and adding the amount of pallets
computed: the company has shipped to its customers, as follows:
PNdays  P P 
1. Cost of retrieving ðC r Þ, which reflects the cost for picking up pal- t¼0 IA ðtÞ þ 7i¼1 SA;i ðtÞ
lets from the delivery points, according to the ‘pallet retrieving PA ¼ ½pallets ð29Þ
Ndays
logic’. We compute its daily value as follows:
2. Asset rotation (AR): the asset rotation measures the number of
PNdays  
times per year where the pallets rotates and is, therefore,
t¼0 cr;i  IPDP;i ðtÞ
Cr ¼ ½€=day ð22Þ expressed in [year1]. This KPI is a measure of the efficiency
Ndays with which a company is deploying its own assets: the higher
2. Cost of purchasing – regular order ðC p;r Þ, which is the cost for the number of rotations, the higher the capacity of Company
purchasing new pallets from the pallets provider, in the case A to exploit its proprietary assets. AR can be computed as the
of a regular order. It is computed starting from the amount of inverse of the cycle time CT [days], which measures the time
pallets purchased and the unitary cost of assets and averaged required for an asset to complete a cycle in the CLSC considered.
over the time horizon considered, as follows: A complete ‘cycle’ consists of the following processes: storage at
PNdays the warehouse of Company A; shipment to the customer; stor-
t¼0 ðcasset;r  OA ðtÞÞ age at the customer’s warehouse; shipment back to Company A.
C p;r ¼ ½€=day ð23Þ
Ndays The same asset cannot be shipped from one customer to
another, so that the cycle time can include the contribution of
3. Cost of purchasing – urgent order ðC p;u Þ, which is the cost to
only one customer. Given the description above, the CT can be
purchase new pallets from the pallets provider, in the case of
computed starting from the following formula:
urgent orders. It is computed starting from the amount of pal-
lets purchased urgently and the unitary cost of urgent assets CT ¼ T stock;A þ T stock;DP þ LT r þ LT d ½days ð30Þ
and averaged over the time horizon considered, as follows:
T stock;A can be estimated starting from the average stock of assets
PNdays
t¼0 ðcasset;u  UOA ðtÞÞ at Company A and the average amount of pallets shipped daily
C p;u ¼ ½€=day ð24Þ by the company, as follows:
Ndays
PNdays P
4. Inventory cost ðC I Þ, which reflects the cost of storing pallets in IA ðtÞ=Ndays
T stock;A ¼ PN t¼0 P7 ½days ð31Þ
the warehouse of Company A. It is computed starting from days
t¼0 i¼1 SA;i ðtÞ=N days
the physical inventory of assets at Company A and the unitary
cost of holding the stock of assets and averaged over the time Similarly, T stock;DP can be estimated as the ratio between the aver-
horizon considered, i.e.: age stock of assets available at the delivery points and the
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 335

average amount of assets shipped or retrieved from the same in 2013 and available in the first spreadsheet. As an example, for
points, i.e.: customer 1, the demand ranges from 0 to 78 pallets/day. The
PNdays P7 P aggregated demands from the final customers compose the
i¼1 I DP;i ðtÞ=N days ODP;i ðtÞ and are used as the input for the whole model (Eqs. (1)–
T stock;DP ¼ PN P t¼0 P  ½days
days 7 (21)).
t¼0 i¼1 SA;i ðtÞ þ i¼i RDP;i ðtÞ =N days
The data related to the inbound flows of Company A in 2013 are
ð32Þ used to assess the correctness of the deferred exchange logic used
Once CT has been determined, AR can be computed according to in the model. More precisely, as can be seen from Eq. (18), the
the following formula (assuming 260 working days per year): amount of assets a customer can return to Company A is estimated
as a random number, ranging from 0 to the theoretical availability
1 of assets of the customer. The same computation is implemented
AR ¼  260 ½year1  ð33Þ
CT in Microsoft Excel™, meaning that DIDP;i ðtÞ is generated as a ran-
3. Pallet utilization rate ðU % Þ: this KPIs is computed as the ratio dom number. This latter is compared to the real inbound flow of
between the time [days] in which the asset is used in the Company A, to ensure that the randomly generated data are in line
CLSC, meaning that it is either at the delivery point or in transit with the real scenario.
(but not in the warehouse of Company A to store goods), and The input data listed in Table 1 were also set in the simulation.
the CT [days] of the asset. It is, therefore, expressed in [%], As far as the OP and MPQ are concerned, none of those parameters
according to the following formula: is used by Company A in its current decision process. Therefore,
they were varied in a range of possible values, to identify the opti-
T stock;DP þ LT r þ LT s mal (i.e., minimum cost) configuration of the asset management
U% ¼ ½% ð34Þ
CT process. Specifically, OP and MPQ were varied from 50 to 1200
4. Out-of-stock (OOS): this KPI is intended to provide a quantita- (pallets). For each combination of OP and MPQ, the spreadsheet
tive measure of those critical situations where Company A does computes the economic and strategic KPIs, according to
not have pallets available to ship products to the final customer. Eq. (22)–(31). The simulation duration was set at N days = 3000 days.
We measure the number of days per year where the OOS situa- When computing the KPIs, outputs were collected starting from
tion is observed, i.e.: day t = 150, to limit warm-up effects of the simulation.

number of out-of-stock situations

OOS ¼  260 ½days=year
Ndays 4. Simulation and optimization
ð35Þ
The simulation model described above was used to evaluate
three different scenarios of the CLSC, which were considered of
3.3. Input data particular interest for Company A. The scenarios differ with respect
to the KPIs that are chosen for optimization and reflect the situa-
To apply the model described in the previous sub-sections to tion where the company is interested in optimizing some specific
the targeted CLSC, several input data were collected from aspects of the pallet management process. The multi-objective
Company A. They are summarized in Table 2. optimization procedure was supported by the commercial soft-
ware ModeFRONTIER™ release 4.5.4 (Esteco S.p.A.).
3.4. Software implementation

The set of equations described in Section 3.1 (and related 4.1. Optimization procedure
sub-sections) was embodied in a Microsoft Excel™ simulation
model. The simulation model consists of two spreadsheets. The ModeFRONTIER™ is a multi-objective optimization and process
first one is the complete database of the inbound/outbound flows integration tool based on the Pareto-optimal frontier of the objec-
Company A handled in 2013, i.e. the shipments to the delivery tives space. The logic of the optimization loop can be set up in a
points and the pallets returned by them. The second spreadsheet graphical way, building up a workflow structure by means of inter-
reproduces the decision process of Company A. The demand from connected nodes. The optimization starts from the interaction of
the delivery points is modeled as a random number and ranges, ModeFRONTIER™ with the Microsoft Excel™ simulation model,
for each customer, from 0 to the maximum demand value recorded with some fixed input data (cf. Table 1) and two variable parame-
ters, i.e. OP and MPQ. The first variation of OP and MPQ in their
Table 2
range was made by setting a Design of Experiments (DoE) proce-
Fixed input data for the CLSC examined.
dure, with a 32 full factorial design, which takes the boundary val-
LT d 2 Days ues and the intermediate one for each parameter (i.e., 50, 625 and
LT r 2 Days
1200). The DoE provides a preliminary investigation of the design
LT o 2 Days
LT u 1 Days space and is useful to guide the subsequent non-dominated sorting
OA 500 Pallets genetic algorithm (NSGA-II) towards the optimal solution. Overall,
%L 2.5% – the NSGA-II carries out 270 simulations. This number is signifi-
%D 1% –
cantly lower than that required if the simulation model was simply
dA;i 362 (i = 1); 358 (i = 2); 606 (i = 3); 352 (i = 4); 232 km
(i = 5); 934 (i = 6); 632 (i = 7)
run under Microsoft Excel™ by varying the OP and MPQ in their
ckm 1.86 €/km/truck range, with the purpose of identifying the optimal configuration.
npallets=truck 500 pallets/ For instance, varying OP and MPQ from 50 to 1200 (step 50) would
truck lead to 24 ⁄ 24 = 576 simulation runs. Indeed, NSGA-II is a fast and
casset;r 9 €/pallet
elitist multi-objective genetic algorithm that allows solving high
casset;u 45 €/pallet
cI 0.025 €/pallet/day complexity multi-objective optimization problems with a rela-
WACC 5% – tively limited number of iterations, preserving elitist solutions
(Deb, Meyarivan, Pratap, & Agarwal, 2002).
336 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

4.2. Simulation scenarios (a) 700

The multi-objective optimization procedure was used to exam- 600

ine three different scenarios of the asset management process of
Company A; these scenarios are briefly described below. 500

Ctot [€/day]
– Scenario 1: in this scenario, the performance parameters set for 400
the multi-objective optimization are C tot (minimum), U % (max-
imum), AR (maximum) and OOS (minimum); 300
– Scenario 2: this scenario is the same as the first one for the
strategic KPIs, while the cost component considered in the opti- 200
mization is the total cost of purchasing new assets, i.e. C p;r þ C p;u
(minimum); 100
– Scenario 3: this scenario is the same as the previous ones for the 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
strategic KPIs, while the cost component considered in the opti- OP [pallets]
mization is the cost of retrieving assets from the customers, i.e.
C r (minimum). (b) 700

600
The analysis was focused on those scenarios because of the fol-
lowing reasons. Scenario 1 basically reflects the current operating
500
conditions of Company A: the primary goal of the company is to

Ctot [€/day]
minimize the total cost of its asset management process, but, at
the same time, the company tries to optimize the assets usage 400

and rotations and to avoid out-of-stock situations. Therefore,

results of the multi-objective optimization for scenario 1 are 300

expected to provide Company A with guidelines for the optimiza-

200
tion of its current asset management process. Scenario 2 reflects
a possible situation where Company A would like to avoid (or
100
limit) the purchase of new assets, with the purpose of enhancing
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
the use of its proprietary pallets. The solutions resulting from the
MPQ [pallets]
multi-objective optimization under this scenario are expected to
improve the utilization of the proprietary assets of Company A, Fig. 3. Trend of C tot as a function of OP (a) and MPQ (b) – scenario 1.
at the same time identifying potential benefits resulting with a
strategy of asset management different from the current one. The
last scenario reflects a possible situation where Company A is no of the correlation analysis, are similar across the scenarios, we pro-
longer able to manage the reverse flows of its assets. Because vide the detailed outcomes for scenario 1, while, for the remaining
managing the reverse flows of assets is, currently, a critical process, scenarios, we discuss some selected outcomes.
it is likely that, in the near future, Company A will leave the man-
agement of this process. Hence, it is useful to Company A to have 5.1. Scenario 1
an estimate of the total cost of assets management in the case
the Company will privilege the purchasing of new pallets against Fig. 3 shows the trend of C tot as a function of OP (a) and MPQ (b).
returning operation.1 From Fig. 3(a) it is easy to see that the design space includes solu-
For all scenarios, a constraint related to the amount of propri- tions with low C tot (approx. 200 €/day) obtained setting very differ-
etary pallets was added in ModeFRONTIER™, so as to consider ent OP (from 100 to 1100 pallets). At the same time, however, some
the finished storage capacity of Company A. Accordingly, we forced configurations with OP  600–700 pallets experience very high
PA < 3000 pallets and limited the analysis to those solutions that C tot . This would suggest that those variables do not have a direct
meet this constraint. Also, only integer values were allowed for this relationship. Indeed, OP has a different impact on the cost compo-
parameter, as well as for OP and MPQ. nents included in C tot . Specifically, increasing OP means that
Company A will place orders for new pallets when a higher stock
5. Multi-objective optimization results of assets is available; consequently, the average inventory level
increases, thus increasing C I . At the same time, orders will be
In this section, we report the main results of the simulation placed more frequently, increasing C p;r . Moreover, because of the
runs, in terms of the trend of the KPIs optimized as a function of higher number of regular orders placed, Company A would (prob-
the operating leverages of the decision process, i.e. OP and MPQ. ably) have lower need for placing urgent orders, generating a lower
The results provided aim at identifying the existence of relation- C p;r . The reduced number of urgent orders generates a decreased
ships between the reorder policy parameters and the model out- C opp , since, despite the fact that regular orders are characterized
puts. Because those relationships cannot be immediately evident by a high number of assets, those assets have a lower value.
from the model description in Section 3, outcomes are substanti- Those relationships are confirmed by the correlation analysis in
ated by a correlation analysis, whose detailed results are reported Appendix A (Table A.1). The same analysis highlights that the cor-
in Appendix A. Because the trends observed, as well as the results relation between OP and C tot , although significant, is very weak
(0.330), thus confirming that these variables do not exhibit a
1
With respect to the number of scenarios analyzed, the discussion in Section 5 and direct relationship (Taylor, 1990). With respect to the trend of
the correlation analysis in Appendix A show that the trends of almost all the model C tot as a function of MPQ, outcomes in Fig. 3(b) show a more evi-
outputs correlate to each other. Because of this correlation, it is likely that, even if we
analyse more than three scenarios with some modifications in the KPIs to optimize,
dent relationship between those variables. Specifically, very high
the results will partially overlap with those presented in Section 5. We have therefore C tot are observed with MPQ < 300 pallets, while, for higher values
limited the analysis to three scenarios with the purpose of avoiding repetitions. of MPQ, C tot is stable and accounts for less than 200 €/year.
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 337

(a) 50 (a) 40

40
30
AR [rotaons/year]

30

OOS [days/year]
20
20

10
10

0 0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
OP [pallets]
OP [pallets]

(b) 50
(b) 40

40
AR [rotaons/year]

30
30

OOS [days/year]
20
20

10 10

0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
MPQ [pallets]
MPQ [pallets]
Fig. 4. Trend of AR as a function of OP (a) and MPQ (b) – scenario 1.
Fig. 5. Trend of OOS as a function of OP (a) and MPQ (b) – scenario 1.

placed, so that out-of-stock situations are less likely to occur.

Correlation analysis confirms that the negative relationships
This is also confirmed by the good correlation coefficient between
between OP and C tot is strong (0.735). This result is the combina-
those variables (0.714). From Fig. 5(a) it can be seen that config-
tion of the different effects MPQ has on the cost components.
urations with OP > 300 pallets always generate null OOS, while for
Specifically, the increase in MPQ generates a higher C opp , because
OP < 200 pallets OOS situations are more likely to occur. The rela-
of the presence of more assets at Company A, which, in turn,
tionship between OOS and MPQ is less evident and also weaker,
increases C I too. Similarly, if more pallets are picked from the cus-
as shown by the lower correlation coefficient (0.370). In general,
tomers in a single shipment, the frequency of picks from the cus-
lower MPQ seems to generate higher OOS; this is probably due to
tomers would be lowered, thus decreasing C r . Under that
the fact that increasing MPQ increases the amount of proprietary
circumstance, Company A would purchase new pallets through
assets of Company A, thus making OOS situations less likely to
regular orders, thus justifying the positive correlations with C p;r .
occur. Nonetheless, some configurations with MPQ < 100 pallets
Fig. 4 shows the trend of AR as a function of OP (a) and MPQ (b).
generate null OOS and, at the same time, some configurations with
Fig. 4(a) shows a trend which is similar to that of Fig. 3(a), in that
higher MPQ (from 300 to 400) generate quite high OOS.
OP and AR do not seem to be directly related. Indeed, the design
As far as the relationship between U % and OP is concerned,
space shows that the highest AR (approx. 40 rotations/year) is
Fig. 6(a) shows that U % tends to decrease with the increase in
obtained with OP < 100 pallets, but there are also configurations
OP. As already mentioned, increasing OP involves a corresponding
with OP < 100 pallets which generate a significantly lower AR.
increase in the number of proprietary assets of Company A, and,
The correlation coefficient (Table A.1) shows a modest negative
consequently, a high CT of assets. Since U % is computed starting
relationship between OP and AR (0.453). This could be motivated
from the inverse of CT, is decreases accordingly. This is also con-
considering that increasing OP involves a corresponding increase in
firmed by a high correlation coefficient between those variables
the number of proprietary assets of Company A, and thus an
(0.903). The relationship between U % and MPQ seems to be oppo-
increase in the CT of assets (corresponding to a decrease in AR).
site (Fig. 6(b)), although the correlation between those variables is
The relationship between AR and MPQ is stronger (0.799) and evi-
very weak (0.159).
dent from Fig. 4(b). When the MPQ is higher, Company A will
decrease retrieving operations, but, at the same time, will increase
the number of regular orders, resulting in an increased number of 5.2. Scenario 2
proprietary assets. As per the previous variables, with more assets
available at Company A, the average cycle time of assets will be The trends of U % , AR and OOS as a function of OP and MPQ
higher, resulting in a lower AR. obtained under scenario 2 reflect those described for the previous
The trend of OOS as a function of OP and MPQ is proposed in scenario, indicating that the relationships between those variables
Fig. 5(a) and (b). The relationship between OP and OOS is obvious: is almost the same under the two scenarios. This is also confirmed
higher OP generates higher inventory available when orders are by the correlation analysis in Table A.2, which shows that the
338 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

(a) 1.0 (a) 210

Feasible
Unfeasible 190
0.9 Error
170

0.8 150

Cp,u + Cp,r [€/day]

130
U% [%]

0.7
110

90
0.6
70
0.5 50

30
0.4 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
OP [pallets]
OP [pallets]

1.0
(b) 210
(b) Feasible
190 Unfeasible
Error
0.9
170

Cp,u + Cp,r [€/day]

150
0.8
130
U% [%]

0.7 110

90
0.6
70

0.5 50

30
0.4 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 MPQ [pallets]
MPQ [pallets]
Fig. 7. Trend of C p;r þ C p;u as a function of OP (a) and MPQ (b) – scenario 2.
Fig. 6. Trend of U % as a function of OP (a) and MPQ (b) – scenario 1.

correlation coefficients of scenario 2 are similar to those of scenario (a) 500

1, thus suggesting the same trends. Therefore, for brevity, we limit
the discussion of the outcomes to the trend of the economic KPI
400
(i.e., the total purchasing cost) as a function of OP and MPQ
(Fig. 7). From Fig. 7(a) one can appreciate that low OP generates
Cr [€/day]

a high purchasing cost, while for OP > 300 the purchasing cost is
almost constant and accounts for approx. 40 €/day; overall, this 300

trend suggests a negative correlation between these variables.

However, looking at the correlation coefficients (Table A.2), it can
be seen that OP has a different effect on C p;r and C p;u ; in particular, 200

C p;r increases with the increase in OP, while C p;u decreases. This
outcome was partly expected: indeed, increasing the OP forces
100
Company A to place more orders (regular), which increases the cor-
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
responding cost. Moreover, the similarity between the correlation
OP [pallets]
coefficients (0.581 vs. 0.580) suggests that C p;r and C p;u are almost
perfectly negatively correlated. This means that whenever (b) 500 Feasible
Company A makes use regular orders, urgent orders will be corre- Unfeasible
spondingly less required (and vice versa). Error

The trend of C p;r þ C p;u as a function of MPQ (Fig. 7(b)) is again 400
negative, although less evident. Low MPQ seems to generate the
highest purchasing cost; however, there is not a strict correspon-
Cr [€/day]

dence between the MPQ and the resulting purchasing cost. The cor- 300
relation analysis confirms this consideration, since both correlation
coefficients are quite low (0.324 and 0.323). As per the previous
case, the effect of MPQ against C p;r and C p;u is opposite. It should also 200
be noted that MPQ > 600 always generates unfeasible solutions.

5.3. Scenario 3 100

0 100 200 300 400 500 600 700 800 900 1000 1100 1200
MPQ [pallets]
As per the previous scenario, the trends of U % , AR and OOS as a
function of OP and MPQ obtained under scenario 3 reflect those Fig. 8. Trend of as a function of OP (a) and MPQ (b) – scenario 3.
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 339

described for the scenario 1. The similar trends are also confirmed generate null OOS and do not require urgent orders (C p;u = 0). The
by the correlation analysis in Table A.3. The related description is amount of proprietary assets is quite high, always exceeding
omitted, for brevity. We focus, instead, on the trend of the eco- 2170 pallets. The first ranked configuration provides the lowest
nomic KPI (i.e., the retrieving cost) as a function of OP and MPQ total cost (182.39 €/day), a significant part of which is due to
(Fig. 8). From Fig. 8(a) it is immediate to see that the C r does not retrieving operations (126.67 €/day). Such configuration is
have any specific relationship with the OP set in the simulation, obtained setting OP = 310 pallets and MPQ = 372 pallets.
meaning that this latter does not affect the retrieving cost to an Nonetheless, OP and MPQ of all configurations vary in a very lim-
appreciable extent. Indeed, the correlation between those variables ited range (from 306 to 312 and from 372 to 386 respectively).
is very weak (0.174). Conversely, MPQ and C r show a more evi-
dent negative relationship, which is also confirmed by the good 6.2. Scenario 2
correlation coefficient (0.755). Such outcome was expected,
because, as already remarked, MPQ has a direct role in driving Table 4 shows the top 5 simulation runs of scenario 2, after
retrieving operations of Company A. ranking. We recall that, in this scenario, Company A is interested
in minimizing C p;r þ C p;u , while the strategic KPIs are the same as
6. Multi-criteria decision making in the previous scenario. The top 5 configurations of Table 4 are
similar in terms of U % (from 83.04% to 85.03%), while significant
The multi-criteria decision making (MCDM) tool of differences can be found against the remaining KPIs. Indeed, the
ModeFRONTIER™ was exploited to rank the 270 configurations of top 2 configurations shows the minimum cost of purchasing (less
each scenario in a rational order and to identify the best dyad than 39 €/day and entirely due to regular orders) as well as a null
MPQ-OP of each scenario. This tool works exactly as the traditional OOS; AR ranges from 11 to 12 rotations/year. Those configurations
multi-criteria decision analysis of operational research. are obtained setting quite high OP and MPQ, accounting for more
Specifically, the different objectives set in the optimization model than 300 and more than 400 pallets, respectively. The remaining
are treated as (conflicting) criteria, and the simulation runs are 3 configurations, instead, are obtained setting significantly lower
ranked on the basis of their score against those criteria, to identify OP and MPQ (less than 200 and less than 80, respectively) and gen-
the optimal configuration. A linear MCMD model was selected, erate a higher AR (more than 30 rotations/year). However, the cost
with the following weights: of purchasing is always higher than 40 €/day, due to the presence
of urgent orders, and thus of OOS situations. Because of the low
 0.40 for the cost criterion. Such a weight reflects the relevance MPQ, under those configurations Company A will be forced to per-
of the economic considerations in the asset management form very frequent retrieving operations at the delivery points,
process; resulting in a high C r . In turn, this leads to a poor performance in
 0.30 for OOS. This choice is motivated by the fact that, as already terms of C tot , which is approx. double compared to the top 2
mentioned, OOS should be possibly avoided for Company A; configurations.
 0.15 for the remaining KPIs, i.e. AR and U % . The first ranked configuration is obtained setting OP = 329 pal-
lets and MPQ = 434 pallets and, overall, appears as the most inter-
The application of the MCDM tool was limited to the feasible esting for a practical implementation, because of the good U % , the
configurations of each scenario. null OOS and the low cost of purchasing.

6.1. Scenario 1 6.3. Scenario 3

Table 3 provides an extract of the top 5 simulation runs of sce- Table 5 lists the top 5 simulation runs of scenario 3, after rank-
nario 1, as they were ranked after the application of the MCDM ing. In this scenario, Company A is interested in minimizing C r , as
tool. Gray highlighting indicates the KPIs that were optimized in well as the same strategic KPIs as in the previous scenarios. A first
this scenario. From Table 3 it can be seen that the top 5 configura- consideration from Table 5 is that the U % is quite similar in all the
tions are quite similar in terms of U % and AR, which are always configurations proposed, ranging from 86.77% to 89.06%, and is sig-
close to 83% and 13 rotations/year, respectively. All configurations nificantly higher compared to the previous scenarios. All

Table 3
Optimal configurations for scenario 1 (Note: gray highlighting = KPIs optimized).

Table 4
Optimal configurations for scenario 2 (Note: gray highlighting = KPIs optimized).
340 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

Table 5
Optimal configurations for scenario 3 (Note: gray highlighting = KPIs optimized).

configurations generate a low OOS, with configuration 1 generating able to retrieve assets from its customers; therefore, the retrieving
null OOS, which could be particularly interesting for a practical cost should be minimized. The optimal setting for this scenario is
implementation. The top 3 configurations are similar in terms of OP = 177 pallets and MPQ = 340 pallets; this setting generates a
the MPQ (from 338 to 340 pallets) and AR (approx. 15 rota- retrieving cost of approx. 128 €/day, resulting in a total cost of
tions/year), while different values are observed in the remaining approx. 180 €/day, which is in line with the optimal values
configurations. obtained in scenario 1. In fact, scenarios 1 and 3 are quite similar
The first ranked configuration is obtained setting in terms to the cost they generate, as the cost of retrieving is the
OP = 177 pallets and MPQ = 340 pallets. This setting generate the most significant part of the total cost of assets management for
minimum cost of retrieving (127.75 €/day), but, overall, this cost Company A.
does not vary significantly from configuration 1 to configuration From a practical perspective, the results summarized above
5, suggesting that Company A could also select an alternative set- provide Company A with an overview of the performance of its
ting and would get similar performance. By comparing these out- asset management process and can be useful in the case the com-
comes with those of scenario 1 (cf. Table 3), it is immediate to pany is interested in changing its current asset management pol-
see that results, in terms of C r and C tot are very similar, meaning icy, by defining a different strategy (e.g., minimizing the
that, when minimizing, e.g., the cost of retrieving, the total cost purchase of new assets or retrieving operations). In this regard,
of asset management will be minimized as well. In turn, this is results obtained in scenario 2 indicate that a strategy aimed at
due to the fact that the cost of retrieving is the most significant reducing the purchase of new assets, by enhancing the use of pro-
part of the total cost of assets management for Company A, so that prietary pallets, would be sustainable from the economic perspec-
C tot has almost the same trend of C r . tive only if Company A sets adequate values of OP and MPQ.
Conversely, a strategy aimed at reducing (or avoiding) retrieving
7. Discussion and conclusions operations at the customer’s sites turns out to be suitable for
implementation by Company A, since such strategy would also
This study has proposed a comprehensive analysis of the perfor- optimize the total cost of assets management for the company.
mance of the asset management process in a real CLSC, consisting Therefore, it would not be problematic, for Company A, to embrace
of a pallet provider, a manufacturer (Company A) and seven retail- this new strategy.
ers. The analysis was supported by a Microsoft Excel™ simulation From the theoretical perspective, the model developed in this
model, which reproduces, through an adapted EOQ policy, the paper is quite detailed, including an articulated decision process
asset management process of Company A and computes the corre- and two optimization logics. Such a model could be used also to
sponding cost and performance. The model was subsequently used reproduce different settings or operating conditions of Company
in a multi-objective optimization procedure, supported by A or, at the same time, it could be adapted to analyse other compa-
ModeFRONTIER™, which examined three different operating con- nies or CLSCs, characterized, for instance, by a different number of
ditions of Company A and generated, for each scenario, the optimal customers. Similarly, the multi-objective optimization procedure
configuration of the reorder policy setting for the manufacturer. could be applied with different settings of the performance param-
The scenarios investigated describe either the current operating eters, so as to explore additional assets management strategies.
conditions of Company A or reflect potential operating conditions Therefore, the model itself represents an interesting addition to
that could be implemented by the company in the near future. the literature about CLSCs.
In particular, under the first scenario of the multi-objective Starting from this study, several future research directions
optimization, which reproduces the current situation of Company could be undertaken. As mentioned, the model developed could
A, we found that the best performance are achieved setting be applied for the analysis of a different CLSC, in terms of input
OP = 310 pallets and MPQ = 372 pallets in the decision process of data or supply chain structure, with the purpose of analyzing the
Company A. Under this configuration, the total cost of asset man- performance of systems different to that investigated. As a further
agement is approx. 182 €/day. Scenario 2 reflects a potential situa- research direction, the model developed in this paper could be
tion where Company A would like to enhance the use of its exploited to investigate how the performance of the CLSC changes
proprietary pallets, thus minimizing the purchase of new assets. as a function of some of the input data. In this study, we used input
The optimal setting of the decision process is OP = 329 pallets data from a real company; those data were not altered to preserve
and MPQ = 434 pallets: such setting generates null OOS, low pur- the correspondence with the company investigated. Therefore, the
chasing cost and good U % , while AR is limited. An alternative set- trend of the KPIs as a function of the input data could be investi-
ting could be OP = 185 pallets and MPQ = 66 pallets. In this case, gated in future studies.
Company A would improve the asset rotation and would experi-
ence very limited out-of-stock situations; however, because of Appendix A. detailed results of correlation analysis
the low MPQ, this setting generates a relevant total cost of assets
management (approx. 421 €/day), since Company A will be forced The following tables (Tables A.1–A.3) report the results of the
to increase its retrieving operations. Scenario 3 reflects a situation, correlation analysis between the operating leverages of the deci-
which is once again hypothetical, where Company A is no longer sion process, i.e. OP and MPQ, and the KPIs measured. The
 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342 341

Table A.1
Correlation between OP and MPQ with strategic and economic KPIs under scenario 1 (N = 270).

Pearson’s correlation coefficient C tot CI C opp C p;r C p;u Cr AR OOS CT U% PA

⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄
OP .330 .906 .421 .709 .711 .061 .453 .714 .241 .903 .303⁄⁄
MPQ .735⁄⁄ .495⁄⁄ .680⁄⁄ .354⁄⁄ .353⁄⁄ .749⁄⁄ .799⁄⁄ .370⁄⁄ .984⁄⁄ .159⁄⁄ .972⁄⁄
⁄
Significant correlation at p < 0.05.
⁄⁄
Significant correlation at p < 0.01.

Table A.2
Correlation between OP and MPQ with strategic and economic KPIs under scenario 2 (N = 270).

Pearson’s correlation coefficient C tot CI C opp C p;r C p;u Cr AR OOS CT U% PA

⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄ ⁄⁄
OP .154 .876 .328 .581 .580 .078 .306 .580 .202 .872 .258⁄⁄
MPQ .762⁄⁄ .530⁄⁄ .706⁄⁄ .324⁄⁄ .323⁄⁄ .764⁄⁄ .841⁄⁄ .337⁄⁄ .987⁄⁄ .206⁄⁄ .978⁄⁄
⁄
Significant correlation at p < 0:05.
⁄⁄
Significant correlation at p < 0:01.

Table A.3
Correlation between OP and MPQ with strategic and economic KPIs under scenario 3 (N = 270).

Pearson’s correlation coefficient C tot CI C opp C p;r C p;u Cr AR OOS CT U% PA

OP .174⁄⁄ .930⁄⁄ .266⁄⁄ .604⁄⁄ .606⁄⁄ .017 .436⁄⁄ .621⁄⁄ .191⁄⁄ .906⁄⁄ .265⁄⁄
MPQ .755⁄⁄ .321⁄⁄ .796⁄⁄ .312⁄⁄ .306⁄⁄ .779⁄⁄ .763⁄⁄ .324⁄⁄ .970⁄⁄ .289⁄⁄ .949⁄⁄
⁄
Significant correlation at p < 0:05.
⁄⁄
Significant correlation at p < 0:01.

Pearson’s correlation coefficient (Stigler, 1989) is used to evaluate Guide, V. D.-R., Jr., & Van Wassenhove, L.-N. (2009). The evolution of closed-loop
supply chain research. Operations Research, 57(1), 10–18.
the relationship between the variables. The correlation analysis
Harris, F.W. (1913). How many parts to make at once. Factory, 10(2), 135–136
was supported by Statistical package for the social science (SPSS), <http://logist.ru/sites/default/files/users/user1/files/eoqmodel-originalpaper.
release 21 for Windows (IBM), and was carried out separately for pdf> Accessed December 2014.
each simulation scenario, exploiting the simulation outcomes Hellström, D., & Johansson, O. (2010). The impact of control strategies on the
management of returnable transport items. Transportation Research Part E:
(N = 270) as input data. Logistics and Transportation Review, 46(6), 1128–1139.
ISO/IEC (2007). ISO/IEC 15459-5:2007 – Information technology – Unique identification
– Part 5: Unique identification of returnable transport items (RTIs).
References Johansson, O., & Hellstrom, D. (2007). The effect of asset visibility on managing
returnable transport items. International Journal of Physical Distribution &
Logistics Management, 37(10), 799–815.
Aberdeen Group (2004). RFID-enabled logistics asset management benchmark report
Karkkainen, M., Ala-Risku, T., & Herold, M., (2004). Managing the rotation of
<http://www.tkrconsulting.com/tkr/Resources/whitepapers/RFID-Enabled%
reusable transport packaging—A multiple case study. In Proceedings of the
20Logistics%20Asset%20Mgmt.pdf> Accessed September 2014.
thirteenth international working seminar on production economics, Innsbruck,
Alinovi, A., Bottani, E., & Montanari, R. (2012). Reverse Logistics: A stochastic EOQ-
February 16–20, 2004.
based inventory control model for mixed manufacturing/remanufacturing
Kelle, P., & Silver, E. A. (1989). Forecasting the returns of reusable containers. Journal
systems with return policies. International Journal of Production Research,
of Operations Management, 8(1), 17–35.
50(5), 1243–1264.
Kim, T., Glock, C. H., & Kwon, Y. (2014). A closed-loop supply chain for deteriorating
Bottani, E., & Bertolini, M. (2009). Technical and economic aspect of RFID
products under stochastic container return times. Omega, 43, 30–40.
implementation for asset tracking. International Journal of RF Technologies:
Kroon, L., & Vrijens, G. (1995). Returnable containers: an example of reverse
Research and Applications, 1(3), 169–193.
logistics. International Journal of Physical Distribution & Logistics Management,
Bottani, E., Rizzi, A., & Vignali, G. (in press). Improving logistics efficiency of
25(2), 56–69.
industrial districts: A framework and exploratory study in the food sector.
Martınez-Sala, A. S., Egea-Lo  pez, E., Garcıa-Sa
nchez, F., & Garcıa-Haro, J. (2009).
International Journal of Logistics: Research and Applications. http://dx.doi.org/10.
Tracking of returnable packaging and transport units with active RFID in the
1080/13675567.2014.945401.
grocery supply chain. Computers in Industry, 60(3), 161–171.
Choong, S. T., Cole, M. H., & Kutanoglu, E. (2002). Empty container management for
McKerrow, D. (1996). What makes reusable packaging systems work. Logistics
intermodal transportation networks. Transportation Research Part E: Logistics
Information Management, 9(4), 39–42.
and Transportation Review, 38(6), 423–438.
Mollenkopf, D., Closs, D., Twede, D., Lee, S., & Burgess, G. (2005). Assessing the
Deb, K., Meyarivan, T., Pratap, A., & Agarwal, S. (2002). A fast and elitist
viability of reusable packaging: a relative cost approach. Journal of Business
multiobjective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary
Logistics, 26(1), 169–197.
Computation, 6(2), 182–197.
Dekker, R., Inderfurth, K., Van Wassenhove, L. N., & Fleischmann, M. (2004). Reverse Özkır, V., & Baslıgil, H. (2013). Multi-objective optimization of closed-loop
logistics – Quantitative models for closed-loop supply chains. Berlin: Springer. supply chains in uncertain environment. Journal of Cleaner Production, 41,
ISBN: 978354040696. 114–125.
Di Francesco, M., Crainic, T. G., & Zuddas, P. (2009). The effect of multi-scenario Rosenau, W. V., Twede, D., Mazzeo, M. A., & Singh, S. P. (1996). Returnable/reusable
policies on empty container repositioning. Transportation Research Part E: logistical packaging: A capital budgeting investment decision framework.
Logistics and Transportation Review, 45(5), 758–770. Journal of Business Logistics, 17(2), 139–165.
European Commission (2007). BRIDGE – Building radio frequency identification for the Stigler, S. M. (1989). Francis Galton’s account of the invention of correlation.
global environment – WP9: Returnable transport items: The market for EPCglobal Statistical Science, 4(2), 73–79. http://dx.doi.org/10.1214/ss/1177012580.
applications <www.bridge-project.eu> Accessed July 2008. Tanimizu, Y., & Shimizu, Y. (2014). A study on closed-loop supply chain model for
Flapper, S. D. P., van Nunen, J. A. E. E., & Van Wassenhove, L. N. (2005). Managing parts reuse with economic efficiency. Journal of Advanced Mechanical Design,
closed-loop supply chains. Berlin: Springer. Systems, and Manufacturing, 8(5), 1–12. http://dx.doi.org/10.1299/jamdsm.
Georgiadis, P., & Athanasiou, E. (2013). Flexible long-term capacity planning in 2014jamdsm0068.
closed-loop supply chains with remanufacturing. European Journal of Taylor, R. (1990). Interpretation of the correlation coefficient: A basic review.
Operational Research, 225(1), 44–58. Journal of Diagnostic Medical Sonography, 6(1), 35–39.
Guide, V. D. R., Jayaraman, V., & Linton, J. D. (2003). Building contingency planning Thoroe, L., Melski, A., & Schumann, M. (2009). The impact of RFID on management of
for closed-loop supply chains with product recovery. Journal of Operations returnable containers. Electron Markets, 19, 115–124. http://dx.doi.org/10.1007/
Management, 21(3), 259–279. s12525-009-0013-3.
342 E. Bottani et al. / Computers & Industrial Engineering 87 (2015) 328–342

Tibben-Lembke, R. (2004). Strategic use of the secondary market for retail consumer Witt, C. E. (2000). Are reusable containers worth the cost? Transportation &
goods. California Management Review, 46(2), 90–104. Distribution, 41(9), 105–108.
Twede, D. (1999). Can you justify returnables? Transportation & Distribution, 40(4), Zhang, S., Zhao, X., & Zhang, J. (2014). Dynamic model and fuzzy robust control of
85–88. uncertain closed-loop supply chain with time-varying delay in
Wang, H.-F., & Hsu, H.-W. (2010). A closed-loop logistic model with a spanning-tree remanufacturing. Industrial and Engineering Chemistry Research, 53(23),
based genetic algorithm. Computers & Operations Research, 37(2), 376–389. 9805–9811. http://dx.doi.org/10.1021/ie404104c.