The Emerald Research Register for this journal is available at The current issue and full text archive

xt archive of this journal is available at

http://www.emeraldinsight.com/researchregister http://www.emeraldinsight.com/0960-0035.htm

IJPDLM
33,4 The impact of increasing
demand visibility on
336
production and inventory
Received January 2003
control efficiency
Revised March 2003 Johanna Småros, Juha-Matti Lehtonen, Patrik Appelqvist and
Accepted March 2003
Jan Holmström
Department of Industrial Engineering and Management,
Helsinki University of Technology, Helsinki, Finland

Keywords Supply-chain management, Demand forecasting, Information exchange,

Production planning, Inventory control
Abstract Information sharing practices such as vendor-managed inventory (VMI) give
manufacturers access to more accurate demand information, e.g. customer sales data, than before.
The value of this type of information sharing has been established in many studies. However, most
of the research has focused on the ideal situation of the manufacturer having access to information
from all downstream parties. In practice, this is rarely the case. In this paper, discrete-event
simulation is used to examine how a manufacturer can combine traditional order data available
from non-VMI customers with sales data available from VMI customers in its production and
inventory control and what impact this has on the manufacturer’s operational efficiency. The
simulation model is based on a real-life VMI implementation and uses actual demand and product
data. The key finding is that even for products with stable demand a partial improvement of
demand visibility can improve production and inventory control efficiency, but that the value of
visibility greatly depends on the target products’ replenishment frequencies and the production
planning cycle employed by the manufacturer.

International Journal of Physical
Distribution & Logistics Management
Vol. 33 No. 4, 2003
pp. 336-354
q MCB UP Limited
0960-0035
DOI 10.1108/09600030310478801
Introduction
Lack of demand visibility has been identified as an important challenge for
supply chain management (Chen, 1998; Lee, 2002; Lin et al., 2002). Commonly,
the only factual demand information companies have access to, are the orders
placed by their customers (Cachon and Fisher, 2000). As both practice and
research has shown, order information often gives a delayed and distorted
picture of end customer demand and what actually happens in the market. The
distortion tends to increase upstream in the supply chain, making demand look
variable and unpredictable even when end customer demand is level.
Controlling production and inventories based on this flawed demand
information easily leads to inefficient capacity utilization, poor product
availability, and high stock levels. (Forrester, 1961; Houlihan, 1987; Burbidge,
1989; Towill et al., 1992; Lee et al., 1997a).
To remedy this problem and to obtain a smoother material flow, companies
have started to develop replenishment methods that operate without orders.
One of the most common types of these automatic replenishment programs is Increasing
vendor-managed inventory or VMI (Daugherty et al., 1999). In VMI, the vendor demand
is given access to its customer’s inventory and demand information. The visibility
vendor monitors the customer’s inventory level and has the authority and the
responsibility to replenish the customer’s stock according to jointly agreed
inventory control principles and objectives (Cachon and Fisher, 1997; Waller
et al., 1999; Kaipia et al., 2002). VMI offers the vendor access to its customer’s
337
sales information, sometimes called sell-through information, rather than its
orders. This means that one level of order batching is removed, allowing for
more accurate, more rapidly available, and more level demand information. In
addition, since the vendor is free to choose the timing of the replenishment
shipments, it can further dampen demand peaks, for example, by delaying non-
critical replenishments (Kaipia et al., 2002).
Daugherty et al. (1999) examined the adoption and performance of automatic
replenishment programs through a survey of US manufacturers and retailers.
The results, although based on a small sample, indicated fairly widespread use,
or plans for future use, of automatic replenishment programs in general, and
VMI in particular. Furthermore, the study found a positive relationship
between automatic replenishment programs and company performance.
Several case studies indicate that the benefits of VMI in the area of
production planning and inventory control can be significant. Kaipia et al.
(2002) have demonstrated that implementing VMI can enable substantial
inventory reductions as well as an opportunity to shift from make-to-stock to
make-to-order production. Success stories from the industry demonstrate the
potential of VMI in practice; companies have reported inventory reductions,
improved customer service, and reduced obsolescence as the results of VMI
adoption (Fraza, 1998; Holmström, 1998; Kaipia et al., 2002).
Still, many companies are struggling to make VMI work for them. Vergin
and Barr (1999) studied ten Fortune 500 consumer products manufacturing
companies involved in VMI. They concluded that although the companies’ VMI
customers had benefited from improved availability and lower stock levels,
only two of the manufacturing companies had been able to realize
improvements in their management of production and only one had achieved
lower internal inventories. Cooke (1998) and Lapide (2001) report similar
findings: although some companies promote VMI, many are retreating from the
concept, and especially manufacturing companies are skeptical about the
benefits of VMI.
A major challenge for manufacturing companies is that usually, only part of
their customer base is involved in VMI. This means that the manufacturers
need to set up their operations in a way that both VMI and non-VMI customers
simultaneously can be efficiently served. However, research has mainly
focused on the ideal situation of the manufacturer having access to sell-through
data from all of its downstream partners, or the relationship between only one
IJPDLM vendor and one customer. Hence it has provided only little support for
33,4 companies struggling with limited visibility and a heterogeneous customer
base.
In this paper, we study how manufacturing companies can benefit even from
a partial increase in demand visibility. We look at a situation where a
manufacturer loads its production with a combination of order data from non-
338 VMI customers and sell-through data from a varying number of VMI
customers. Using simulation, we examine how the manufacturer’s benefit,
measured as reduced variability of its production load, increases as the number
of VMI customers increases. We also demonstrate how the benefit of VMI is
dependent on certain product demand characteristics and on the production
planning cycle employed by the manufacturer.

Literature review
Supply chain visibility
In his seminal work on industrial dynamics, Forrester (1961) first demonstrated
the potentially devastating phenomenon of variability amplification along the
supply chain. He showed that amplification is caused by system structure,
delays in decisions and actions, as well as inventory control policies. Later on,
other researchers such as Houlihan (1987), Towill (1991), and Lee et al. (1997a)
have further developed the theory of industrial dynamics.
During the last few years, following the advent of enabling technologies, the
role of information sharing as a means of reducing bullwhip in the supply chain
has received great interest. Forrester (1961, p. 177) himself touched on this
issue, only implicitly, by demonstrating how amplification could be reduced by
removing the distributor echelon in the simulations.
Based on a thorough analysis of the bullwhip effect, Lee et al. (1997b)
suggested that “one remedy [to the bullwhip problem] is to make demand data
at a downstream site available to the upstream site”. Later, Lee et al. (2000)
validated this finding using mathematical modeling. Several other researchers
have also examined this claim. Chen et al. (2000) used mathematical modeling
to examine the effect of information sharing, forecasting and lead times on the
bullwhip effect, and concluded that although access to customer demand
information does not completely remove the problem of variability
amplification it can significantly reduce it. Yu et al. (2000) used a case
example to illustrate the benefits of different levels of information sharing and
concluded that a VMI approach brings remarkable benefits to the vendor, and
profits the customer as well.
To summarize, several researchers present the same conclusion: information
sharing in the supply chain is valuable (Chen et al., 2000; Lee et al., 2000; Yu
et al., 2000). Many also present the same observation of upstream parties
benefiting more from access to downstream demand data than the downstream
parties sharing the information (Lee et al., 2000; Yu et al., 2000). Interestingly,
this latter research finding is in striking contrast with the observations of, for Increasing
example, Lapide (2001), Cooke (1998) and Vergin and Barr (1999) who describe demand
the difficulties of manufacturing companies in benefiting from VMI and how visibility
the interest in VMI has decreased among manufacturers. Cachon and Fisher
(2000) have also, based on a simulation study, concluded that for stationary
demand the benefits of more frequent ordering and timely processing outweigh
the benefits of information sharing.
339
Lapide (2001) suggests that the main reason why manufacturing companies
have failed to benefit from VMI is that they have only implemented the
execution part of VMI, i.e. the sales and distribution transactions. He claims
that the companies have not managed to link the demand information, i.e. the
customer sell-through information available through VMI to their production
planning and inventory control systems. This corresponds with our own
experience; most of the VMI implementations the authors have seen have
lacked this link to supply chain planning. Consequently, one can conclude that
linking demand information to supply chain planning seems to be of critical
importance to benefiting from visibility efforts such as VMI.

Supply chain planning systems

Clark and Scarf (1960) were among the first to study the problem of
replenishing a complex supply chain in a reliable, cost-efficient way. They
developed a model for calculating optimal stock levels and re-order quantities
in a two-echelon supply chain. The work of Clark and Scarf (1960) was further
developed by companies creating models that enabled dynamic and continuous
supply chain optimization. Building on industrial work (e.g. Stenger and
Cavinato, 1979), Bookbinder and Heath (1988) developed the concept of
distribution requirements planning (DRP). The idea of DRP was to minimize
inventory and cost in the distribution system for a certain service level and
demand forecast through periodic planning of inventory levels and
replenishments.
Today, the term advanced planning and scheduling is typically used for
software products in the distribution and supply chain area. DeKok and
Fransoo (2002) and Fransoo et al. (2001) have analyzed these supply chain
planning solutions and concluded that the development focus has been on
improved optimization – virtually to the exclusion of practical implementation
challenges. This focus on optimization can be considered problematic.
Optimization algorithms typically require accurate and complete data to
function correctly. In practice, manufacturing companies can seldom achieve
this state of affairs more often than once a month or once a week (Euwe and
Wortmann, 1997). It is also difficult to include customer or supplier information
in the plans since the data available is usually not accurate enough. In addition,
even if accurate information is available from some partners, it is usually not
available from all, which means that the completeness requirement is not met.
IJPDLM This is also the typical situation in VMI. Generally, only part of the
33,4 manufacturer’s customer base is involved in VMI and increasing the adoption
rate takes time, making it difficult to utilize the additional demand visibility
available from VMI partners.
In practice, it is our experience that many companies have chosen to link
VMI to their production planning and inventory control systems in a way that
340 emulates order-based operation. Often, when a VMI replenishment is triggered,
this leads to an order being generated in the vendor’s system. The vendor’s
planning system then treats this “VMI order” as any other order and uses it to
load production and update forecasts. The customer sell-through data is, thus,
neglected in order to make all customers look alike from a system point of view.
A question that arises is whether there are opportunities to implement VMI in a
way that the more accurate sell-through data, available from VMI-customers,
can be efficiently utilized?

Benefiting from demand visibility in production and inventory control

Some general ideas on how manufacturers can benefit from increased demand
visibility in their production planning and inventory control have been
presented. Silver et al. (1998) present the idea of basing production planning
and inventory control on sell-through data attained from customers, but does
not discuss the practical implementation. Lee et al. (2000) develop an analytical
model to study the effect of information sharing in a two-level supply chain.
The model quantifies the inventory and cost levels when information on non-
stationary demand is shared openly between the retailer and manufacturers in
comparison to the situation where only orders are placed. Disney and Towill
(2001) present a similar idea: controlling the manufacturer’s production and
inventory using the customer’s sell-through data, instead of using order or
delivery data.
However, most of the visibility research only considers the ideal situation of
demand data being readily available from all downstream partners, or focuses
on relationships involving only one vendor and one customer. There is little
research discussing the problems related to dealing with a mix of VMI and non-
VMI customers. In the rare cases where partial information availability is
discussed, the link to production planning and inventory control is still not
explicitly examined. Waller et al. (1999), for example, use simulation to examine
the effect of VMI adoption rates on inventory levels in a supply chain. The core
of their VMI model is, however, increased inventory review and replenishment
triggering frequency. No production and inventory control solution that would
utilize the demand information available from the VMI customers is presented.

Research focus
In this paper, we build on the research and visibility models presented by Lee
et al. (2000), Disney and Towill (2001) and Silver et al. (1998). We expand their
ideas by including several customers – both VMI and non-VMI – in the supply
chain model, and by employing a simple production planning and inventory Increasing
control scheme that allows the manufacturer to use VMI customers’ sell- demand
through information as well as traditional orders from non-VMI customers in visibility
parallel. Using simulation, we then examine how the level of visibility, i.e. the
number of VMI customers, affects the manufacturer’s production efficiency. In
addition, we analyze how products’ different demand characteristics and the
manufacturer’s production planning cycle length affect the benefits of
341
increased visibility.

Methodology and simulation model design

Based on the identified gap in the literature, the following research problem
was formulated. What is the effect of increased visibility of customer sell-
through data on the performance of a manufacturer’s production and inventory
control?
To study the impact of increased visibility we chose to use discrete-event
simulation. This is in accordance with Maloni and Benton’s (1997)
recommendation to use simulation models as a way to critically evaluate the
benefits of supply chain partnerships. A simulation model provides a
convenient lab environment for testing the effects of different factors. In
addition, simulation facilitates the involvement of experts from industry as
both the model and the results are usually reasonably easy to understand.
To be able to construct a simulation model that was as realistic as possible
we worked closely with a fast moving consumer goods manufacturer involved
in VMI. The company currently has a VMI agreement with one of its largest
customers, but also serves several non-VMI customers. The VMI
implementation has already proven valuable both to the manufacturer and
the distributor. It has, among other things, reduced the distributor’s workload
and improved communication between the companies. Still, the manufacturer
believes that additional benefits could be attained if the demand information
available through VMI could be used more effectively in its production and
inventory control processes.
The supply chain model used in the simulation runs was designed to model
the situation of the case manufacturer in one of its primary markets. Actual
sell-through data available from the VMI implementation, as well as the
examined products’ actual minimum replenishment quantities, were used in the
simulation runs. In addition, after the simulation runs had been completed, the
results were reviewed with the manufacturer.

The simulation model

The supply chain model used in the simulation runs consists of one
manufacturer serving three distributors, which in turn serve several retail
outlets. Although Figure 1 illustrates a situation where one of the distributors
IJPDLM
33,4

342

Figure 1.
The supply chain model
used in the simulation
runs

is involved in VMI and the two others employ traditional orders, all VMI
adoption rates – zero, one, two or three VMI distributors – are examined.
The retail outlets in the simulation model order products from the
distributors daily. VMI is not employed between the distributors and the retail
outlets, although this could easily be incorporated in the model. The reason
why orders are used is that this is how the case supply chain currently
operates. The real-life demand data used in the model, therefore, represent a
situation in which retail outlets place orders. However, as the focus of the
simulation study is on the manufacturer’s operational performance and how it
changes with the adoption rate of VMI among the distributors, the
replenishment model employed downstream in the supply chain does not
affect the results as long as it is consequently the same throughout the study,
as it is here.
In the simulation runs, actual sell-through data from the case
manufacturer’s VMI implementation is used to model the retail outlets’
orders. From a 60-week period of daily sell-through data from the VMI
distributor, 21 products having stable demand and belonging to one
product category were selected. Seasonal products and products introduced
or discontinued during the observed period were discarded. To model
multiple distributors, the original sell-through data was divided into three
20-week sections, which were then used as the demand for the three
distributors in the simulation model. The resulting demand pattern is
reasonably realistic as the VMI distributor stands for approximately one-
third of the case manufacturer’s total sales and the ordering patterns of the
different retail outlets can be assumed to be quite similar.
The distributors fulfill the retail outlets’ orders and manage their own
inventories using a re-order point control mechanism. For non-VMI distributors
this means that they place orders when a product’s inventory level has dropped
below a specified re-order point. For VMI distributors, the manufacturer Increasing
monitors the distributors’ inventories and generates a replenishment request demand
whenever a re-order point is reached. For both VMI and non-VMI customers, visibility
the re-order points are determined using statistical independent inventory
control formulas with a normal distribution assumption (Vollman et al., 1997,
p. 712). In reality, the inventory control parameters would be regularly updated
and include forecast information. However, since only products with stable
343
demand were selected for examination, both parameter updates and forecasts
could be omitted. In addition, in our model, the re-order points and inventory
control logic of the distributors are unaffected by their status as VMI or non-
VMI distributors, although, in reality, the inventory control parameters in VMI
would be determined by the manufacturer and the VMI distributors co-
operatively. The main difference between VMI and non-VMI distributors in our
model is, thus, the degree of demand visibility. Whereas the manufacturer has
access to replenishment requirements as well as sell-through data (i.e. the
distributors’ deliveries to the retail outlets) from the VMI distributors, it only
has access to order information from non-VMI distributors.
The manufacturer ships products to the distributors on a daily basis
according to purchase orders and replenishment requirements. For production
scheduling the manufacturer uses the period batch control for standard
products presented by Burbidge (1994). A weekly, bi-weekly or monthly
planning and production cycle is employed. The manufacturer loads its
production with a combination of the VMI distributors’ sell-through data from
the previous period as well as the non-VMI distributors’ orders. Manufacturing
is not capacity constrained. At the beginning of each simulation run, the
manufacturer’s safety stock is initialized to be high enough so that all orders
and replenishment requirements can be fulfilled.
The simulation model was created using the AutoMod 9.0 simulation
software. The detailed logic of the simulation model is presented in the
Appendix (see Figure A1).

Variables
The main relationship we set out to examine was between the level of visibility,
i.e. access to distributor sell-through data, and the bullwhip or demand
variability amplification experienced by the manufacturer. We also looked at
two control variables in order to examine how the main external factor, i.e. the
retail outlets’ demand fed into the model, and the main internal factor, i.e. the
production planning frequency used in the model, affected the results of the
simulation runs.
The independent variable, level of visibility, was studied by moving from
none to one, two, and all three distributors being involved in VMI. The demand
characteristics control variable was examined by looking at 21 different
products with different demand volumes and minimum replenishment
IJPDLM quantities. The other control variable, production planning frequency, was
33,4 examined by looking at three different scenarios: weekly, bi-weekly and
monthly planning and production cycles.
The dependent variable, bullwhip, was measured as the relative or absolute
standard deviation of the manufacturer’s production load.
Bullwhip is defined by Fransoo and Wouters (2000) as:
344
cout
Bullwhip ¼ ; ð1Þ
cin
where cout and cin represent the relative standard deviation of demand
measured upstream in the supply chain (cout) and downstream in the supply
chain (cin), respectively. The relative standard deviation c is defined as the
standard deviation (s) of demand (D) divided with the demand mean (m) for a
specified time interval ½t; t þ T
:
c ¼ sðDðt; t þ TÞÞ
: ð2Þ
mðDðt; t þ TÞÞ
In this article, cin refers to the relative standard deviation of demand
experienced by the distributors and cout to the relative standard deviation of
demand experienced by the manufacturer. Since production is unconstrained
and loaded directly with the demand experienced by the manufacturer (VMI
distributors’ sell-through data and non-VMI customers’ orders), the relative
standard deviation of the production load is equal to cout. In addition, since the
same customer demand is used in all simulation runs, cin does not change
between runs and does not need to be discussed when presenting the results.
Therefore, from here on, we will use the production load standard deviation,
either relative or absolute, to express the different factors’ impact on the
manufacturer’s production and inventory control efficiency. Reducing
production load standard deviation is key to achieving lower inventories,
improving delivery accuracy and improving capacity utilization.

Simulation results
In order to study all factor combinations, a total of 252 simulation runs were
conducted (see Figure 2). First, the impact of the distributors’ VMI adoption
rate on the manufacturer’s production efficiency was examined. This was done
by increasing the number of VMI distributors one by one, and measuring the

Figure 2.
Simulation runs
impact on the manufacturer’s production load variability. Second, 21 products Increasing
with different replenishment frequencies were studied in order to establish how demand
a product’s demand characteristics, i.e. demand volume and minimum visibility
replenishment quantity, affect the value of increased visibility. Third, three
different production planning cycle lengths – one week, two weeks and four
weeks – and their relationship to the value of increased demand visibility were
examined.
345

Effect of increasing visibility

The primary objective of the research was to examine how increased demand
visibility, i.e. increased access to distributor sell-through data, affects the
manufacturer’s production efficiency. This was examined by increasing the
number of VMI distributors one by one and measuring the effect on the
manufacturer’s production load variability. Although the simulations were
conducted for three different production planning cycle lengths, only the
results attained using the shortest planning cycle will be discussed here. We
will return to the impact of planning cycle length on the value of visibility in a
later paragraph.
The first change from no VMI distributors to one VMI distributor, as can be
seen in Table I, resulted in a moderate average reduction of production load
variability. The average variability reduction was 14 percent, but there were
noticeable differences between the individual products. For some products, the
reduction was not important, only a few per cent, whereas the maximum
reductions were over 20 percent.
Next, the number of VMI distributors was increased to two, resulting in two-
thirds of the manufacturer’s production requirements being based on
distributor sell-through data and one-third on traditional orders. The effect
on production load variability was noticeable. Compared to the base case of no
VMI distributors, the reduction in production standard deviation was on
average 28 percent for the products examined.
The full visibility scenario, i.e. all three distributors involved in VMI, further
reduced variability. Compared to the base case of no VMI distributors, full
visibility reduced production load standard deviation with 41 percent on average.
The results were, once again, different for different products. The smallest
reduction was, however, over 20 percent and the largest as much as 65 percent.
Figure 3 illustrates what happens when the number of VMI distributors
increases. As the number of VMI distributors increases, the manufacturer can
replace more and more order data with sell-through data in its production
planning and inventory management. This means that the distorting effect of
distributor order batching is removed, leveling out the highest peaks and
lowest valleys in demand. Of course, due to the natural variation in demand as
well as the order batching of the retail outlets, some variation still remains in
the data even in the situation that all distributors are involved in VMI.
IJPDLM
Change in bullwhip
33,4 1/3 VMI (%) 2/3 VMI (%) All VMI (%)

Product 20 2 33 245 2 44
Product 11 2 24 237 2 40
Product 17 2 23 237 2 45
346 Product 5 2 23 234 2 54
Product 7 2 18 234 2 48
Product 15 2 18 220 2 40
Product 19 2 17 232 2 46
Product 4 2 16 226 2 44
Product 12 2 16 231 2 37
Product 9 2 16 229 2 65
Product 3 2 13 218 2 44
Product 2 2 13 211 2 31
Product 13 2 13 239 2 49
Product 21 2 11 227 2 43
Table I. Product 10 28 220 2 24
Change in weekly Product 14 26 228 2 24
standard deviation Product 18 25 213 2 39
of production load Product 8 25 228 2 29
(i.e. bullwhip Product 16 25 225 2 40
experienced by Product 6 24 226 2 45
manufacturer) Product 1 22 220 2 30
compared to the Average change 2 14 228 2 41
base case of no VMI Minimum change 22 211 2 24
distributors Maximum change 2 33 245 2 65

Figure 3.
Example product’s
production load at
different levels of
visibility
Effect of product demand characteristics Increasing
When assessing the effect of increased visibility on the manufacturer’s production demand
load variability, considerable differences between the products could be observed. visibility
Whereas the reduction in variability for some products was remarkable, up to 65
percent, for other products the value of information sharing seemed much smaller.
In order to examine these differences a bit closer, we focused on the two
characteristics differentiating the products: the products’ average demand and
347
their minimum replenishment quantities. By plotting the change in production
load variability against a combination of these characteristics – the average
number of replenishments per week – a pattern could be detected. This pattern
revealed that the impact of increasing demand visibility was, in general,
greater for products with low replenishment frequencies, i.e. typically
C-products, than for products with high replenishment frequencies, i.e.
typically A-products. This is illustrated graphically in Figure 4 for the change
from no visibility to full visibility in the model supply chain. The
replenishment frequency and the relative reduction in production load
variability in Figure 4 are also statistically correlated (R2 ¼ 0:62, p , 0:001).
The most important benefit of distributor sell-through data when compared
to traditional orders is that one level of order batching is removed. This means
that the impact of access to distributor sell-through data is typically greater for
products with significant order batching, i.e. products whose minimum
replenishment batches are large in proportion to their average daily or weekly
demand. For these products, access to distributor sell-through data
significantly speeds up the information flow within the supply chain. For

Figure 4.
Impact of increased
visibility on products
with different
replenishment
frequencies
IJPDLM products with high replenishment frequencies, the change is less dramatic. As
33,4 can be seen in Table II, an increase in demand visibility reduces the production
load variability for all products. However, the change is larger for the products
with low replenishment frequencies than those with high replenishment
frequencies. When the proportion of VMI distributors is high, the production
loads of all products behave similarly.
348
Effect of planning frequency
The third factor that we studied was the impact of the manufacturer’s
production planning frequency on the value of increased demand visibility.
Although some companies employ a weekly planning cycle or, in some extreme
cases, a daily planning cycle, many companies still use a longer, monthly
planning cycle that enables them to optimize production capacity utilization.
The relationship between planning cycle length and value of increased
visibility is, therefore, of great practical interest.
In order to examine this relationship, simulation runs with three different
planning cycle lengths were conducted. Flexible production employing a
weekly planning cycle was contrasted with less flexible production using
planning cycles of two weeks and one month (see Table III).

Relative standard deviation of production load

No VMI 1/3 VMI 2/3 VMI All VMI
Replenishments/week (%) (%) (%) (%)

Product 14 1,5 44 37 31 15
Product 20 2,0 41 36 29 23
Product 8 2,1 41 38 30 23
Product 1 2,2 39 32 26 21
Product 6 2,2 35 31 21 18
Product 18 2,3 38 32 28 21
Product 17 2,6 33 27 22 17
Product 15 2,7 26 19 16 15
Product 10 2,8 25 21 17 15
Product 5 2,8 30 26 24 16
Product 2 3,1 21 18 18 14
Product 3 3,1 34 26 22 15
Table II. Product 12 3,1 35 32 28 26
Relative weekly Product 11 3,4 23 18 18 13
standard deviation Product 7 3,4 25 19 15 13
of manufacturer’s Product 4 3,5 27 25 23 16
production load for Product 13 3,5 28 27 21 15
products with Product 19 3,7 22 22 18 15
different Product 16 3,9 23 22 16 16
replenishment Product 21 4,2 28 18 15 15
frequencies at Product 9 4,5 22 20 16 16
different visibility Average 30 26 22 17
levels Standard deviation 7 7 5 4
Our analysis showed that the three situations significantly differed from one Increasing
another right from the start. When using traditional orders, the relative demand
standard deviation of the production load was, on average, 30 percent for the visibility
weekly planning cycle. When using a two-week cycle, the average relative
standard deviation was smaller, around 21 percent. Finally, a monthly
planning cycle reduced deviation even further, down to an average of 13
percent. These figures show how the importance of the order-batching
349
phenomenon is reduced when using a longer production planning cycle. The
long planning period also smoothens out variation between weeks.
Against this background, it comes as no surprise that the results of our
simulations indicate that, other things being equal, increased visibility is more
valuable when using a shorter production planning cycle. The average reduction
in production load standard deviation when moving from a situation with no VMI
distributors to full VMI adoption was 41 percent when using a weekly planning
cycle, whereas it was only 29 percent for the two longer production cycles. That is,
when full adoption of VMI reduced the average production load variability from
30 percent to 17 percent when using a weekly planning cycle, the reduction was
from 21 percent to 14 percent for the bi-weekly planning cycle, and from 13
percent to 9 percent for the monthly planning cycle (see Table III).
To summarize, the simulation runs demonstrate that there is a strong link
between the manufacturer’s production planning cycle and the potential benefit
of VMI. However, one can also look at the situation from the opposite point of
view: the results indicate that VMI can make it possible for a manufacturer to
benefit from shorter planning cycles, without having to face the full negative
impact of production load variability on capacity utilization. VMI thus reduces
the trade-off between efficient capacity utilization and flexible production.

Conclusions
The simulation model presented in this paper builds on the work of, for
example, Disney and Towill (2001) and Lee et al. (2000), but adds some
important new factors to the analysis.
First, the shift from one customer to multiple customers made it possible to
examine the impact of partial visibility on the manufacturer’s production and
inventory control. This is a very important point of view since situations in
which demand information, such as sell-through data, is available from all

Average production load standard deviation (standard deviation of results) Table III.
Weekly planning (%) Bi-weekly planning (%) Monthly planning (%) Production load
standard deviation
No VMI 30 (7) 21 (6) 13 (4) in weekly, bi-weekly
1/3 VMI 26 (7) 19 (5) 11 (4) or monthly periods
2/3 VMI 22 (5) 17 (4) 10 (3) for different
All VMI 17 (4) 14 (3) 9 (2) visibility levels
IJPDLM customers are rare. The simulations demonstrated that, by combining traditional
33,4 order data with sell-through information available from VMI customers, the
manufacturer could benefit even from a partial increase in visibility. This
presents a significant practical challenge to manufacturers as well as enterprise
resource planning and advanced planning and scheduling solution providers,
since the planning and control solutions currently used typically provide only
350 limited support for combining different types of demand information.
Second, the examination of several products with somewhat different
replenishment frequencies enabled important observations. Cachon and Fisher
(2000) found, in their single-product simulation study, that for stationary
demand, the impact of more frequent ordering typically outweighs the benefits
of information sharing. Approaching the problem from a product range
perspective, as in this study, reveals that there is an important relationship
between the value of information sharing and a product’s order frequency. Our
simulation study indicates that products with low replenishment frequencies
are likely to benefit more from increased visibility. The result is in accordance
with the findings of Kaipia et al. (2002), who used analytical modeling to
develop a measure for the value of demand visibility and made similar
observations regarding the different effects on different product types.
The managerial implications of this finding are significant. VMI presents an
opportunity to increase inventory management and production efficiency
especially for C-products, without having to face the potential inefficiencies,
such as less-than-pallet deliveries, caused by increased ordering frequency. The
simulation results suggest that the benefits of VMI are likely to be greater
when the product range is wide, i.e. when there are several C-products. VMI
also presents an opportunity to extend the product range without decreasing
production efficiency. This is an important observation for companies
considering the implementation of VMI and which products to involve in it.
Third, the examination of the impact of production planning frequency on
the value of visibility was very fruitful. Order batching and, therefore, demand
variability amplification are more important problems when short planning
cycles are used. This means that the impact of increased visibility is greater for
short planning cycles. Hence, the selected planning cycle length directly affects
the results of the study. This suggests that more attention should be paid to
this factor when further studying VMI and supply chain visibility. The results
also illustrate the need to test the main factors embedded in simulation models
to examine what their impact on the results is.
The examination of the production planning frequency variable also
provided some interesting managerial insights. The simulation results indicate
that companies should not expect significant benefits for products with stable
demand when employing long planning periods. This can be one of the reasons
why the benefits of VMI sometimes have been disappointing. However, VMI
also presents an opportunity to move towards shorter planning periods and the
positive effects it brings, such as increased opportunities to react, without Increasing
sacrificing production efficiency. demand
visibility
Limitations and further research
This paper examined one type of automatic replenishment program – VMI. An
important question is whether the simulation results are generalizable to such 351
other replenishment programs as continuous replenishment planning or quick
response. We believe that the results are generalizable at least to some extent.
The main characteristic of the simulation model studied was that the
manufacturer gained access to distributor sell-through data and thus could
remove the impact of the distributors’ order batching on the demand
information used for production and inventory control. The findings should,
therefore, be applicable to a wide range of situations in which manufacturers
get access to downstream demand data, such as customer sell-through data or
retail outlet point-of-sales data. However, as the different automatic
replenishment programs give the manufacturers different levels of authority
and different rights to organize replenishments, they can, in some situations,
find it difficult to realize the potential benefits of increased visibility. If, for
example, a manufacturer has access to its customers’ demand information but
still has to operate based on the customers’ sometimes surprising orders, it may
be difficult to eliminate buffers in the supply chain (Kaipia et al., 2002).
Another notable limitation of the research is that it only looks at products
with stable demand. From a supply chain management point of view, this type
of demand is the least challenging and the easiest to manage. Interestingly, the
simulation results indicated some notable efficiency improvement
opportunities resulting from the implementation of VMI even for this type of
demand. Limiting the examination to products with stable demand may,
however, cause the model to appear unrealistic. Although the situation is
indeed unrealistic, we argue that the results are still valid. There is no reason to
assume that the positive effects of eliminating distributor order batching would
be any smaller for, for example, seasonal products. In fact, since VMI provides
an opportunity to look forward by revealing demand downstream in the supply
chain, the arrangement is likely to be more valuable for products with changing
demand. However, in these situations the importance of other factors, such as
accurate forecasts, increase and these need to be incorporated in the supply
chain model. More research on VMI for products with varying demand is, thus,
needed. The next step is to, as Lee et al. (2000) did for their single-product,
single-customer analytical model, include different types of changing demand
in the simulation model and examine how this affects the results.
Although the production planning and inventory control model used in the
simulation is somewhat more complex than earlier models, it still provides a
very simplified picture of reality. To be able to deal with strong seasonality,
product introductions, and promotions, forecasts need to be included in the
IJPDLM model. This is, in fact, the same step that the fast moving consumer goods
33,4 industry has taken when moving from automatic replenishment programs,
such as VMI, to more comprehensive collaboration models, such as
collaborative planning forecasting and replenishment (CPFR) that also
include planning and forecasting. In addition, different types of production
constraints, such as batch constraints and capacity constraints, as well as more
352 sophisticated production planning algorithms could be introduced to make the
model more realistic and, thus, more valuable for companies trying to benefit
from demand visibility by implementing VMI.
Finally, it is important to keep in mind that models are always
simplifications of reality. Therefore, more case studies focusing on
companies that have implemented VMI and studying the benefits they have
attained as well as the problems they have faced in practice are needed. This is
the only way to get reliable information on the actual processes employed by
the companies as well as on the important relationship issues that are so critical
to the success of collaborative approaches such as VMI.

References
Bookbinder, J. and Heath, D. (1988), “Replenishment analysis in distribution requirements
planning”, Decision Sciences, Vol. 19 No. 3, pp. 477-89.
Burbidge, J.L. (1989), Production Flow Analysis – for Planning Group Technology, Clarendon
Press, Oxford.
Burbidge, J.L. (1994), “The use of period batch control (PBC) in the implosive industries”,
Production Planning & Control, Vol. 5 No. 1, pp. 97-102.
Cachon, G. and Fisher, M. (1997), “Campbell Soup’s continuous replenishment program:
evaluation and enhanced inventory decision rules”, Production and Operations
Management, Vol. 6 No. 3, pp. 266-76.
Cachon, G. and Fisher, M. (2000), “Supply chain inventory management and the value of shared
information”, Management Science, Vol. 46 No. 8, pp. 1032-48.
Chen, F. (1998), “Echelon reorder points, installation reorder points, and the value of centralized
demand information”, Management Science, Vol. 44 No. 12, pp. 221-34.
Chen, F., Drezner, Z., Ryan, J.K. and Simchi-Levi, D. (2000), “Quantifying the bullwhip effect in a
simple supply chain: the impact of forecasting, lead-times and information”, Management
Science, Vol. 46 No. 3, pp. 436-43.
Clark, A.J. and Scarf, H. (1960), “Optimal policies for a multi-echelon inventory problem”,
Management Science, Vol. 6 No. 4, pp. 475-90.
Cooke, J. (1998), “VMI: very mixed impact?”, Logistics Management and Distribution Report,
Vol. 37 No. 12, pp. 51-3.
Daugherty, P.J., Myers, M.B. and Autry, C.W. (1999), “Automatic replenishment programs: an
empirical examination”, Journal of Business Logistics, Vol. 20 No. 2, pp. 63-82.
DeKok, T. and Fransoo, J. (2002), “Handbook of operations research: supply chain management”,
manuscript, 9 September.
Disney, S.M. and Towill, D.R. (2001), “The impact of vendor managed inventory on the bullwhip
effect in supply chains”, Proceedings of the 16th International Conference on Production
Research, Prague, 30 July-2 August.
Euwe, M.J. and Wortmann, H. (1997), “Planning systems in the next century (I)”, Computers in Increasing
Industry, Vol. 34 No. 2, pp. 233-7.
Forrester, J. (1961), Industrial Dynamics, MIT Press, Cambridge, MA.
demand
Fransoo, J. and Wouters, M.J.F. (2000), “Measuring the bullwhip effect in the supply chain”, visibility
Supply Chain Management: An International Journal, Vol. 5 No. 2, pp. 78-89.
Fransoo, J., Wouters, M.J.F. and DeKok, T. (2001), “Multi-echelon multi-company inventory
planning with limited information exchange”, Journal of the Operational Research Society, 353
Vol. 52, pp. 830-8.
Fraza, V. (1998), “Streamlining the channel”, Industrial Distribution, Vol. 87 No. 9, pp. 73-4.
Holmström, J. (1998), “Business process innovation in the supply chain: a case study of
implementing vendor managed inventory”, European Journal of Purchasing & Supply
Management, Vol. 4 No. 2-3, pp. 127-31.
Houlihan, J.B. (1987), “International supply chain management”, International Journal of Physical
Distribution & Materials Management, Vol. 17 No. 2, pp. 51-66.
Kaipia, R., Holmström, J. and Tanskanen, K. (2002), “What are you losing if you let your
customer place orders?”, Production Planning & Control, Vol. 13 No. 1, pp. 17-25.
Lapide, L. (2001), “New developments in business forecasting”, Journal of Business Forecasting
Methods & Systems, Vol. 20 No. 4, pp. 11-13.
Lee, H.L. (2002), “Unleashing the power of intelligence”, ECR Journal, Vol. 2 No. 1, pp. 61-73.
Lee, H.L., Padmanabhan, V. and Whang, S. (1997a), “Information distortion in a supply chain: the
bullwhip effect”, Management Science, Vol. 43 No. 4, pp. 546-58.
Lee, H.L., Padmanabhan, V. and Whang, S. (1997b), “The bullwhip effect in supply chains”, Sloan
Management Review, Vol. 38 No. 3, pp. 93-102.
Lee, H.L., So, K.C. and Tang, C.S. (2000), “The value of information sharing in a two-level supply
chain”, Management Science, Vol. 46 No. 5, pp. 626-43.
Lin, F., Huang, S. and Lin, S. (2002), “Effects of information sharing on supply chain performance
in electronic commerce”, IEEE Transactions on Engineering Management, Vol. 49 No. 3,
pp. 258-68.
Maloni, M.J. and Benton, W.C. (1997), “Supply chain partnerships: opportunities for operations
research”, European Journal of Operational Research, Vol. 101, pp. 419-29.
Silver, E.A., Pyke, D.F. and Peterson, R. (1998), Inventory Management and Production Planning
and Scheduling, 3rd ed., Wiley, New York, NY.
Stenger, A.J. and Cavinato, J.L. (1979), “Adapting MRP to the outbound side – distribution
requirements planning”, Production and Inventory Management, Vol. 20 No. 4, pp. 1-13.
Towill, D.R. (1991), “Supply chain dynamics”, International Journal of Computer Integrated
Manufacturing, Vol. 4 No. 4, pp. 197-208.
Towill, D.R., Naim, M.M. and Wikner, J. (1992), “Industrial dynamics simulation models in the
design of supply chains”, International Journal of Physical Distribution & Logistics
Management, Vol. 22 No. 5, pp. 3-14.
Vergin, R.C. and Barr, K. (1999), “Building competitiveness in grocery supply through
continuous replenishment planning: insights from the field”, Industrial Marketing
Management, Vol. 28 No. 2, pp. 145-53.
Vollmann, T., Berry, W. and Whybark, C. (1997), Manufacturing Planning and Control Systems,
4th ed., Irwin/McGraw-Hill, New York, NY.
Waller, M., Johnson, M.E. and Davis, T. (1999), “Vendor-managed inventory in the retail supply
chain”, Journal of Business Logistics, Vol. 20 No. 1, pp. 183-203.
Yu, Z., Yan, H. and Cheng, T.C.E. (2000), “Benefits of information sharing with supply chain
partnerships”, Industrial Management & Data Systems, Vol. 101 No. 3, pp. 114-9.
IJPDLM Appendix
33,4

354

Figure A1.
Logic of the simulation
model