Simulation in the supply chain context a survey Sergio Terzia,.docx
•Download as DOCX, PDF•
0 likes•14 views
Simulation in the supply chain context: a survey
Sergio Terzia,*, Sergio Cavalierib a Politecnico di Milano, Department of Economics, Industrial and Management Engineering, Piazza Leonardo da Vinci 32, 20133 Milan, Italy b Department of Industrial Engineering, Universita` di Bergamo, Viale Marconi 5, 24044 Dalmine, Italy Received 29 January 2003; accepted 13 June 2003
Abstract
The increased level of competitiveness in all industrial sectors, exacerbated in the last years by the globalisation of the economies and by the sharp fall of the final demands, are pushing enterprises to strive for a further optimisation of their organisational processes, and in particular to pursue new forms of collaboration and partnership with their direct logistics counterparts. As a result, at a company level there is a progressive shift towards an external perspective with the design and implementation of new management strategies, which are generally named with the term of supply chain management (SCM). However, despite the flourish of several IT solutions in this context, there are still evident hurdles to overcome, mainly due to the major complexity of the problems to be tackled in a logistics network and to the conflicts resulting from local objectives versus network strategies. Among the techniques supporting a multi-decisional context, as a supply chain (SC) is, simulation can undoubtedly play an important role, above all for its main property to provide what-if analysis and to evaluate quantitatively benefits and issues deriving from operating in a co-operative environment rather than playing a pure transaction role with the upstream/downstream tiers. The paper provides a comprehensive review made on more than 80 articles, with the main purpose of ascertaining which general objectives simulation is generally called to solve, which paradigms and simulation tools are more suitable, and deriving useful prescriptions both for practitioners and researchers on its applicability in decision-making processes within the supply chain context. # 2003 Elsevier B.V. All rights reserved. Keywords: Parallel and distributed simulation; Supply chain management; High level architecture; Survey 1. Introduction Modern industrial enterprises operate in a rapidly changing world, stressed by even more global competition, managing world-wide procurement and unforeseeable markets, supervising geographically distributed production plants, striving for the provision of outstanding products and high quality customer service. More than in the past, companies which are not able to revise periodically their strategies and, accordingly, to modify their organisational processes seriously risk to be pulled out from the competitive edge. In the 1990s, companies have made huge efforts for streamlining their internal business processes, identifying and enhancing the core activities pertaining to the product value chain, and invested massively in new intra-company information and communicat.
1 of 52
More Related Content
Simulation in the supply chain context a survey Sergio Terzia,.docx
1. Simulation in the supply chain context: a survey
Sergio Terzia,*, Sergio Cavalierib a Politecnico di Milano,
Department of Economics, Industrial and Management
Engineering, Piazza Leonardo da Vinci 32, 20133 Milan, Italy b
Department of Industrial Engineering, Universita` di Bergamo,
Viale Marconi 5, 24044 Dalmine, Italy Received 29 January
2003; accepted 13 June 2003
Abstract
The increased level of competitiveness in all industrial sectors,
exacerbated in the last years by the globalisation of the
economies and by the sharp fall of the final demands, are
pushing enterprises to strive for a further optimisation of their
organisational processes, and in particular to pursue new forms
of collaboration and partnership with their direct logistics
counterparts. As a result, at a company level there is a
progressive shift towards an external perspective with the
design and implementation of new management strategies,
which are generally named with the term of supply chain
management (SCM). However, despite the flourish of several IT
solutions in this context, there are still evident hurdles to
overcome, mainly due to the major complexity of the problems
to be tackled in a logistics network and to the conflicts resulting
from local objectives versus network strategies. Among the
techniques supporting a multi-decisional context, as a supply
chain (SC) is, simulation can undoubtedly play an important
role, above all for its main property to provide what-if analysis
and to evaluate quantitatively benefits and issues deriving from
operating in a co-operative environment rather than playing a
pure transaction role with the upstream/downstream tiers. The
paper provides a comprehensive review made on more than 80
articles, with the main purpose of ascertaining which general
objectives simulation is generally called to solve, which
paradigms and simulation tools are more suitable, and deriving
useful prescriptions both for practitioners and researchers on its
2. applicability in decision-making processes within the supply
chain context. # 2003 Elsevier B.V. All rights reserved.
Keywords: Parallel and distributed simulation; Supply chain
management; High level architecture; Survey 1. Introduction
Modern industrial enterprises operate in a rapidly changing
world, stressed by even more global competition, managing
world-wide procurement and unforeseeable markets, supervising
geographically distributed production plants, striving for the
provision of outstanding products and high quality customer
service. More than in the past, companies which are not able to
revise periodically their strategies and, accordingly, to modify
their organisational processes seriously risk to be pulled out
from the competitive edge. In the 1990s, companies have made
huge efforts for streamlining their internal business processes,
identifying and enhancing the core activities pertaining to the
product value chain, and invested massively in new intra-
company information and communication platforms, as data
warehouse or ERP systems. In the last years, globally active
companies, as well as SMEs, are realising that the efficiency of
their own Computers in Industry 53 (2004) 3–16
*Corresponding author. Fax: þ39-02-2399-2700. E-mail
address: [email protected] (S. Terzi). 0166-3615/$ – see front
matter # 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0166-3615(03)00104-0 businesses is heavily
dependent on the collaboration and co-ordination with their
suppliers as well as with their customers [1]. This external
perspective is termed in literature under the broad concept of
supply chain management (SCM), which is concerned with the
strategic approach of dealing with trans-corporate logistics
planning and operation on an integrated basis [2]. Adopting a
SCM strategy means to apply a business philosophy where more
industrial nodes along a logistic network act together in a
collaborative environment, pursuing common objectives,
exchanging continuously information, but preserving at the
same time the organisational autonomy of each single unit. This
business vision is applied to different industrial processes (e.g.
3. procurement, logistics, marketing, etc.) and implementing
different policies (e.g. continuous replenishment, co-marketing,
etc.). Integrated management frameworks (as the SCOR [3]
model) support the development of collaboration among
multiple tiers through mutually designed planning and execution
processes along the entire supply chain (SC). From the IT
perspective, a new wave of solutions is arising with the main
hype to overcome all the physical, organisational and
informational hurdles which can seriously jeopardise any co-
operation effort. Advanced planning and scheduling (APS)
systems aim to step over the intra-company integration supplied
by ERP systems by providing a common inter-organisational
SCM platform, which supports the logistics chain along the
whole product life-cycle, from its initial forecast data, to its
planning and scheduling, and finally to its transportation and
distribution to the end customer [4]. Despite the various
solutions currently available on the market, the common
features of the APS products reside on the intensive usage of
quantitative methods in order to provide users with the best
solution at time. An example is given by mixed integer linear
programming techniques and genetic algorithms for solving
multi-site or transportation planning problems, or timeseries
and regressive techniques for demand planning problems.
Among these quantitative methods, simulation is undoubtedly
one of the most powerful techniques to apply, as a decision
support system, within a supply chain environment. In the
industrial area, simulation has been mainly used for decades as
an important support for production engineers in validating new
lay-out choices and correct sizing of a production plant (e.g.
[5,6]). Nowadays, simulation knowledge is considered one of
the most important competences to acquire and develop within
modern enterprises in different processes (business, marketing,
manufacturing, etc.) [7]. Within the Visions for 2k-enterprises
[8], simulation is considered one of the most relevant key-
success factors for companies surviving, thanks to its
predictable features. Several organisations consider simulation
4. as an essential decision support system, for example, since
1996, the USA Department of Defence (DoD) has been asking to
all its services and parts suppliers to furnish a simulation model
of the product/service provided [9]. In particular, as the topic of
the paper, supply chain is a typical environment where
simulation (in particular, discrete-event simulation) can be
considered a useful device. In fact, it is quite evident to find out
how, by using simulation technology, it is possible to reproduce
and to test different decision-making alternatives upon more
possible foreseeable scenarios, in order to ascertain in advance
the level of optimality and robustness of a given strategy. Aim
of the paper is to survey how simulation techniques (in
particular, discrete-event simulation) could represent one of the
main IT enablers in a supply chain context for creating a
collaborative environment among logistics tiers. After an
introduction to simulation specifications and terminology
(Section 2), a detailed literature review is proposed (Section 3)
in order to analyse the scope of use, the paradigms employed
and the main benefits reported from the adoption of simulation
techniques in the supply chain context. In Section 4, final
considerations from the authors are provided. 2. The role of
simulation techniques in the supply chain context Despite the
great emphasis given in the last decade on the need for
companies to smooth their physical boundaries in favour of a
more integrated perspective, there is often among practitioners a
lot of confusion and a flawed use of the term ‘‘integration’’.
Stevens [10] provides a framework for achieving an integrated
supply chain, highlighting that integration of logistics functions
requires a progressive evolution from intra-company functional
integration (i.e. change 4 S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3–16 from a functional to a process view of
internal activities) to an internal corporate logistics integration
(supported by ERP, DRP systems), and finally to an external
integration in a logistic network extended upstream to suppliers
and downstream to customers. The last step is undoubtedly the
most challenging one. However, in addition to the classical
5. morphological scheme in corporate logistics, a logistics network
requires, among others, alignment of network strategies and
interests, mutual trust and openness among tiers, high intensity
of information sharing, collaborative planning decisions and
shared IT tools [1]. These requirements represent often the
major hurdles inhibiting the full integrability of a logistics
chain: even in presence of a strong partnership and mutual trust
among logistics nodes, there are in practice evident risks of
potential conflict areas of local versus global interests and
strong reluctance of sharing common information related to
production planning and scheduling as for example inventory
and capacity levels. Hence, from the IT point of view there is
the strong requirement to adopt distributed collaborative
solutions, which could preserve at the same time the local
autonomies and privacy of logistics data. Moreover, these
solutions must necessarily be platform independent and easily
interfaceable with companies’ legacy systems. These
requirements are profoundly changing also the traditional
paradigms underlying the world of simulation. In literature,
there is a progressive shift of research and application works
from local, single node simulation studies to modelling of more
complex systems, as logistics channels are. Generally,
simulation of such systems can be carried out according to two
structural paradigms: using only one simulation model, executed
over a single computer (local simulation), or implementing
more models, executed over more calculation processors
(computers and/or multi-processors) in a parallel or distributed
fashion [11]. Consequently, a simulation model of a supply
chain can be designed and realised either traditionally as a
whole single model reproducing all nodes (Fig. 1), or using
more integrated models (one for each node), which are able to
run in parallel mode in a single cooperating simulation (Fig. 2).
Fig. 1. Local simulation paradigm. Model Model Model Model
Co-operative Simulation Fig. 2. Parallel and distributed
simulation paradigm. S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3–16 5 The next section will be mainly
6. addressed to the specification of the parallel and distributed
simulation (PDS) paradigms. 2.1. The parallel and distributed
simulation paradigms Parallel discrete-event simulation (PS) is
concerned with the execution of simulation programs on
multiprocessor computing platforms, while distributed
simulation (DS) is concerned with execution of simulations on
geographically distributed computers interconnected via a
network, local or wide [11]. Both cases imply the execution of a
single main simulation model, made up by several sub-
simulation models, which are executed, in a distributed manner,
over multiple computing stations. Hence, it is possible to use a
single expression, PDS, referred to both situations. PDS
paradigm is based upon a co-operation and collaboration
concept in which each model co-participates to a single
simulation execution, as a single decision-maker of a
‘‘federated’’ environment. The need of a distributed execution
of a simulation across multiple computers derives from four
main reasons [9,11,12]. To reduce execution simulation time: A
large simulation can be split in more models and so executed in
a shorter time. To reproduce a system geographic distribution:
Some systems (as supply chain systems or military applications)
are geographically distributed. Therefore, reducing them into a
single simulation model is a rough approximation. By
preserving the geographic distribution, the execution of a PDS
over distributed computers enables the creation of virtual
worlds with multiple participants that are physically located at
different sites. To integrate different simulation models that
already exist and to integrate different simulation tools and
languages: Simulation models of single local sub-systems may
already exist before designing a PDS (e.g. flight simulators in
military application, but also local production systems in a
supply chain context) and may be written in different simulation
languages and executed over different platforms. By using a
PDS paradigm, it is possible to integrate existing models and
different simulation tools into a single environment, without the
need to adopt a common platform and language and to re-write
7. the models. To increase tolerance to simulation failures: This is
a potential benefit for particular simulation systems. Within a
PDS, composed by different simulation processors, if one
processor fails, it may be possible for others processors to go on
with simulation runs without the down processor. PDS paradigm
derives from studies that academic laboratories and also
military agencies have been realising since 1970. These studies
can be classified according to Fujimoto [11] in two major
categories. Analytic simulation: This type of simulation is used
to analyse quantitatively the behaviour of systems. In this case,
PDS paradigm is applied to execute as fast as possible the
simulation experimental campaigns. Distributed virtual
environment: A virtual environment is composed by more
simulation applications that are used to create a virtual world
where humans can be embedded for training (e.g. soldiers
training in battlefields) and also for entertainment (e.g.
distributed video games) purposes. In recent years, PDS
paradigm has been mainly used in military applications, but also
in several civil domains (e.g. navy in [13], emergency
management in [14], transportation in [15]). PDS practical
execution needs a framework, which enables to model the
information sharing and synchronicity among single local
simulations. In literature, it is possible to distinguish two
different PDS frameworks, separated by their basic co-
ordination logic. A network structure, based on a distributed
protocol logic, in which single nodes are mutually
interconnected (Fig. 3a). A centralised structure, founded on a
centric logic, in which a single process manager is responsible
for linking participant nodes (Fig. 3b). For the purposes of the
paper, it is possible to synthesise the two frameworks as
follows. Distributed protocols map interaction messages that
each participant model sends continuously to other nodes, to
bring their update of proper simulation state. MPI-ASP [16] and
GRIDS [17] are examples of distributed protocols logic. 6 S.
Terzi, S. Cavalieri / Computers in Industry 53 (2004) 3–16 The
centric logic provides a software instrument that is able to
8. receive standard messages from each connected node, and,
therefore, to sort out needed communications between single
participant simulation nodes. The last logic, as it will be
possible to understand by the following literature survey, is
becoming the most widely used, since it clearly divides
connection and model activity. In fact, in a PDS centric
structure a user is only interested in the model creation, while
the central software solves all connection problems. High level
architecture (HLA) [12] is the most known PDS framework.
HLA is a standard PDS architecture developed by the US DoD
for military purposes and nowadays is becoming an IEEE
standard. A PDS in HLA is named ‘‘federation’’ while
participant models are termed ‘‘federates’’. One HLAPDS is
based on the ‘‘federation and federate rules’’, which establish
10 ground rules for creating and managing the simulation. In
particular, 10 ‘‘rules’’ identify: the HLA interface
specification, that defines services for federation execution; the
Object Modelling Template (OMT) language, for the
specification of communications amongst federates. Within the
HLA framework, a distributed simulation is accomplished
through a ‘‘federation’’ of concurrent ‘‘federates’’, interacting
between themselves by means of a shared data model and
federation services (basically time and data distribution
management services). The federation services are provided by
the Run Time Infrastructure (RTI) software tool, compliant to
the HLA interface specification. 2.2. PDS and supply chain
simulation Many software vendors (e.g. i2 in [18], or IBM in
[19]), universities and consultancy companies have traditionally
used a local simulation approach in the supply chain context.
Only in recent years, some of the features of PDS were
recognised as important benefits for enabling sound simulation
models in support of SCM policies [20,21]. PDS ensures the
possibilities to realise complex simulation models which cross
the enterprise boundaries without any need of common sharing
of local production system models and data; as previously
discussed, companies that do not belong to the same enterprise
9. might not be willing to share their data openly. Gan et al. [22]
explain that PDS paradigm guarantees the ‘‘encapsulation’’ of
different local models within one overall complex simulation
system, so that, apart from the information exchanged, each
model is self-contained. PDS provides a connection between
supply chain nodes that are geographically distributed
throughout the globe, guaranteeing that each single simulation
model is really linked to its respective industrial site. In some
cases, the execution of a PDS model allows to reduce the time
spent for simulation, since separated models run faster than a
single complex model. 3. Literature survey The survey has been
conducted over the scientific literature in order to ascertain
which general objectives simulation is generally called to solve,
using which paradigms and simulation tools or languages, and
derive useful prescriptions both for practitioners and
researchers on its applicability in decision-making processes
within the supply chain context. More than 80 papers have been
reviewed. Introductive papers on supply chain simulation were
also analysed, but they are not classified within the tables.
Reader may note that the survey considers only papers and
references that propose applications of supply chain simulation,
as (i) industrial test cases, or (ii) simulation software
specifically designed for modelling supply chains or (iii)
simulation tests conducted over a logistics network. Fig. 3. PDS
frameworks. S. Terzi, S. Cavalieri / Computers in Industry 53
(2004) 3–16 7 Table 1 Literature survey—local simulation
paradigm Papers Alfieri and Brandimarte [32] Archibald etal.
[28] Bagchi etal. [19] Belhau etal. [24] Berry and Naim [50]
Botter etal. [25] Burnett and Le Baron [51] Cavalieri etal. [52]
Chen etal. [40] Hafeez etal. [54] Hirsch etal. [23] Ingalls etal.
[55] Jain etal. [56] Luo etal. [57] Mielke [58] Persson and
Olhager [59] Petrovic [60] Phelps etal. [42] Phelps etal. [61]
Promodel [27] Ritchie Dunham and Anderson [37] Siprelle etal.
[29] Schunk [26] Van der Vorst etal. [31] Zhang etal. [30]
Zhang etal. [38] Scope and objective Objective Network design
Design Localisation Strategic decision Management archetype
10. Strategic model Process Demand and sales planning SC
planning Inventory planning Distribution and transportation
planning Production planning and scheduling Morphology SC
ownership SC single ownership SC multi-ownership SC levelsa
Na 2 Na 2 2þ Na Na 2 Na 2þ Na Na Na 2þ Na Na Na Na Na Na
Na Na Na Na 2þ Na Simulation paradigm and technology Local
Specific tool General tool Other (simulation tools and
languages) ModSim IBM SCA IBM SCA Create! Dynamo Arena
Automod Java IBM SCA Dynamo LOCOMOTIVE Arena Arena
Arena Arena Taylor II General purpose SDI SDI SCGuru SDI
Supply Solver General purpose Arena Development stageb Ex,
Cn Ex Sw Ex, Cn Ex Ex Ex Ex Cn Ex Ex Ex Ex, Cn Ex Ex Ex
Ex, Cn Sw Sw Sw Cn Sw Sw Ex, Cn Ex Cn a Na means
information not available. b Cn: conceptual; Sw: software; Ex:
experience; Ts: testing. 8 S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3 –16 Table 2 Literature survey—parallel
and distributed simulation paradigm Papers Barnett and Miller
[39] Brun et al. [35] Gan et al. [16] Gan et al. [22,33,43] Gan et
al. [21,44] Gan and McGinnis [53] Kim et al. [36] Seliger et al.
[48] Strasburger et al. [34,45,46] Sudra et al. [17]
Ventateswaran et al. [62] Zulch et al. [63] Scope and objective
Objective Network design Design Localisation Strategic
decision Management archetype Strategic model Process
Demand and sales planning SC planning Inventory planning
Distribution and transportation planning Production planning
and scheduling SC features SC ownership SC single ownership
SC multi-ownership SC levelsa Na 2 2 2 2 Na Na 2 2 Na 2 Na
Simulation paradigm and technology PDS Network logic Centric
logic () Other (simulation tools and languages, PDS
frameworks) HLA HLA (WILD)MPI-HLA DP HLA HLA
DEVS/CORBA HLA HLA GRIDS HLA Osim Development
stageb Cn Ts, Ex Ts Ts Ts Ts, Cn Cn Ts Ts Ts Ts Cn a Na
means information not available. b Cn: conceptual; Sw:
software; Ex: experience; Ts: testing. S. Terzi, S. Cavalieri /
Computers in Industry 53 (2004) 3 –16 9 The survey makes use
of a chart classification and its results are summarised in Tables
11. 1 and 2. Before detailing the content of the tables, it is
necessary to introduce the classification criteria adopted. 3.1.
Classification criteria Three classification criteria have been
adopted for categorising the reviewed articles. Scope and
objectives: It is related to the specific context, the objectives
and the scale of the problem (strategic, tactic, operative) the
simulation technique was addressed to. Simulation paradigm
and technology: It states the simulation paradigm (e.g. local
versus distributed simulation) and the simulation tools and
languages adopted. Development stage: It refers to the different
levels of development of the simulation application reported in
the articles (from the conceptual level to testing activities or
commercial applications). 3.1.1. Scope and objectives This
classification driver is further structured in three sub-criteria:
(1) objectives, (2) processes, (3) morphology. (1) Objectives: It
is possible to highlight two macro objectives. (a) Network SC
design: Simulation can be used as a decision support system
within the design phases (e.g. design of a logistics network,
design of production nodes). Two sub-levels are defined. (i)
Design: It stands for logical modelling and industrial nodes
configuration. It is possible to notice that all papers illustrating
specific simulation tools stress this objective. For example, in
Hirsch et al. [23], a specific supply chain simulation tool,
named LOCOMOTIVE, is adopted to verify and test more
solutions into a logistic network for packing eco-reusing and
recovery. (ii) Node localisation: It relates to the activity of
placing a supply chain node in a determined geographic site.
Only a few simulation models and tools, among those reviewed,
deal with the problem of geographic disposition of industrial
nodes. For example, in Belhau et al. [24], a simulation model is
conceived in order to identify the right geographic disposition
for distribution centres, aiming to minimise transport costs
through the use of proper cost functions. (b) SC strategic
decision support: Simulation is applied over a supply chain to
evaluate more strategic alternatives, as strategies based on
quick response, collaborative planning and forecasting or
12. outsourcing to third-parties. As an example, in Botter et al. [25]
simulation is applied on a Brazilian beer logistics network in
order to evaluate the possibility of entirely outsourcing the
logistic process to an external provider. (2) Processes: The
survey investigates which processes are addressed and which
decision levels (strategic, tactic, operative [7]) are pondered in
the simulation applications under scrutiny. The classification
makes use of the same categorisation of most APS systems [4].
(a) Demand and sales planning: Simulation processes dealing
with stochastic demand generation (e.g. customer process
generation) and forecasting planning definition. (b) Supply
chain planning: Simulation processes supporting production
planning and distribution resources allocation, under supply and
capacity constraints; as an example, Schunk [26] describes a
simulation tool, Supply Solver, which is interfaced with an
external module, which optimises the solution for distribution
and production allocation problem. (c) Inventory planning:
Simulation processes supporting multi-inventory planning; the
commercial simulation tool programmed by Promodel, SCGuru,
proposes, a specific module for inventory management and
optimisation [27]. (d) Distribution and transportation planning:
Simulation of distribution centres, sites localisation and
transport planning, in terms of resources, times and costs; it is
one of the most recurrent simulation processes reported 10 S.
Terzi, S. Cavalieri / Computers in Industry 53 (2004) 3–16 in
literature; for example, as described in Bagchi et al. [19] and
Archibald et al. [28], IBM supply chain analyser has two
separated modules (distribution and transportation planning) to
simulate distribution centres, transport type (train, truck, etc.)
and relative management processes (material handling, loading
and unloading, etc.). (e) Production planning and scheduling:
Simulation processes related to production management. Each
logistics node is simulated at its manufacturing layer, as a
specific set of machines, cells and lines. Manufacturing
planning is implemented by simulation models (and tools),
which integrate different model layers, from single production
13. lines to the entire factory and to the whole logistics chain. SDI
Industry Pro [29] is one of the most important examples of
manufacturing planning implementation; SDI is a simulation
tool specifically developed for logistics chains, which allows
the development of models from single production machines to
more complex distribution centres. (3) Morphology: The
morphology of the supply chains addressed by simulation
models can be further refined as follows. (a) Supply chain
ownership: Which distinguishes two possible conditions. (i)
Single ownership: This is the typical case of multinational
companies, whose industrial nodes (manufacturers, distributors,
financial sites, etc.) are distributed all over the world; an
example is the case of IBM and its supply chain analyser [19]
simulation tool specifically developed as a decision support
system for solving company’s supply chain issues. (ii) Multi-
ownership: In this case, there is a fair balance of power among
more autonomous enterprises joining a logistics network. In the
LOGSME-ESPRIT 22633 European project, a simulation tool
has been developed in order to support the decision-making
process of a logistics network made up by SMEs [30]. (b)
Supply chain levels: With regards to the number of tiers along a
supply chain, from the survey, it came out that most of the
articles reviewed do not provide clear information about the
physical dimension of the simulated systems. 3.1.2. Simulation
paradigm and technology As reported in Section 2.1, there are
in literature two main alternative approaches adopted, with
different choices in terms of tools and languages adopted. Local
simulation paradigm: With: (i) specific commercial simulation
tools developed by software vendors only for simulation
purposes within a supply chain context (e.g. SDI Industry Pro in
[29], IBM SCA in [19], SCGuru in [27], LOCOMOTIVE in
[23], Supply Solver in [26]). (ii) General-purpose simulation
tools or languages, as Arena [25], Create! [24], CPLEX [31],
ModSim [32]. Parallel and distributed simulation paradigm:
With (i) a network logic approach (e.g. CMB-DIST in [22],
MPI-ASP in [33], GRIDS in [17]); (ii) a centric logic approach
14. (e.g. HLA in [34,35], DEVS/CORBA in [36]). 3.1.3.
Development stage From the literature review, it is possible to
argue that the reported simulation studies are at a different level
of development, ranging from as follows. Conceptual level:
Papers which still denote a conceptual content, since simulation
models appear not yet implemented and tested [36–38], or are
mainly proposals for new descriptive methodologies supporting
the adoption of simulation in supply chain environments
[24,31,39,40], or reporting the application of novel simulation
paradigms, as web-based simulation [41]. Software description:
Papers which explain features of tools specifically created for
design and development of simulation models. Examples of this
category of articles are the two papers presented at the 1998 and
2000 Winter Simulation Conference (WSC) by two simulation
software development teams, IBM supply chain analyser [19]
and SDI Industry Pro [42]. Experience description: Papers
which describe real applications of supply chain simulation. For
example, Archibald et al. [28] describe a simulation of a food
logistics network aiming to verify the S. Terzi, S. Cavalieri /
Computers in Industry 53 (2004) 3–16 11 effectiveness of
alternative logistics management strategies, in particular, the
adoption of continuous replenishment policies. Testing activity:
Papers which verify simulation technology portability in a
supply chain context. In particular, in these papers IT platforms
and software solutions are tested. The stability of distributed
simulation paradigms is the most experimented problem, as it
appears by the papers presented by the research groups of the
University of Singapore [16,21,22,33,43,44] and by the
University of Magdeburg [34,45,46]. 4. Survey analysis From
the literature survey, it is possible to draw some useful
indications for recognising the future trends of simulation
applications in a supply chain context. At first, it is important
to notice the clear difference that exists between local
simulation and PDS paradigms. In fact, after this evidence,
authors decided to divide Tables 1 and 2 in local simulation and
PDS experiences. Next considerations are reported having in
15. mind this first sharp separation. 4.1. Local simulation paradigm
The local simulation paradigm is still the most applied approach
in literature. It is mainly applied for supply chain network
design, but also for verifying strategic models and management
archetypes. The most implemented simulation processes are
related to distribution, transportation and inventory planning.
With regards to the simulation tools adopted, with more
powerful simulation tools (e.g. IBM SCA and SDI), based upon
modular construction, it is possible to describe detailed industry
models and more complex supply chain processes; on the other
hand, general-purpose simulation languages guarantee more
programs flexibility, but with more complexity, so that they
appear not suitable for simulation of multi-tier logistics
networks. In synthesis, the local simulation paradigm: is used in
many experiences, with heterogeneous objectives, from supply
chain design to strategic decisions, within several industrial
sectors and with different company scales; is often realised,
within the industry environment, with specific simulation tools,
whilst academic users mostly apply general simulation tools; is
usually applied to a single-ownership supply chain (e.g. as in
the IBM case), while only for some experiences is applied to a
multi-ownership supply chain, for the main reason that each
company normally is not willing to share its own simulation
models and data with the other tiers of the network. 4.2. PDS
paradigms The literature survey on PDS applications points out
clearly that PDS paradigm has not become a steady applied
approach and probably, at this time, the critical research mass
for advancing development and userfriendly employment has
not been yet reached. Certainly, this is due to the major IT
complexity that PDS paradigm causes. Among the studies
reporting the use of PDS paradigm, it is worthwhile to report
two particular experiences. The Web Integrated Logistics
Designer (WILD) project [47], conducted by the authors, which
makes use of heterogeneous simulation models, each
reproducing an industrial node of an aeronautical multi-
ownership logistics chain, written in different languages and
16. intertwined through the use of the HLA framework; the main
objective of the project was to integrate the local production
planning and scheduling activities at each node by means of
interaction among distributed simulation models; in each
simulation model, local production systems, production
management and scheduling activities were simulated. The
Osim project [48], conducted by the University of Karlsruhe,
which aims to create a hierarchical simulation where more
interconnected simulation models reproduce different
‘‘industrial’’ processes and layers (production physical cells,
production management, customers, business control, etc.) in
order to model a single supply chain node. These coupled
models could be (not at the present version) interconnected in a
more extensive supply 12 S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3–16 chain simulation with models of other
industrial nodes. Both experiences highlight the increasing
attention of the scientific and industrial community for parallel
and distributed supply chain simulation, which is being
developed in different ways: in the research world, it is in a
testing phase, above all for solving IT stability problems; there
is not yet a sufficient critical research mass for expanding PDS
application; it is applied mainly to multi-ownership supply
chains, for their main property to solve any information-sharing
issue among nodes, thanks to the provision of a common
information bus where each simulation model, even if written
with proprietary language, can be plugged in and synchronised;
at IT implementation level, it is possible to observe an
evolutionary trend from a network structure, based on
distributed protocols approach, to a centric structure, especially
based on the HLA standard framework. 5. Conclusions
According to Chang and Makatsoris [49]: ‘‘discrete-event
simulation allows the evaluation of operating performance prior
to the implementation of a system since: (a) it enables
companies to perform powerful what-if analyses leading them to
better planning decisions; (b) it permits the comparison of
various operational alternatives without interrupting the real
17. system and (c) it permits time compression so that timely policy
decisions can be made’’. These features are the common
background coming out from the survey reported in this paper,
which shows how simulation is successfully adopted in different
studies related to logistics network. In particular, the local
simulation paradigm is preferably used within intra-company
supply chain projects (typical of large multinational logistics
networks) for evaluating and quantitatively ranking different
project solutions or for verifying more strategic policies. On the
contrary, if the supply chain is composed by independent
enterprises, sharing information becomes a critical obstacle,
since each independent actor typically is not willing to share
with the other nodes its own production data (as production
capacity, internal lead times, production costs, etc.). This
problem is further exacerbated in geographically distributed
networks. Each simulation model of a local production site of a
company needs be locally resident on each plant. In fact, the
maintenance of the simulation model cannot be carried out
centrally, since only the technical personnel directly working on
the plant is able to maintain and update it whenever the plant is
subjected to any reconfiguration (like installing new machines
or lay-out modifications). Unlike local simulation, PDS
paradigm fulfils powerfully these requirements. Within the PDS
approach, each simulation model can run in its own local
environment; the data exchange and, above all, the
synchronisation with the other distributed simulation models are
ensured by a shared protocol. Thus, in a supply chain context,
collaborating nodes need only to define at the beginning which
information will be shared and the time steps or the production
events which will trigger the data exchange. In addition, each
model can be developed with different simulation tools or
languages and executed on heterogeneous platforms, since the
establishment of the shared network is rather similar to a plug-
in tool. This sounds quite important whenever simulation
models already exist: no substantial revisions on the simulation
code need to be produced in order to scale it up from a local
18. running to a distributed experimentation. PDS can be
implemented with the two frameworks described in Section 2.1,
which propose different solutions for the two most important
PDS problems: (i) data exchanging and (ii) simulation time
synchronisation. From literature survey, it is possible to argue
that the centric logic is becoming the most used framework. In
particular, HLA can be considered the reference, adopted in
several simulation projects within different domains (military,
civil, scientific). This HLA supremacy derives certainly from
the free distribution policy decided by the USA DoD (developer
of the HLA framework), but it also comes out from evidence:
HLA promises a relative simple approach to PDS and it
guarantees all necessary devices and support. Once available the
proper IT tools, it is possible to assert that in the future,
simulation models developed with the PDS approach could
better enlarge their S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3–16 13 current scope of application as a
support to the decision-making processes of SCM. Their
intensive use will certainly contribute to the elimination of the
current barriers in the accomplishment of a real integration of
logistics networks. By providing a systematic quantitative and
objective evaluation of the outcomes resulting from different
possible planning scenarios, from demand planning to
transportation and distribution planning, simulation techniques
can make companies more aware of the benefits coming out
from an integrated and co-operating strategy with their
upstream/downstream nodes rather than following myopically
an antagonistic behaviour with them. Acknowledgements The
paper reports some of the results achieved by the authors within
the WILD project (refer to its website
http://st.itim.unige.it/wild/ for detailed information), a project
involving seven Italian universities and funded by the Italian
Ministry of Universities Research & Scientific Technologies.
The authors wish to thank all the collaborating researchers
within the WILD project, in particular, researchers working at
the Dipartimento di Ingegneria Gestionale of Politecnico di
19. Milano, namely Prof. Marco Garetti, Prof. Alessandro Pozzetti,
Dr. Alessandro Brun, Dr. Maria Caridi, Roberto Cigolini and
Marco Macchi. References [1] R. Hieber, Supply chain
management: a collaborative performance measurement
approach, VDF Zurich, 2002. [2] H.C.W. Lau, W.B. Lee, On a
responsive supply chain information system, International
Journal of Physical Distribution and Logistics Management 30
(7–8) (2000) 598– 610. [3] SCOR, Supply Chain Operations
Reference Model, Supply Chain Council, 2002 (www.supply-
chain.org). [4] H. Stadtler, C. Kilger, Supply Chain
Management and Advanced Planning, Springer-Verlag, Berlin,
2000. [5] B. Martinoli, Guida alla simulazione, Franco Angeli
Editore, 1993. [6] R. Brooks, in: S. Robinson (Ed.), Simulation,
Operational Research Series, Palgrave, 2001. [7] J. Kosturiak,
M. Gregor, Simulation in production system life cycle,
Computers in Industry 38 (1999) 159–172. [8] Integrated
Manufacturing Technology Initiative (IMTI), Modelling and
Simulation Roadmap, 2000 (www.imti.net). [9] Defense
Modeling and Simulation Office (DMSO), High level
architecture (HLA) overview, 2002 (hla.dmso.mil). [10] G.
Stevens, Integrating the supply chain, Physical Distribution &
Materials Management 19 (8) (1989) 3–8. [11] R. Fujimoto,
Parallel and distributed simulation, in: Proceedings of the 1999
Winter Simulation Conference, IEEE, Piscataway, NJ, pp. 122–
131. [12] Defense Modeling and Simulation Office (DMSO),
RTI Programmer’s Guide, Version NG, 1999 (hla.dmso.mil).
[13] G.A. Bruzzone, R. Mosca, R. Revetria, YACHTS—yet
another cooperative high level architecture training software,
in: Proceedings of the 2001 Winter Simulation Conference,
2001. [14] U. Klein, Distributed simulation for emergency
management based on the high level architecture, in:
Proceedings of the 5th Conference of the International
Emergency Management Society (TIEMS’98), 1998. [15] T.
Schulze, U. Klein, Migration of HLA into civil domains:
solutions and prototypes for transportation applications,
Simulation 73 (5) (1999) 296–303. [16] B.P. Gan, L. Liu, J.S.
20. Turner, W. Cai, S. Jain, W.J. Hsu, Distributed supply chain
simulation across enterprise boundaries, in: Proceedings of the
2000 Winter Simulation Conference, 2000. [17] R. Sudra, S.J.
Taylor, T. Janahan, Distributed supply chain simulation in
GRIDS, in: Proceedings of the 2000 Winter Simulation
Conference, 2000. [18] J. Padmos, B. Hubbard, T. Duczmal, S.
Saidi, How i2 integrates simulation in supply chain
optimisation, in: Proceedings of the 1999 Winter Simulation
Conference, 1999. [19] S. Bagchi, S. Buckley, M. Ettl, G. Lin,
Experience using the IBM supply chain simulator, in:
Proceedings of the 1998 Winter Simulation Conference, 1998.
[20] C. McLean, F. Riddick, The IMS Mission architecture for
distributed manufacturing simulation, in: Proceedings of 2000
Winter Simulation Conference, 2000. [21] B.P. Gan, W. Cai, S.
Turner, Distributed supply-chain simulation using high level
architecture, Transactions of the Society for Computer
Simulation 18 (2) (2001a) 98–109. [22] B.P. Gan, Y. Low, C.
Lim, S. Jain, Bottleneck based modelling of semiconductor
supply chains, in: Proceedings of the International Conference
on Modelling and Analysis of Semiconductor Manufacturing,
2000. [23] B.E. Hirsch, T. Kuhlmann, T.J. Schumacher,
Logistics simulation of recycling networks, Computer in
Industry 24 (1998) 31–38. [24] T. Belhau, C. Strothotte, D.
Ziems, A. Schurholz, M. Schmitz, Modeling and Simulation of
Supply Chain, 1999 (www.uni-magdeburg.de). [25] R.C. Botter,
A. Bergsten Mendes, R. Ferreira de Souza, Simulation model
for the redesign of a supply distribution system, in: Proceedings
of the 1998 Summer Computer Simulation Conference, 1998.
[26] D. Schunk, Using simulation to analyze supply chain, in:
Proceedings of the 2000 Winter Simulation Conference, 2000.
14 S. Terzi, S. Cavalieri / Computers in Industry 53 (2004) 3–
16 [27] Promodel Corporation, Supply Chain Guru, 2002 (www.
promodel.com). [28] G. Archibald, N. Karabakal, P. Karlsson,
Supply chain vs. supply chain: using simulation to compete
beyond the four walls, in: Proceedings of the 1999 Winter
Simulation Conference, 1999. [29] A.J. Siprelle, D. Parsons,
21. R.A. Phelps, SDI Industry Pro: simulation for enterprise wide
problem solving, in: Proceedings of the 1999 Winter Simulation
Conference, 1999. [30] J. Zhang, I. Hunt, C. O’Connor,
Simulating and modeling supply chain activities in the food
industry, 1998 (www.cimru.nuigalway.ie). [31] J.G. Van der
Vorst, A.J. Beulens, P. Van Beek, Modelling and simulating
multi-echelon food systems, European Journal of Operational
Research (2000) 354–366. [32] A. Alfieri, P. Brandimarte,
Object-oriented modelling and simulation of integrated
production/distribution systems, Computer Integrated
Manufacturing Systems 10 (1997) 261– 266. [33] B.P. Gan, Y.
Low, C. Lim, S. Jain, W. Cai, W. Hsu, S. Huang, S. Turner,
Parallel discrete-event simulation of a supply chain in
semiconductor industry, in: Proceedings at HPC ASIA, 2000.
[34] S. Strasburger, T. Schulze, M. Schumann, H. Menzler,
Distributed traffic simulation based on the HLA, in:
Proceedings of the 1998 Simulation Interoperability Workshop,
Fall, 1998. [35] A. Brun, S. Cavalieri, M. Macchi, A. Portioli-
Staudacher, S. Terzi, Distributed simulation for supply chain
co-ordination, in: Proceedings of the 12th International
Working Seminar on Production Economics, Igls, Austria, 2002.
[36] D. Kim, H. Cao, S. Buckley, Modelling and simulation of
supply chain management based on DEVS and CORBA
framework, in: Proceedings of the 1999 Winter Simulation
Conference, 1999. [37] J. Ritchie Dunham, E. Anderson, A
strategic supply chain simulation model, in: Proceedings of the
2000 Winter Simulation Conference, 2000. [38] J. Zhang, I.
Hunt, J. Browne, The development of an extended enterprise
supply chain simulator, 1998 (www.cimru.nuigalway.ie). [39]
M.W. Barnett, C. Miller, Analysis of the virtual enterprise using
distributed supply chain modelling and simulation: an
application of e-SCOR, in: Proceedings of the 2000 Winter
Simulation Conference, 2000. [40] H.B. Chen, O. Bimber, C.
Chatre, E. Poole, S. Buckley, eSCA: A thin client/server/web-
enabled system for distributed supply chain simulation, in:
Proceedings of the 1999 Winter Simulation Conference, 1999.
22. [41] E.H. Page, S. Griffin, S. Rother, Providing conceptual
framework support for distributed web-based simulation within
the HLA, SPIE, Orlando, 1998, pp. 287–292. [42] R. Phelps,
D.J. Parsons, A. Siprelle, The SDI industry product suite:
simulation from the production line to the supply chain, in:
Proceedings of the 2000 Winter Simulation Conference, 2000.
[43] B.P. Gan, J. Turner, W. Cai, P. Xavier, Visualization of a
distributed semiconductor supply chain simulation, in:
Proceedings of the 2001 European Simulation Interoperability
Workshop, 2001. [44] B.P. Gan, J. Turner, W. Cai, Hierarchical
Federations: an architecture for information hiding, PADS USA,
2001. [45] S. Strasburger, T. Schulze, M. Schumann, E.
Bluemel, Using HLA for factory simulation, in: Proceedings of
the 1998 Fall Simulation Interoperability Workshop, 1998. [46]
S. Strasburger, T. Schulze, U. Klein, Migration of HLA into
civil domains: solutions and prototypes for transportation
applications, Simulation 73 (5) (1999) 296–303. [47] WILD
Consortium, in: A. Bruzzone, R. Mosca, R. Revetria (Eds.), The
Wild Book, University of Genoa, Italy, 2001. [48] G. Seliger,
D. Krutzfeldt, P. Lorenz, S. Strasburger, On the HLA and
Internet-based coupling of commercial simulation tools for
production networks, in: WebSim Symposium, 1999. [49] Y.
Chang, H. Makatsoris, Supply chain modelling using
simulation, International Journal of Simulation 1 (2001). [50]
D. Berry, M. Naim, Quantifying the relative improvements of
redesign strategies in a PC supply chain, International Journal
of Production Economics 46–47 (1996) 181–196. [51] D.
Burnett, T. Le Baron, Efficiently modeling warehouse systems,
in: Proceedings of the 2001 Winter Simulation Conference,
2001. [52] S. Cavalieri, V. Cesarotti, S. Grassi, A. Spandri,
Coordinated planning models for managing spare parts
inventory in after sales service, in: Proceedings of the 12th
International Working Seminar on Production Economics, Igls,
Austria, 2002. [53] B.P. Gan, L. McGinnis, Distributed
simulation with incorporated APS procedures for high-fidelity
supply chain optimisation, in: Proceedings of the 2001 Winter
23. Simulation Conference, 2001. [54] K. Hafeez, M. Griffiths, J.
Griffiths, M. Naim, Systems design of two-echelon steel
industry supply chain, International Journal of Production
Economics 45 (1996) 121–130. [55] R.G. Ingalls, C. Kasales,
CSCAT: the Compaq supply chain analysis tool, in: Proceedings
of the 1999 Winter Simulation Conference, 1999. [56] S. Jain,
L. Collins, R. Workman, E. Ervin, Development of a high-level
supply chain simulation model, in: Proceedings of the 2001
Winter Simulation Conference, 2001. [57] Y. Luo, P.
Wirojanagud, R. Caudil, Network-based Optimisation and
Simulation of Sustainable e-Supply Chain Management, IEEE,
Piscataway, NJ, 2001. [58] R. Mielke, Applications for
enterprise simulation, in: Proceedings of the 1999 Winter
Simulation Conference, 1999. [59] F.J. Persson, D. Olhager,
Performance simulation of supply chain designs, International
Journal of Production Economics 77 (2002) 231–245. [60] D.
Petrovic, Simulation of supply chain behaviour and performance
in an uncertain environment, International Journal of Production
Economics 71 (2001) 429–438. [61] R.A. Phelps, D.J. Parsons,
A.J. Siprelle, SDI supply chain builder: simulation from atoms
to the enterprise, in: S. Terzi, S. Cavalieri / Computers in
Industry 53 (2004) 3–16 15 Proceedings of the 2001 Winter
Simulation Conference, 2001. [62] J. Ventateswaran, M.
Jafferali, Y. Son, Distributed simulation: an enabling
technology for the evaluation of virtual enterprises, in:
Proceedings of the 2001 Winter Simulation Conference, 2001.
[63] G. Zulch, U. Jonsson, J. Fischer, Hierarchical simulation of
complex production systems by coupling of models,
International Journal of Production Economics 77 (2002) 39–51.
Sergio Terzi is a PhD student of Politecnico di Milano,
Department of Economics, Industrial and Management
Engineering, Laboratory of Production Systems Design and
Management. He is also taking PhD in conjunction with CRAN
laboratories, University of Nancy I, France. He received his
MSc in management engineering degrees from the University of
Castellanza in 1999 and from the same university he received
24. his BS degrees in economics in 2002. His current research
interests are parallel and distributed simulation applied to
industry and supply chain context, technologies enabling
product lifecycle management within SME and modelling of
production systems. Sergio Cavalieri is currently associate
professor at the Department of Industrial Engineering of the
University of Bergamo. Graduated in July 1994 in management
and production engineering, in 1998 he got the PhD title in
management engineering at the University of Padua. His main
fields of interest are modelling and simulation of manufacturing
systems, application of multi-agent systems and soft-computing
techniques (genetic algorithms, ANNs, expert systems) for
operations and supply chain management. He has been
participating to various research projects at national and
international level. He has published two books and about 40
papers on national and international journals and conference
proceedings. He is currently co-ordinator of the IMS Network of
Excellence Special Interest Group on Benchmarking of
Production Scheduling Systems and member of the IFAC-TC on
Advanced Manufacturing Technology. 16 S. Terzi, S. Cavalieri /
Computers in Industry 53 (2004) 3–16
Shelby Shelving Case StudyShelby ShelvingShelby Shelving is
a small company that manufactures two types of shelves for
grocery stores. Model S is the standard model, and model LX is
a heavy-duty model. Shelves are manufactured in three major
steps: stamping, forming, and assembly. In the stamping stage, a
large machine is used to stamp, i.e., cut, standard sheets of
metal into appropriate sizes. In the forming stage, another
machine bends the metal into shape. Assembly involves joining
the parts with a combination of soldering and riveting. Shelby's
stamping and forming machines work on both models of
shelves. Separate assembly departments are used for the final
stage of production.The hours required on each machine for
each unit of product are shown in the range B5:C6 of the
Accounting Data sheet. For example, the production of one
25. model S shelf requires 0.25 hour on the forming machine. Both
the stamping and forming machines can operate for 800 hours
each month. The model S assembly department has a monthly
capacity of 1900 units.The model LX assembly department has a
monthly capacity of only 1400 units. Currently Shelby is
producing and selling 400 units of model S and 1400 units of
model LX per month.Model S shelves are sold for $1800, and
model LX shelves are sold for $2100. Shelby's operation is
fairly small in the industry, and management at Shelby believes
it cannot raise prices beyond these levels because of the
competition. However, the marketing department feels that
Shelby can sell as much as it can produce at these prices. The
costs of production are summarized in the Accounting Data
sheet. As usual values in blue cells are given where as other
values are calculated from these.Management at Shelby just met
to discuss next month's operating plan. Although the shelves are
selling well, the overall profitability of the company is a
concern. Doug Jameson, the plant's engineer suggested that the
current production of model S shelves be cut back. According to
him, "Model S shelves are sold for $1800 per unit, but our costs
are $1839. Even though we're only selling 400 units a month,
we're losing money on each one. We should decrease production
of model S." The controller, Sarah Cranston disagreed.She said
that the problem was the model S assembly department trying to
absorb a large overhead with a small production volume. "The
model S units are making a contribution to overhead. Even
though production doesn't cover all of the fixed costs, we'd be
worse off with lower production."Your job is to develop an LP
Model of Shelby's problem, then run Solver and finally make a
recommendation to Shelby management with a short verbal
argument supporting Doug or Sarah.Notes on Accounting Data
calculations: The fixed overhead is distributed using activity-
based costing principles. For example, at current production
levels, the forming machine spends 100 hours on model S
shelves and 700 hours on model LX shelves. The forming
machine is used 800 hours of the month, of which 12.5% of the
26. time is spent on model S shelves and 87.5% is spent on model
LX shelves. The $95,000 of fixed overhead in the forming
department is distributed as $11,875 (= 95,000 x 0.125) to
model S shelves and $83,125 (= 95,000 x 0.875) to model LX
shelves. The fixed overhead per unit of output is allocated as
$29.69 (= model LX shelves. The fixed overhead per unit of
output is allocated as $29.69 (= 11,875/400) for model S and
$59.38 (= 83,125/1400) for model LX. In the calculation of the
standard overhead cost, the fixed and variable costs are added
together, so that the overhead cost for the forming department
allocated to a model S shelf is $149.69 (= 29.69 + 120, shown
in cell G20 rounded up to $150). Similarly, the overhead cost
for the forming department allocated to a model LX shelf is
$229.38 (= 59.38 + 170, shown in cell H20 rounded down to
$229).For this problem you will submit the final product which
will be an Excel spreadsheet used to create the model and either
a Word document or a Power Point presentation. The final
project will be graded not only on the accuracy of the
quantitative solutions, but also the analytical approach used and
the presentation of the results. Keep in mind that this course is
designed for individuals interested in Business Management.
As such, the final presentation should be appropriate for a
presentation in a professional setting. It will be necessary to
clearly explain the case study and present the results in a
professional, yet easily understood manner.The presentation
should clearly state the objective, the constraints in obtaining
that objective, the factors that can be varied, the sensitivity of
the model to the variable factors, and the potential weakness of
the conclusions.
Accounting DataShelby Shelving Data for Current Production
ScheduleMachine requirements (hours per unit)Given monthly
overhead cost dataModel SModel LXAvailableFixed Variable
SVariable LXStamping0.30.3800Stamping$125,000$80
Chris Albright: Overhead per unit of model S
$90
27. Chris Albright: Overhead per unit of model
LXForming0.250.5800Forming$95,000$120$170Model S
Assembly$80,000$165$0Model SModel LXModel LX
Assembly$85,000$0$185Current monthly
production400.01400.0Standard costs of the shelves -- based on
the current production levelsHours spent in departmentsModel
SModel LXModel SModel LXTotalsDirect
materials$1,000$1,200Stamping120.0420.0540.0Direct
labor:Forming100.0700.0800.0 Stamping$35$35
Forming$60$90Percentages of time spent in departments
Assembly$80$85Model SModel LXTotal direct
labor$175$210Stamping22.2%77.8%Overhead
allocationForming12.5%87.5% Stamping$149$159
Forming$150$229Unit selling price$1,800$2,100
Assembly$365$246Total overhead$664$635Assembly
capacity1900.01400.0Total cost$1,839$2,045
System Dynamics Modeling: TOOLS FOR LEARNING IN A
COMPLEX WORLD.
Authors:
Sterman, John D.1,2
Source:
California Management Review. Summer2001, Vol. 43 Issue 4,
p8-25. 18p. 3 Diagrams, 1 Graph.
Document Type:
Article
Subject Terms:
*SYSTEM analysis
*INDUSTRIAL management
*DECISION making
*ORGANIZATIONAL change
*SIMULATION methods & models
*MATHEMATICAL models
*INDUSTRIAL organization (Economic theory)
28. *INDUSTRIAL designLEARNINGCOMPUTER simulation
Abstract:
Today's problems often arise as unintended consequences of
yesterday's solutions. Business and public policy settings suffer
from policy resistance, the tendency for well-intentioned
interventions to be defeated by the response of the system to the
intervention itself. Just as an airline uses flight simulators to
help pilots learn, system dynamics enables us to create
management flight simulators to avoid policy resistance and
design more effective policies. System dynamics is also a
process for working with high-level teams designed to improve
the chances for implemented results. This article discusses how
system dynamics can be used effectively to design high-
leverage policies for sustainable improvement and introduces
the next three articles in this issue discussing the application of
system dynamics to a variety of critical issues facing business
leaders today. [ABSTRACT FROM AUTHOR]
Contents
1. Dynamic Complexity
2. Feedback
3. Time Delays
4. Stocks and Flows
5. Attribution Errors and False Learning
6. Tools of System Dynamics
7. Applications
8. Notes
9. Table 1. Examples of Policy Resistance
10. Table 2. Dynamic Complexity
11. Dynamic Complexity Arises Because Systems Are:
ListenSelect:
Accelerating change is transforming our world, from the prosaic
(such as the effect of information technology on the way we use
the telephone) to the profound (such as the effect of greenhouse
gases on the global climate). Some of these changes amaze and
delight us; others impoverish the human spirit and threaten our
29. survival. More important, thoughtful leaders increasingly
suspect that the tools they have been using have not only failed
to solve the persistent problems they face, but may in fact be
causing them. All too often, well-intentioned efforts to solve
pressing problems create unanticipated side effects. Our
decisions provoke unforeseen reactions. The result is policy
resistance, the tendency for interventions to be defeated by the
response of the system to the intervention itself. From
California's failed electricity reforms, to road building programs
that create suburban sprawl and actually increase traffic
congestion, to the latest failed change initiative in your
company, our best efforts to solve problems often make them
worse. Table 1 lists some examples, including economic, social,
and environmental issues.
While we like to imagine that new technologies and accelerating
change present us with new and unique challenges, policy
resistance is nothing new. In 1516, Sir Thomas More wrote in
Utopia about the problems of policymaking, saying "And it will
fall out as in a complication of diseases, that by applying a
remedy to one sore, you will provoke another; and that which
removes the one ill symptom produces others." The late
biologist and essayist Lewis Thomas, in an essay entitled "On
Meddling," provided both a diagnosis and a solution:
When you are confronted by any complex social system, such as
an urban center or a hamster, with things about it that you're
dissatisfied with and anxious to fix, you cannot just step in and
set about fixing with much hope of helping. This realization is
one of the sore discouragements of our century . . . You cannot
meddle with one part of a complex system from the outside
without the almost certain risk of setting off disastrous events
that you hadn't counted on in other, remote parts. If you want to
fix something you are first obliged to understand . . . the whole
system. . . . Intervening is a way of causing trouble.(n1)
However, how can one come to understand the whole system?
How does policy resistance arise? How can we learn to avoid it,
to find the high-leverage policies that can produce sustainable
30. benefit?
For many, the solution lies in systems thinking--the ability to
see the world as a complex system, in which we understand that
"you can't do just one thing" and that "everything is connected
to everything else." With a holistic worldview, it is argued, we
would be able to learn faster and more effectively, identify the
high leverage points in systems, and avoid policy resistance. A
systemic perspective would enable us to make decisions
consistent with our long-term best interests and the long-term
best interests of the system as a whole.(n2)
The challenge facing us all is how to move past slogans about
accelerating learning and systems thinking to useful tools that
help us understand complexity, design better operating policies,
and guide effective change. System dynamics is a method to
enhance learning in complex systems. Just as an airline uses
flight simulators to help pilots learn, system dynamics is,
partly, a method for developing management flight simulators
(often based on formal mathematical models and computer
simulations) to help us learn about dynamic complexity,
understand the sources of policy resistance, and design more
effective policies.
However, successful intervention in complex dynamic systems
requires more than technical tools and mathematical models.
System dynamics is fundamentally interdisciplinary. Because
we are concerned with the behavior of complex systems, system
dynamics is grounded in the theory of nonlinear dynamics and
feedback control developed in mathematics, physics, and
engineering. Because we apply these tools to the behavior of
human as well as technical systems, system dynamics draws on
cognitive and social psychology, organization theory,
economics, and other social sciences. To solve important real
world problems, we must learn how to work effectively with
groups of busy policymakers and how to catalyze change in
organizations.
To introduce this special section on system dynamics, I briefly
discuss how policy resistance arises from the mismatch between
31. the dynamic complexity of the systems we have created and our
cognitive capacity to understand that complexity. I then
summarize the system dynamics approach, illustrate some tools,
and discuss some of the limitations and pitfalls. Finally, I
summarize the applications discussed in the articles in this
special section. Readers interested in learning more about
system dynamics and about successful applications should refer
to the growing scholarly and practitioner literature.(n3)
Dynamic Complexity
Policy resistance arises because, as wonderful as the human
mind is, the complexity of the world dwarfs our
understanding.(n4) Our mental models are limited, internally
inconsistent, and unreliable. Our ability to understand the
unfolding impacts of our decisions is poor. We take actions that
make sense from our short-term and parochial perspectives, but
due to our imperfect appreciation of complexity, these decisions
often return to hurt us in the long run. To understand the
sources of policy resistance, we must therefore understand both
the complexity of systems and the mental models that we use to
make decisions.
Most people think of complexity in terms of the number of
components in a system or the number of possibilities one must
consider in making a decision. The problem of optimally
scheduling an airline's flights and crews is highly complex, but
the complexity lies in finding the best solution out of an
astronomical number of possibilities. Such problems have high
levels of combinatorial complexity. However, most cases of
policy resistance arise from dynamic complexity--the often
counterintuitive behavior of complex systems that arises from
the interactions of the agents over time. Dynamic complexity
can arise even in simple systems with low combinatorial
complexity. For example, courses in system dynamics often
begin with the "Beer Distribution Game," a role-playing board
game simulation representing a manufacturing supply chain.(n5)
The game is highly simplified--there is only one SKU, not tens
of thousands. Each player has exactly one customer and one
32. supplier. Yet players consistently generate wild fluctuations in
production and inventory, and average costs are ten times
greater than optimal. Complex and dysfunctional dynamics arise
from a game you can play on your dining room table and whose
rules can be learned in 15 minutes.
Table 2 describes some of the characteristics of complex
dynamic systems. These attributes are common, but
counterintuitive. Where the world is dynamic, evolving, and
interconnected, we tend to make decisions using mental models
that are static, narrow, and reductionist. Among the elements of
dynamic complexity people find most problematic are feedback,
time delays, stocks and flows (accumulations), and nonlinearity.
Feedback
One cause of policy resistance is our tendency to interpret
experience as a series of events, for example, "inventory is too
high," or "sales fell last month." Accounts of who did what to
whom are the most common mode of discourse, from the
mailroom to the boardroom, from the headlines to the history
books. We are taught from an early age that every event has a
cause, which in turn is an effect of some still earlier cause:
"Inventory is too high because sales unexpectedly fell. Sales
fell because the competitors lowered their price. The
competitors lowered their price because . . ." Such event-level
explanations can be extended indefinitely. They allow us to
blame others for our difficulties, but also, as a consequence,
reinforce the belief that we are powerless.
The event-oriented, open-loop worldview leads to an event-
oriented, reactionary approach to problem solving (Figure 1).
We assess the state of affairs and compare it to our goals. The
gap between the situation we desire and the situation we
perceive defines our problem. For example, suppose your firm's
profits fall below Wall Street expectations. You need to boost
profits, or you'll be searching for a new job. You consider
various courses of action, select the options you deem best, and
implement them. You might initiate various process
improvement programs to boost productivity, increase the
33. number of new products in the development pipeline to boost
sales, and announce a round of layoffs to cut expenses. Your
consultants, spreadsheets, and pro forma analyses suggest these
decisions will restore growth and profitability. The consultants
move on, and you turn to other pressing issues. Problem solved-
-or so it seems.
Contrary to the sequential, open-loop view in Figure 1, real
systems react to our interventions. There is feedback: The
results of our actions define the situation we face in the future.
The new situation alters our assessment of the problem and the
decisions we take tomorrow (see the top of Figure 2). Moreover,
as shown in the bottom of Figure 2, our actions may also trigger
side effects we didn't anticipate. Other agents, seeking to
achieve their goals, react to restore the balance we have upset.
Policy resistance arises because we do not understand the full
range of feedbacks operating in the system. For example, the
improvement initiatives you mandated never got off the ground
because layoffs destroyed morale and increased the workload
for the remaining employees. New products were rushed to
market before all the bugs were worked out, so now warranty
claims explode while sales slump. Rising customer complaints
overwhelm your call centers and service organization. Stressed
by long hours, budget cuts, and continual crisis, your best
engineers and most experienced managers quit to take better
jobs with your competitors. Yesterday's solutions become
today's problems. Without an understanding of the feedback
processes that create these outcomes as a consequence of our
own decisions, we are likely to see these new crises as more
evidence confirming our view that the world is unpredictable,
unpleasant, and uncontrollable--that all we can do is react to
events.
Time Delays
Time delays between taking a decision and its effects on the
state of the system are common and particularly troublesome.
Delays in feedback loops create instability and increase the
tendency of systems to oscillate. As a result, decision makers
34. often continue to intervene to correct apparent discrepancies
between the desired and actual state of the system long after
sufficient corrective actions have been taken to restore the
system to equilibrium. Research shows convincingly that people
commonly ignore time delays, even when the existence and
contents of the delays are known and reported to them, leading
to overshoot and instability.(n6)
More subtly, delays reduce our ability to accumulate
experience, test hypotheses, and learn. A 1988 study estimated
the improvement half-life in a wide range of firms. The
improvement half-life is the time required to cut the defects
generated by a process in half. Improvement half-lives were as
short as a few months for simple processes with short cycle
times (for example, reducing operator error in a job shop) while
complex processes with long cycle times (such as product
development) had improvement half lives of several years or
more.(n7)
Stocks and Flows
Stocks and flows--the accumulation and dispersal of resources--
are central to the dynamics of complex systems. A population is
increased by births and decreased by deaths. A firm's inventory
is increased by production and decreased by shipments,
spoilage, and shrinkage. It is only in the past decade or so that
the strategic management community has begun to consider the
role of stocks and flows explicitly, as the resource-based view
of the firm has grown in popularity. The resource-based view
expanded the definition of a firm's resources beyond tangible
stocks such plant, equipment, cash, and other traditional balance
sheet items to include less obvious but more important stocks
underlying firm capabilities, such as employee skills, customer
loyalty, and other forms of intangible human, social, and
political capital.(n8)
Nevertheless, research shows that people's intuitive
understanding of stocks and flows is poor. Figure 3 illustrates
the problem with one of the simplest stock-flow structures: a
bathtub. The stock of water in the tub is filled by the inflow and
35. drained by the outflow. From the graphs of the flows it is easy
to infer the trajectory of the stock, and without use of calculus
or any mathematics beyond simple arithmetic. However, the
average performance of graduate students at an elite business
school was only 46%. In this and related stock-flow problems,
many people drew trajectories that violated basic laws of
physics such as conservation of matter.(n9)
Attribution Errors and False Learning
Some people believe that experience and market forces enable
good managers to learn quickly about the feedbacks and side
effects of their decisions, including, as in the example above,
the morale and workload impacts of layoffs or the low quality
resulting from rushing a product to market. Unfortunately, few
of us can say we've never faced such situations or been
blindsided by unanticipated side effects of our own actions. The
heuristics we use to judge causal relationships systematically
lead to cognitive maps that ignore feedbacks, nonlinearities,
time delays, and other elements of dynamic complexity. To
judge causality, we use cues such as temporal and spatial
proximity of cause and effect, temporal precedence of causes,
covariation, and similarity of cause and effect. In complex
systems, however, cause and effect are often distant in time and
space, and the delayed and distant consequences of our actions
are different from and less salient than their proximate effects--
or are simply unknown. The interconnectedness of complex
systems causes many variables to be correlated with one
another, confounding the task of judging cause. Research shows
that few mental models incorporate any feedback loops. For
example, studies have found virtually no feedback loops in the
cognitive maps of political leaders; rather, the leaders focused
on particular decisions they might make and their likely
consequences--an event-level representation.(n10) Experiments
in causal attribution show people tend to assume each event has
a single cause and often cease their search for explanations
when the first sufficient cause is found.(n11)
A fundamental principle of system dynamics states that the
36. structure of the system gives rise to its behavior. In complex
systems, different people placed in the same structure tend to
behave in similar ways. However, people have a strong
tendency to attribute the behavior of others to dispositional
rather than situational factors--that is, to character (and, in
particular, character flaws) rather than to the system in which
these people are embedded. The tendency to blame other people
instead of the system is so strong that psychologists call it the
"fundamental attribution error."(n12) In a famous study,
psychologists Robert Rosenthal and Lenore Jacobson told a
group of grade school teachers that test scores showed a
particular 20% of their students would bloom academically in
the year ahead. At the end of the year, those students showed
larger increases in IQ than the others. There was only one
problem: the apparently "gifted" students had been chosen
entirely at random.(n13) The teachers, without realizing it
themselves, set higher expectations for the students labeled as
gifted, gave them more help, provided more praise. Thus
nurtured, these lucky students did bloom, though they were no
different at the start than any of the other children in the class.
The others necessarily received less attention, less help, and
less praise, falling farther and farther behind. Without the
ability to see how they themselves were part of the classroom
and community system, how their own behavior helped some to
excel while undermining others, the teachers interpreted events
such as test grades and class participation as evidence
confirming their preconceptions: The high performance of the
students in the gifted group proved that they were truly gifted,
and the poor performance of the rest proved that these were in
fact the low achievers.
Because they were unaware of the ways in which the system
structure shaped their behavior, the teachers learned a false
lesson with pernicious consequences. The attribution of
behavior to individuals and their character rather than system
structure diverts our attention from the high leverage points
where redesign of the system can have significant, sustained,
37. beneficial effects on performance. When we attribute behavior
to people rather than system structure the focus of management
becomes scapegoating and blame rather than the design of
organizations in which ordinary people can achieve
extraordinary results.
Tools of System Dynamics
To improve our ability to learn about and manage complex
systems, we need tools capable of capturing the feedback
processes, stocks and flows, time delays, and other sources of
dynamic complexity. The tools must also enable us to
understand how these structures create a system's dynamics and
generate policy resistance. They must help us evaluate the
consequences of new policies and new structures we might
design. These tools include causal mapping and simulation
modeling.
Much of the art of system dynamics modeling lies in
discovering and representing the feedback processes and other
elements of complexity that determine the dynamics of a
system. One might imagine that there is an immense range of
different feedback processes to be mastered before one can use
system dynamics effectively. In fact, all dynamics arise from
the interaction of just two types of feedback loops, positive (or
self-reinforcing) and negative (or self-correcting) loops.
Positive loops tend to reinforce or amplify whatever is
happening in the system: The more nuclear weapons NATO
deployed during the Cold War, the more the Soviet Union built,
leading NATO to build still more. If a firm lowers its price to
gain market share, its competitors may respond in kind, forcing
the firm to lower its price still more. The larger the installed
base of Microsoft software and Intel machines, the more
attractive the Wintel architecture became as developers sought
the largest market for their software and customers sought
systems compatible with the most software; the more Wintel
computers sold, the larger the installed base. These positive
feedback loops are what chemists call autocatalytic--self-
stimulating processes that generate their own growth, leading to
38. arms races, price wars, and the phenomenal growth of Microsoft
and Intel, respectively.
Negative loops counteract and oppose change. The less nicotine
in a cigarette, the more smokers must consume to get the dose
they need. The more attractive a neighborhood or city, the
greater the migration from surrounding areas will be--increasing
unemployment, housing prices, crowding in the schools, and
traffic congestion until the city is no more attractive than other
places people might live. The larger the market share of
dominant firms, the more likely is government antitrust action
to limit their monopoly power. These loops all describe
processes that tend to be self-limiting, processes that create
balance and equilibrium.(n14)
As an illustration, suppose your firm is about to launch an
innovative new product, one that creates an entirely new
category with substantial market potential, but for which no
market yet exists (e.g., personal computers in the early 1980s).
You need to understand how quickly and in what fashion the
market might develop, how you can stimulate adoption, how the
market will saturate, how to design the marketing mix and
pricing strategy, and a host of other issues. You could begin by
identifying some of the positive feedback processes that could
stimulate adoption, and you could map them with a causal loop
diagram (CLD).
Figure 4a shows two of the feedback processes you could
identify. If the new product is sufficiently attractive, the early
adopters will generate favorable word of mouth (WOM),
stimulating further adoption, increasing the adopter population,
and leading to still more WOM, in a positive feedback. The
arrows in the diagram indicate the causal relationships. The
positive (+) signs at the arrowheads indicate that the effect is
positively related to the cause. Here, an increase in the adopter
population causes the number of word of mouth encounters to
rise above the number that would have occurred without the
increase (and vice versa: a decrease in adopters causes the
volume of WOM to fall below what it would have been).
39. Similarly, more favorable WOM leads to a greater adoption
rate, adding to the adopter population, and leading to still more
WOM. The loop is self-reinforcing, hence the loop polarity
identifier R. The loop is named the contagion loop to capture
the process of social contagion by which the innovation spreads.
If the contagion loop were the only one operating, the adoption
rate and adopter population would both grow exponentially.
Of course, no real quantity can grow forever. There must be
limits to growth. These limits are created by negative feedback.
Negative loops are self-correcting. They counteract change. In
the example, growing adoption of the innovation causes various
negative loops to reduce adoption until use of the innovation
comes into balance with its "carrying capacity" in the social and
economic environment. As shown in Figure 4a, the adoption
rate depends not only on word of mouth generated by adopters,
but also on the number of potential adopters: The greater the
number of potential adopters, the greater the probability that
any adopter will come into contact with a potential adopter and,
through word of mouth, cause that individual to adopt the
innovation (hence the positive polarity on the link from
Potential Adopters to the Adoption Rate). However, the greater
the adoption rate, the smaller the remaining population of
Potential Adopters will be, limiting future adoption through
market saturation (hence the negative (-) polarity for the link
from the Adoption Rate to Potential Adopters). The B in the
center of a loop denotes a balancing feedback.
The diagram shown here is deliberately simplified, showing
only the most basic feedbacks. Your mapping process will likely
identify a host of other loops, both reinforcing and balancing,
that might be relevant in the diffusion process. These might
include the learning curve (greater production experience
lowers costs and price, increasing sales and experience still
further) and scale economies (larger production volumes lead to
efficiencies and greater purchasing power, allowing lower
prices that lead to still more sales). Others might include
positive network externalities arising from compatibility and the
40. development of complementary assets (e.g., the Wintel vs.
Macintosh case). You could also identify negative feedbacks
relating to, for example, entry of competitors, cannibalization of
product sales as new generations of the product are introduced,
and so on. Though not shown in the simple diagram in Figure 4,
you could add each such loop to your diagram, creating a rich
map of the feedbacks from which the product life cycle
emerges.
Though there are only two types of feedback loop, complex
systems can easily contain thousands of loops of both types,
coupled to one another with multiple time delays,
nonlinearities, and accumulations. The dynamics of all systems
arise from the interactions of these networks of feedbacks. We
can infer the dynamics of isolated loops such as those shown in
Figure 4a. However, when multiple loops interact, it is
generally impossible to determine what the dynamics will be by
intuition. When intuition fails, we must turn to computer
simulation.
To develop the simulation model, it is useful to augment the
causal diagram to show the important stocks and flows
explicitly, as shown in Figure 4b. The rectangles represent the
stocks, in this case the populations of potential and actual
adopters. The "pipe" connecting the two stocks represents the
flow; in this case, adoption moves people from the potential
adopter population into the adopter population. Figure 4b also
shows how the word of mouth process works in more detail.
Adoption resulting from word of mouth can be modeled as the
product of the rate at which potential adopters have word of
mouth encounters with adopters and the probability of adoption
after such a contact. The more word of mouth encounters or the
more persuasive each encounter, the greater the adoption rate.
The rate at which potential adopters have word of mouth
encounters depends on the total rate at which they have social
contacts and the probability of contacting an adopter. That
probability, in turn, depends on the proportion of adopters in
the social networks to which the potential adopters belong. The
41. total rate at which potential adopters contact others depends on
the size of the potential adopter population and the frequency of
social interactions in that group. Figure 4c shows the equations
for this simple model.
Before simulating, you must estimate the parameters and initial
conditions (e.g., the probability of adoption after contact with
an adopter and the contact frequency). These parameters might
be estimated using statistical means, market research data,
analogous product histories, expert opinion, and any other
relevant sources of data, quantitative or judgmental.
The overall dynamics of the system depend on which feedback
loops are dominant. Figure 4d shows a simulation of the model
compared to the data for the diffusion of a successful new
computer. For a sufficiently attractive innovation, the self-
reinforcing word of mouth loop dominates initially, and the
adoption rate and adopter population grow exponentially. The
growing rate of adoption, however, drains the stock of potential
adopters, eventually constraining the adoption rate due to
market saturation. The dominant feedback loop shifts from the
positive contagion loop to the negative saturation loop. The
shift in loop dominance is a fundamentally nonlinear process,
which arises in this case because adoption requires a word of
mouth encounter between an adopter and a potential adopter.
The shift in loop dominance occurs at the point where the
adoption rate peaks. The behavior of the system shifts from
acceleration to deceleration, and the system gradually
approaches equilibrium.
In this fashion, the modeling process can continue. The model
should be augmented to include the other important loops
identified through causal mapping. Simulation experiments may
suggest new data to collect and new types of experiments to run
to resolve uncertainties and improve the model structure. The
model can also be used to design and evaluate new policies
before implementing them in the real world. The results of these
experiments in the real world can then lead to revisions and
improvements in both the simulation model and the mental
42. models of the decision makers, thus speeding the learning
process.
Simulations are not tools to predict the future. Rather, they are
virtual worlds or microworlds in which managers can develop
decision-making skills, conduct experiments, and play.(n15)
Management flight simulators can be physical models, board
games, or computer simulations. In systems with significant
dynamic complexity, computer simulation will typically be
needed.
Modern system dynamics modeling software makes it possible
for anyone to participate in the modeling process. Graphical
user interfaces enable modelers to quickly sketch a causal
diagram, capturing the feedbacks, stocks and flows, time delays,
and nonlinearities they identify. Equations can be written using
so-called "friendly algebra" so that advanced mathematical
training is no longer necessary (see Figure 4c). Modeling can
now be done in real-time, and with groups. Simulation results
can be viewed immediately. Sensitivity analysis, optimization,
and calibration to data can be largely automated. A model can
easily be converted into an interactive game with an intuitive
interface. Of course, while the software has become easier and
easier to use, modeling is not computer programming and
remains a demanding activity. Better hardware and software do
not replace the thinking process; rather, they provide a means to
improve our mental models and design more effective policies.
They make it possible for everyone to participate in the
modeling process and increase the time available to focus on the
issues of concern.
Tools for learning about complexity must also facilitate the
process of systems thinking and policy design. While the virtual
world enables controlled experimentation, it does not require us
to apply the principles of scientific method. Similarly,
defensive routines and groupthink that thwart learning in teams
can operate in the learning laboratory just as in the real
organization. Effective modeling often requires members of the
client team to recognize the limitations of their inquiry skills
43. and address their own defensive behaviors. Managers
unaccustomed to disciplined scientific reasoning and an open,
trusting environment with learning as its goal will have to build
these basic skills before a system dynamics model--or indeed,
any model--can prove useful. Developing these skills takes
effort and practice.(n16)
The list of successful interventions using system dynamics is
growing. Of course there are also failures, as the community of
modelers continues to learn and improve the tools and process.
Recent successful projects in the business world include
strategy design for a highly successful wireless communications
startup, leasing strategy for a large automaker, supply chain
reengineering in a number of major high-technology firms, a
new marketing strategy for a major credit card organization,
long-range market forecasts and strategy development for a
major commercial aircraft manufacturer, clinical trial and
marketing strategies for new pharmaceuticals, models for
effective management of large-scale projects in software, civil
construction, shipbuilding, aerospace, defense, and commercial
product development--and many others.
Applications
The articles that follow in this issue of the California
Management Review apply system dynamics to some of the
most difficult issues faced by organizations today. How can an
organization escape the trap of firefighting, in which continual
crisis fosters a short-term orientation that prevents investment
in organizational capabilities that could prevent the crises? Why
do so many process improvement programs fail? Why does
product and service quality drift down despite an organization's
efforts to maintain standards and satisfy their customers? Why
don't people learn on their own how to avoid policy resistance
and overcome these problems?
In "Past the Tipping Point: The Persistence of Firefighting in
Product Development," Nelson Repenning, Paulo Goncalves,
and Laura Black develop a formal model of organizational
firefighting. Their model shows how well-intentioned, hard-
44. working engineers and managers can inadvertently slip into a
trap in which low organizational capabilities are self-
perpetuating. For example, in many firms new product
development projects are routinely plagued by unexpected
rework and low quality, forcing the team into last-minute
heroics to hit launch dates. These heroics, with their long hours
and single-minded focus on getting the product out, prevent
people from devoting effort to upstream work on the next-
generation product, which then reaches the launch stage even
farther behind, triggering a new round of crises and the need for
still more heroic firefighting. They show that many policies
undertaken to escape the trap--including many programs to
implement new product development processes and tools--are
self-defeating, and they explore effective policies to overcome
the trap.
Nearly every firm in the U.S. has made quality and customer
satisfaction a centerpiece of their mission and values, spending
billions on quality programs in the process, yet the American
Customer Satisfaction Index is stagnant at about 80% for
manufacturing and only 70% for services, down nearly 7% since
1995. In "Tradeoffs in Responses to Work Pressure in the
Service Industry," Rogelio Oliva examines this paradox.
Obviously service quality can fall if the demand for service
outstrips an organization's resources. Oliva shows that quality
can erode steadily even when demand and resources are, on
average, sufficient. Random variations in workload lead to
temporary periods of high workload that often cause service
workers to cut corners and spend less time with customers in an
attempt to meet throughput and cost targets. These shortcuts
gradually become embedded in norms for customer interaction.
Since service quality is intrinsically subjective and less salient
than cost and throughput metrics, management often interprets
the reduction in the time spent with each customer as a
productivity gain, justifying a reduction in service resources.
Workload during peak times increases still further, forcing
employees to cut corners still more. Oliva shows how these
45. dynamics played out in a major commercial bank, leading to
steady quality erosion and reduced revenue.
Why don't people, particularly senior managers, learn to
recognize and avoid these traps through experience? Why do
firefighting, quality erosion, and short-term thinking persist?
Part of the answer lies in the way our mental models lead us to
interpret the data we receive from complex systems. As in the
example of the teachers discussed above, we tend to assume
cause and effect are closely related in time and space,
attributing events such as low test scores, late product launches,
or customer complaints to the intrinsically low IQ,
undisciplined work habits, or poor attitude of the students,
engineers, or customer service representatives, rather than to
the pressures created by the system in which they are embedded.
In "Nobody Ever Gets Credit for Fixing Problems that Never
Happened: Creating and Sustaining Process Improvement,"
Nelson Repenning and I show how managers in a large
automaker erroneously attributed their difficulties to the poor
attitudes and work habits of employees. Though these
attributions were wrong, the feedback managers received from
the system caused their false beliefs to be strongly self-
fulfilling, crippling their efforts to improve the product
development process. Worse, some managers involved in the
failed effort came away with stronger prejudices and
stereotypes about the low skills and poor attitudes of the
employees, further intensifying cynicism and eroding trust in
the organization, thus making genuine improvement even less
likely. The article closes with case examples of organizations
that have successfully used system dynamics and management
flight simulators to overcome these dynamics and achieve
dramatic results. These successes show that what often prevents
us from overcoming policy resistance and achieving high
performance is not a lack of resources, technical knowledge, or
a genuine commitment to change. What thwarts us is our lack of
a meaningful systems-thinking capability, the capability to learn
about complexity and find the high leverage policies through
46. which we can create the future we truly desire.
Notes
(n1.) Lewis Thomas, The Medusa and the Snail: More Notes of
a Biology Watcher (New York, NY: Viking Press, 1979), p. 90.
(n2.) There are many schools of systems thinking. For a survey,
see George Richardson, Feedback Thought in Social Science
and Systems Theory (Philadelphia, PA: University of
Pennsylvania Press, 1991). See also Peter Senge, The Fifth
Discipline: The Art and Practice of the Learning Organization
(New York, NY: Doubleday, 1991). Some emphasize qualitative
methods, others stress formal modeling. For sources of method
and metaphor, they draw on fields as diverse as anthropology,
biology, engineering, linguistics, psychology, physics, and
Taoism, and seek applications in fields still more diverse. All
agree, however, that a systems view of the world is still rare.
Jay Forrester developed system dynamics in the 1950s at MIT.
(n3.) See John Sterman, Business Dynamics: Systems Thinking
and Modeling for a Complex World (New York, NY:
Irwin/McGraw-Hill, 2000), <www.mhhe.com/sterman>.
Includes extensive references to the literature and a disc
containing over 60 simulation models.
(n4.) See the late Herbert Simon's concept of bounded
rationality. Herbert Simon, Sciences of the Artificial, 3rd ed.
(Cambridge, MA: The MIT Press, 1996).
(n5.) For descriptions of the Beer Game, see Sterman, op. cit.
and Senge, op. cit.
(n6.) See Sterman, op. cit., chapter 17, for discussion and
examples.
(n7.) See Art Schneiderman, "Setting Quality Goals," Quality
Progress, 21/4 (April 1988): 55-57. Sterman et al. show how
these differential improvement rates led to difficulty at a
leading semiconductor manufacturer. J. Sterman, N. Repenning,
and F. Kofman, "Unanticipated Side Effects of Successful
Quality Programs: Exploring a Paradox of Organizational
Improvement," Management Science, 43/4 (1997): 501-521.
(n8.) See I. Dierickx and K. Cool, "Asset Stock Accumulation
47. and Sustainability of Competitive Advantage," Management
Science, 35/12 (December 1989): 1504-1511. Intangibles have
long been included in system dynamics models. See, for
example, Jay Forrester, Collected Papers of Jay W. Forrester
(Waltham, MA: Pegasus Communications, 1975). System
dynamics modeling stresses the importance of and methods to
operationalize and quantify such so-called soft variables
(variables for which no numerical data may be available).
Omitting such concepts assumes their impact is zero, one of the
few assumptions we know to be wrong.
(n9.) For the solution, discussion, and other examples, see
Linda Booth Sweeney and John Sterman "Bathtub Dynamics:
Initial Results of a Systems Thinking Inventory," System
Dynamics Review, 16/4 (2000): 249-294.
(n10.) See Robert Axelrod, The Structure of Decision: The
Cognitive Maps of Political Elites (Princeton, NJ: Princeton
University Press, 1976); Dietrich Dorner, The Logic of Failure
(New York, NY: Henry Holt, 1996).
(n11.) See Scott Plous, The Psychology of Judgment and
Decision Making (New York, NY: McGraw Hill, 1993).
(n12.) See Plous, op. cit.; L. Ross, "The Intuitive Psychologist
and His Shortcomings: Distortions in the Attribution Process,"
in L. Berkowitz, ed., Advances in Experimental Social
Psychology, Volume 10 (New York, NY: Academic Press,
1977).
(n13.) See R. Rosenthal and L. Jacobson, Pygmalion in the
Classroom, expanded edition (New York, NY: Irvington, 1992).
(n14.) Negative loops do not always result in a smooth and
stable adjustment to equilibrium. Time delays can cause
overshoot and oscillation as corrective actions persist too long.
Such delays are pervasive and so too are fluctuations, from the
fluctuations in your blood sugar level (caused by delays in the
synthesis of insulin) to boom and bust cycles in real estate,
semiconductors, shipbuilding, and other industries (caused by
delays in adjusting production and production capacity to
changes in demand and prices). See Sterman, op. cit.