Proceedings of the 2005 IEEE
International Conference on Robotics and Automation
Barcelona, Spain, April 2005
Bullwhip effect in Integrated Manufacturing and
Service Networks

N. Viswanadham, Vijay Desai, Roshan Gaonkar


The Logistics Institute – Asia Pacific


National University of Singapore, Singapore 119260


tlinv@nus.edu.sg; tlivd@nus.edu.sg; roshan@gaonkar.com



Abstract - An Integrated Manufacturing and Service
maintenance and serviceability in addition to
Network (IMSN) is a grouping of companies, working
manufacturability at the design and planning stage in order
together to offer a bundle of products and product-related
to be efficient and effective.
services, that deliver value to customers over the entire useful

In this paper, we attempt to study the benefits of such
life of the product, from purchase to disposal. The success of
an approach based on integrating manufacturing and
such an alliance is highly dependent on the seamless
integration and interaction of business strategy, operational
service operations. More specifically, we want to quantify
processes and enterprise systems between constituent the benefits of information sharing between the design,
companies and their design, manufacturing and sales
manufacturing and sales divisions of a manufacturer and
operations. In this paper, we present system dynamics models
its associated service centers. Towards this end we develop
to study the behavior of an IMSN and investigate the
cause-effect diagrams and system dynamic models to
bullwhip effects that arise within such networks due to
mimic the operation of an IMSN.
mismatches and delays between the manufacturing and

An important consideration in designing an IMSN is
service systems within the network. We show that integration
the bullwhip effect in the integrated system which can lead
and collaboration between the manufacturing and service
to the service system exhibiting oscillatory behavior in
operations with two-way information flow between them
enhances profitability and minimizes the bullwhip effect
resource levels for repair crews, spare parts, etc. We
within repair centers.
develop system dynamic models for an IMSN and employ

it to establish the benefit of:
Index Terms – Manufacturing, Service Network, Supply
1. Exploiting feedback from the service centre during
Chains, Service Center Design.
product design to improve product serviceability and

reliability.
I. INTRODUCTION
2. Employing sales information in planning future service

The increasingly competitive nature of markets, driven
capacity requirements and reducing bullwhip effect.
by globalization and the maturation of markets in
developed countries, has put the customer in the driving
Due to the unique nature of issues addressed, our study
seat. Consequentially, customer loyalty has become a key
here on the development of models for IMSNs and
driver of a company's competitiveness and profitability.
analysis of their behavior is a significant contribution to
Hence, in order to retain customers and maintain market
the literature on design of manufacturing-service networks.
share, companies have no choice but to design,
II. LITERATURE SURVEY
manufacture and sell offerings that customers want and
demand. Increasingly customers are looking for simple and
One can view IMSN from multiple perspectives, namely
convenient end-to-end solutions that satisfy its immediate
the evolution in customer preferences, business integration
needs and desires, without the need for getting tangled in
trends, economic development and planning management
associated activities and processes that do not directly
trends at the level of the product lifecycle, as shown in
serve its interests.
Figure 1.
The primary characteristic of an IMSN that
In terms of literature tracking the development of
differentiates it from its competitors is its ability to
IMSNs from the customer perspective, there are lots of
appreciate and exploit the interaction between the product
white papers and reports from consultants, such as
development, manufacturing and after-sales operations of a
Accenture and McKinsey & Co., on the benefits of selling
product. Constituent companies within the network solutions rather than products [4]. It is now established
continually interact and share information and knowledge
through various surveys that profit margins in service are
across all phases of the product development and much higher than in product sales. For certain types of
deployment cycle, namely design, manufacturing, sales
industrial equipment such as heavy machines and aircraft
and after-sales service. Decisions made during design and
engines service is an integral part of product sales
manufacturing phases affect the performance and costs
proposition. In terms of the economic development, from
during the service phase. On the other hand, provision of
manufacturing to service based economies, there have
good service boosts product sales and the feedback from
been studies on the importance of manufacturing in a
service operations to design can help produce more service economy [3]. Additionally, while the supply chain
reliable products. It is thus important to take a holistic and
process integrates all activities from product development
integrated view of a company's operations incorporating
to sales of a product, the integration of the service phase,
0-7803-8914-X/05/$20.00 ©2005 IEEE.
3005

---

physically, informationally and organizationally, with the
deteriorates and it needs to be serviced at a service center,
rest of the supply chain is an emerging trend.
which typically would employ a variety of resources such

as repair crews, disassembly tools, diagnostic equipment,
and spare parts. The resources of the service center
determine the service levels in terms of the delivery time.
The service center capacity is determined by the
availability of spare parts and that of skilled repairmen. In
this study we assume that the capacity is variable and can
be changed with a delay of four months.

It is known from the literature that after sales service
quality influences the product sales particularly from the
past customers. We incorporate this in our system
dynamics model by making service level a variable that
influences product sales. Better service level would entice
Figure 1: Perspectives on Product Service Convergence
a customer to buy the product, thereby ramping up its

sales. Information exchange between the design teams,
Traditionally, most attention in the Product Lifecycle
service centers, manufacturing plants, and marketing and
(PLC) has been given to the direct chain of events i.e. from
sales can lead to dramatic improvements in the
the design to the customer, but the service and recycling
performance of the manufacturing service system, by
phases are not integrated with other phases in the PLC. In
ensuring proper coordination between all the parties
service literature, preventive maintenance [5], the concerned. In particular, future service center resource
repairman problem [5] and spare parts management have
planning can be enhanced by sharing information on past
received a lot of attention. Cohen et al. [2] report on a
regional sales, whereas information on current service
benchmarking study of 14 companies in electronics, levels provided by repair centers can help improve the
computing and communication industries. Cohen et al [1]
sales forecasts of the marketing department, by adjusting
build a product life-cycle model to study a set of strategic
for drops or gains in service levels as we show in the
choices facing the manufacturer of durable good as he
following sections.
designs the joint product/service bundle.

Our motivation here is slightly different from the
IV. MODELING IMSNS
previous literature, since we focus on proving the benefits

We first present a simple model to show benefits of
of collaboration between manufacturers and their service
collaboration between the manufacturing, marketing and
partners, so that each phase in the PLC can benefit and
after-sales service teams. The model demonstrates that
contribute towards the creation of better products or
sharing information between manufacturing and service
attainment of higher service levels.
phases can have a huge impact on the performance of the
III.
system. In particular, the interaction between the
MODEL DESCRIPTION
production rates, sales levels, population of deployed
A typical IMSN is characterized by interaction goods, repair rates, service levels amongst others is
between the design, manufacturing, sales and service
considered in this model, as illustrated in Figure 3.
operations, as shown in Figure 2. We focus on complex
products that are economically attractive to maintain and
service rather than replace.

Figure 3: Cause-Effect diagram for Conceptual Model

On the supply side, we assume that goods are
produced at a constant rate and they are sold at a rate
dependent on the utility function. A higher utility to the

Figure 2: A typical manufacturing-service network
customer would result in higher sales, whereas a lower

utility would result in lower sales. The sold goods enter the
The manufacturer designs the product, produces it and
population of deployed equipment while excess goods are
subsequently markets and sells it to a pool of customers.
inventoried for later sale. As the deployed product is used,
With time and usage the performance of the product
over time its performance deteriorates. Eventually when it
3006

---

fails, it comes in for repairs at a service center. At the
in service capacity such that service deterioration is
service center the product is repaired, within the service
minimized. Information visibility and collaboration that is
capacity limits of the service center, by employing the
pervasive between the manufacturing, sales and service
right mix of manpower, spare parts and repair tools.
operations of an IMSN is well suited to support a
As the population of goods sold increases, the number
feedforward based control for service capacity
of products coming in for repair also increases. Hence in
management.
order to maintain service quality and service times, the
In our model, we have assumed the lead time for
capacity of the repair centers need to be increased, or
service capacity addition, to be 120 days and accordingly
synchronized with the population growth, so that they can
every 120 days the population growth is forecasted and the
handle the higher inflow of goods coming in for repairs.
expected daily repair events or equivalently the additional
This phenomenon is represented in Loop 1 in Figure 3,
required service capacity is determined as follows:
wherein it may be noticed that as the population increases,
Capacity ordered every 120 days = Population growth in
the number of products requiring repair increases, in turn
120 days * Failure rate
leading to backlog of repair jobs, lower service levels and

consequentially lower product sales. This drop in sales
Similarly, we assume that excess capacity is removed from
needs to be controlled by appropriately controlling the
the system based on the following relationship:
service capacity, which has a direct positive influence on
Amount of capacity freed = Current Capacity - 1.2*(Job
the service levels and an indirect influence on the product
arrival rate) – Backlog – Capacity to support expected
sales. To address this problem, we explore two alternative
population growth
mechanisms, feedback and feedforward, to manage service

center capacity as described below, and represented by the
The cause-effect diagram for an IMSN, as shown in Figure
dotted lines in the cause-effect diagram in Figure 3.
3, was developed as a system dynamics model in the
a. Feedback system: The first mechanism is a feedback
IThink software and simulated to observe the long-run
mechanism wherein performance degradation at the behaviour of IMSNs. For our analysis we assumed that the
service centers leads to a corrective addition of service
IMSN was in the automobile industry with the following
capacity that reduces the backlog of repair jobs. This
parameters and assumptions.
mechanism is represented in Loop 2 in Figure 3. It may be
1. A production rate of 1000 units/day.
noticed that whenever jobs get accumulated for repair, the
2. Car life of 10yrs or 3650 days.
service capacity augmentation process is initiated and
3. Service level being the sole factor influencing sales
consequentially new capacity is added to the service center
based on the following relationship.
after a fixed lead time, which corresponds to the time
Product Sales = (Mean_Service_level - 0.7)*10000/3.
required to train repairmen and procure required spares
4. Mean number of failure per day = 0.01*Population
and tools. Such a mechanism is reactive in nature and most
5. Service capacity acquisition lead-time of 120 days
service centers typically operate in this fashion when they
attributable to delays in hiring and training.
have no information about product sales. The capacity
6. Product life-cycles 10 years or longer.
requisition in our feedback model is based on the 7. A single repair job completed in one day if one unit of
following relation:
capacity dedicated to the job.
Capacity ordered on any day = (Rate of arrival of new
8. Service level for a particular day defined as the ratio
jobs)*1.1 + (Backlog of jobs) / 5 – Existing Capacity –
of repairs completed in the day to the total number of
Capacity in the process of procurement
incoming repair jobs in the day plus previous backlogs.

Since the number of repairs completed is at most
Similarly, when it is noticed that the service center has
equal to the capacity, Service level = Min
excess capacity it is eliminated. In our model we assume
(Capacity/Total number of failures, 1).
that the service center is monitored daily for excess
Under the above assumptions, the ideal service center
capacity, which is removed based on the following capacity is equivalent to the mean number of failures
equation:
which is one percent of the current population in our
Amount of capacity freed = Current Capacity - 1.2*(Job
model. Our analysis proceeds by a simulation of both the
arrival rate) - Backlog
feedback and feedforward strategies outlined earlier.

4.1 Bullwhip effect in Manufacturing Service Networks
b. Feedforward system: The second mechanism that

An IMSN based on a feedback mechanism responds to
may be used to manage service levels is one based on the
the accumulation of jobs by increasing the service center
concept of feedforward. In a feedforward system sales data
capacity. However since there is an inherent delay in
would be utilized to firstly forecast the population growth
acquiring and deploying service capacity, the service
of deployed products, secondly to predict the number of
capacity addition is not immediate. Consequentially the
future failure events and ultimately to calculate the service
service levels continue to decline and jobs continue to
capacity requirements to support the expected repair needs
accumulate in the interim period. Hence, to rapidly clear
of the forecasted population. Hence, taking into account
the backlog of jobs and to bring the service up to the
the lead-times for service capacity addition, in a desired levels, sometimes more capacity than required
feedforward-based system, we can ensure timely increase
needs to be added. Thus the reactive feedback system
3007

---

results in a bull whip effect: a sudden increase in product
information sharing and collaboration between the
sales after a lag creates a need for excess service capacity.
manufacturing and service networks. However, our
Because of the leadtimes involved in service capacity
analysis above was based on a very simplistic black-box
addition, the service centers are unable to maintain their
representation of a service center based on the assumption
service levels and accordingly the sales decline as shown
that the service center is characterized solely by its
in Figure 4. The fluctuations in service capacity observed
capacity and that it utilizes a single resource type.
in the system (as shown in Figure 4) can be prevented by
promoting collaboration between the manufacturing, sales
and the service operations of the IMSN.
Backlog of jobs
Capacity


Figure 4: Service capacity buildup to clear accumulated jobs

Service level
for system 2
Population for system 2

Figure 6: Service Center work flow
Service level

for system 1
A more realistic representation of a service center would
model the progression of a repair job through various
Population for system 1
sequential stages such as diagnosis, repair and testing and
would highlight the interaction between the utilization of
various resource types such as manpower, tools and spare

parts at each stage. In this section, we attempt to develop
Figure 5: System with Feed-forward and Feedback maintain higher
service level than system with only feedback mechanism
just such a model for a generic service centre and simulate

the operation of the service centre to illustrate the bullwhip
In addition the performance of an IMSN based only on
effect witnessed in the capacity and resource levels of the
feedback was compared to that of an IMSN with both
service centre. We also establish that information sharing
feedback and feedforward mechanisms, as shown in Figure
across the various operations of the service centre can
5. It may be noticed that the feed-forward system is able to
mitigate the effect.
maintain very high service levels while the feedback
For our analysis we consider a generic repair centre,
system shows fluctuations. This is because the feed-
where jobs arrive at the steady rate of 10 jobs per unit of
forward system is better able to forecast service capacity
time. Each job is processed through a sequential step of
requirements ahead of time and can prepare for it better.
activities beginning with diagnosis and followed by repair,
On the other hand, an IMSN based on a feedback testing and a final touchup as shown in Figure 6. The
mechanism shows considerable deviations in the service
resources required for the completion of each activity are
level over time, which in turn effects the population
different. For example, diagnosis requires diagnostic
growth. From Figure 5, it may be noticed that due to
equipment whereas the repair process requires technicians,
consistently high service levels for an IMSN with equipment, and spare parts.
feedforward, as compared to an IMSN with feedback, the
The service center was simulated using a system
sales of products and consequentially the deployed dynamics framework and the initial value for each of the
population for an IMSN with feedforward will be higher
resources was assumed to be zero. As the job starts
over time.
arriving the resources are hired to match the demand. It
V. MODELING SERVICE CENTERS
was observed that the variability in resources increases
with each successive stage as shown in Figure 7. This is

From the analysis above, it may be observed that
similar to the bullwhip effect observed in the supply
service quality management is greatly enhanced by chains.Thus we see that resource management is a
3008

---

challenging problem in service centres and even a small
deterioration. The influence diagram for an IMSN
change in the rate of job arrivals can have drastic effects
manufacturing and servicing a single product under the
on resource requirements and their levels, which get
above operating assumptions is presented in Figure 8.
magnified down the line as well. This further reinforces

the importance of information exchange in better
managing service networks.
Crew
Testers
Mechanics
Diagnostics
Resource


Figure 8: Influence diagram for IMSN with single product
Figure 7: Bullwhip effect in the resources of service center

VI. MODELING IMSNS WITH A SINGLE PRODUCT

The profits of the IMSN are derived from the sales of
both products and services, less the cost of production,
The basic model presented in Section 4 is simplistic in
service provision and cost of providing free services
its assumption that the consumer's decision to buy a
during warranty. In addition, we assume that a fixed
product is based primarily on the quality of the service.
percentage of the profits are continually invested in R&D.
However, in reality the purchase decision is influenced by
These investments bear fruit at a later time in the form of
multiple factors such as the price and quality of both the
improved product quality, reliability and reduced product
product and the service and the quality and length of the
prices, in turn stimulating further sales and profits.
warranty period. In this section we attempt to incorporate
Concurrently, as the product quality improves, the
these features to build a more realistic model of an IMSN
reliability of the product increases and the number of
that manufactures and services a single product.
failure events and consequentially repairs is reduced. In
For our analysis we assume that the utility of a product
this manner, the concept of design for serviceability is
to a consumer is a linear function of product price and
included in our model of an IMSN. However for the sake
quality, service price and quality and the length of the
of simplicity, we have assumed aggregated models for
warranty period.
service centres, each of which are geographically

Utility = -Product price + 1000*Product quality - Service
distributed. One key differentiating factor in the extended
price +1000*Service level + Free service contract.
model presented in this section, as compared to our earlier
Product price and service price have a negative analysis, is the assumption that the product failure rate
influence on the utility function since higher prices deter
increases with the age of the product. With an increase in
customers from buying the product. Similarly, free service
product sales over time the population grows at a higher
contract periods, better product and service qualities, have
rate, resulting in greater number of repair jobs. Also, as a
a positive influence on the utility as perceived by product gets old it comes more often for repair. Hence, to
customers. Furthermore, we also incorporate a reference
predict the load on the service centre we need to track both
utility function, similar to the utility described above,
product sales and the age of the installed base.
which defines a target utility level for the company to
Additionally, we incorporate the feedforward and
aspire towards. In reality this reference utility is equivalent
feedback mechanisms, explained in the earlier sections, for
to the utility of the products offered by the competitors. In
controlling the interaction between the manufacturing and
this manner we are able to account for competitive service sectors and for ensuring that service levels are
pressures facing an IMSN. As can be expected, the better
maintained at the required level. For example, in a
the product offered by the IMSN, relative to the feedback based system, if the service level drops below the
competition, higher its sales. Hence, the number of sales
reference service level, service capacity is added. On the
that take place in a day is directly proportional to the
other hand, in a feedforward system, the sales data is
difference between the reference utility (utility of employed to forecast and plan for additional capacity
competitive offerings) and utility of the product offered by
requirement.
the IMSN. If a product service bundle offers superior value
The model was simulated in IThink to demonstrate
the demand for the product will expectedly increase.
that the feed-forward system is better than the feedback
However increased sales and growth in the product system in terms of its ability to improve service levels and
population will over time result in more service demands
the bottom-line for the IMSN. For the simulation study, it
putting a strain on service levels and quality, requiring
was assumed that the lead time for service capacity
some kind of monitoring process to remedy the service
addition was 120 days and the initial capacity was set to be
3009

---

adequate to meet all demands occurring in the first 120
feedback is eventually eliminated as the backlog of service
days of operations. Based on growth of the population and
jobs is cleared. The models for IMSNs with both feedback
age of installed base a forecast is made for number of
and feedforward were simulated over the entire lifecycle of
failures per day after 240 days and the required extra
the product and their profits compared. The profits in an
capacity is added as appropriate. From the influence
IMSN with both feedback and feedforward mechanisms
diagram it may be noticed that the dynamics within this
are 50% higher than the profits for an IMSN with only
extended and realistic model are more intricate than the
feedback.
basic model presented in Section 4 (Figure 3), establishing
the fact that it is harder to predict the behavior of realistic
IMSN models. Our analysis was based on the following
parameters and assumptions.
1. One unit of service capacity costs 100 units of money
per period.
2. The price charged for service dependent on the service
level offered as dictated by the following relationship.
Service Price = Service capacity cost*(1 + service
level).
3. Profit margin of 10% incorporated in the price of
products

4. Product quality & Manufacturing cost improves non-
Figure 10: Comparison of service capacity for system with only feedback
linearly for incremental investments in R&D, meaning
and system with feedback and feedforward mechanisms
that the ROI for R& D investments is more than
100%.
VII. CONCLUSION
5. The probability of a product failing increases over
time.

In this paper, we have studied the integration of
6. The product becomes obsolete in 6 years, in a non-
manufacturing and service networks through information
linear manner, as it rapidly becomes outdated in its
sharing and collaboration and highlighted the behavior of
first few days before being gradually replaced over
such systems. While there are several studies in the form
time.
of business cases and white papers on the benefits of such
integration, ours is the first systematic study providing a
framework for the analysis and design of integrated
manufacturing systems and service operations planning in
support of strategic manufacturing and sales objectives.
We plan to develop queueing network and Petri-net
models to further study and develop analytical models for
IMSNs.
REFERENCES
[1] Cohen, M., Whang S., “Competing in product and service: A product

life-cycle model”, Management Science, Vol. 43, No. 4, April 1997.
Figure 9: Comparison of service levels for an IMSN with feedback and
[2] Cohen, M., Y Zheng, and V Agarwal, “Service Parts Logistics: A
one with both feedforward and feedback mechanisms
benchmark analysis”, IIE Transactions, Special issue on supply chain

management, 1997.
The service levels for IMSNs with only feed-back and
[3] John Zysman, “Production in the Digital Era: Commodity or
Strategic Weapon?”, Berkeley Roundtable on the International
with both feed-back and feedforward were simulated and
Economy, 2002.
compared as shown in Figure 9. Clearly an IMSN with
[4] Juliet E. Johansson, Chandru Krishnamurthy, and Henry E.
feedforward and feedback mechanisms is able to provide
Schlissberg, “Solving the solutions problem”, Mckinsey quarterly,
2003, No. 3.
consistent service levels, inspite of the dynamics involved
[5] U Dinesh Kumar, John Crocker, J. Knezevic, M El-Haram,
in the system. It may be observed from Figure 10 that the
“Reliability, maintenance and logistic support – A life cycle
service capacity growth in an IMSN with a feedforward
approach”, Kluwer Academic Publishers, 2000.
system is more planned in comparison to an IMSN with
just a feedback system. Since one unit of service capacity
is needed to complete a single repair job in a day, the
service capacity over time in a well synchronized
manufacturing service network, such as one with a
feedforward system, will trace the number of incoming
repairs jobs. However, due to the long lead-times in
service capacity acquisition the service capacity increases
only in a stepwise manner. It may also be noticed that the
excess service capacity added in an IMSN with only
3010

---

Document Outline

• MAIN MENU

---