Content uploaded by Mauricio Camargo
Author content
All content in this area was uploaded by Mauricio Camargo on Feb 07, 2015
Content may be subject to copyright.
Available via license: CC BY
Content may be subject to copyright.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
61
Multiproduct supply chain
- Strategic planning and forecasting
Jose F. Roig Zamora
Industrial Engineering School
Engineering Faculty, University of Costa Rica
Santo Domingo, Costa Rica
E-mail: roigjose@gmail.com
Raymundo Q. Forradellas
Logistics Studies and Applications Centre.
Facultad de Ingeniería. Universidad Nacional de Cuyo.
Centro Universitario, Mendoza, Argentina.
E-mail: kike@uncu.edu.ar
Mauricio Camargo
INPL. Institut National Polytechnique de Lorraine
France
E-mail: mauricio.camargo@ensgsi-inpl-nancy.fr
Abstract
The relation among the actor’s of the Supply Chain defi nes its main characteristics,
and therefore the Distribution and Manufacturing Strategy that the actors must follow
in order to fulfi ll the Service Equation. In a Multiproduct Supply Chain, the different
Negotiating Force of the different actors will truly infl uence in the fi nal design on the
Chain Confi guration. Depending on which actor has more power, the Supply Chain must
react to different supply policies. Forecasting Tools are presented as an option to predict
the product Distribution and Manufacturing needs and as a way to counterbalance the
different negotiating force among actors.
Key words: Logistics, Business Strategy, Forecast, Distribution & Multiproduct
Manufacturing.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
62
INTRODUCTION
Today’s world is becoming a global market with disappearing boundaries. Nowadays,
one of the critical constraints for companies, are the accuracy of manufacturing,
movement, and storage for the products along the Supply Chain, within the functions
that make it possible to happen: according to Chopra et al. (2004) “distribution refers
to the steps taken to move and store a product from the supplier stage to a customer
stage in the Supply Chain”. In order to have products moved and commercialized, the
manufacturing function is needed along the Supply Chain since it works as the chain’s
supplier. According to Bowersox et al. (2002) “manufacturers add value by converting
raw materials into consumer or industrial products”, since manufacturing takes time in
terms of production processes, production lead times tends to be longer than distribution
lead times and so manufacturing processes are more forecast-dependent than distribution
processes; in a MTP/MTS context (Make-to-Plan/ Make-to-Stock); see Bowersox et al.
(2002)). Transportation, Distribution, Storage and Manufacturing Logistics play a critical
role in the Service Equation: Delivery Time, Place, Quantity and Cost.
The relationship between the actors of the Supply Chain defi nes the Distribution
and Manufacturing Strategy that the actors must follow. It is required then, to analyze
the business characteristics and to determine which is the most convenient strategy
to fulfi ll the Service Equation, and later on, materialize this strategy in the Company’s
logistic procedures in order to make it happen.
This paper is organized as follows: section 2 gives an overview of what we propose
as a Basic Supply and Distribution Network model; in section 2.1 we propose an
application of the Quantitative Forecasting Tools within this Basic Distribution Network;
later on we propose the Multi-product Distribution Network model; in section 2.2 we
propose an application for Qualitative and Quantitative Forecasting Methods in terms
of the “Method Category-Aggregation Level Matrix” applied to a Manufacturing context;
this matrix proposes an application of forecasting techniques for a Multiproduct
Manufacturing environment, which consists in an integration of the Quantitative and
Qualitative Forecasting methods and the different potential aggregation degrees of the
products. In section 3 we propose a categorization of the different Negotiating Force
scenarios between Customer and Supplier that must be taken into account in order to
plan the Distribution and Manufacturing Strategy, to strategically deal with important
customers. Section 4 proposes some fi nal conclusions.
STRATEGIC PLANNING FOR THE SUPPLY CHAIN NETWORK
Knowing the market and the environment where the business develops is an
important step to defi ne the Supply, Production, and Distribution Policies. There
are many different kinds of Distribution Network confi gurations that have evolved
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
63
during the years, depending on the nature of its Business and the power of its actors:
suppliers, customers and market. In order to study these networks we propose to reduce
this diversity and to study the simplest network. Once it has been studied, it will be
possible to make conclusions and, later on, generalize them to the complexity of the
entire Network. Let’s propose the following Basic Supply and Distribution Network:
Demand Cd
Demand Cc
Demand C
a
Demand Cb
C
a
C
d
C
c
C
b
S
X
0
400
800
1200
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
200
400
600
800
1000
1 2 34 5 67 8 91011121314151617181920212223
0
500
1000
1500
2000
2500
1 2 34 5 67 8 91011121314151617181920212223
0
1000
2000
3000
Ene Feb Mar Abr May Jun Jul Ago Sep Oct Nov Dic
Figure 1. Basic Supply and Distribution Network and Demand Profi le for each Ci
Let Ci. i be the Customer who demands product from S. Ci may have other demands
confi rmed by other(s) Customer(s) not showed in the drawing.
Let S be the Supplier for Cii. i (for i=a,b,c and d; the Basic Network could have n
Customers).
Let X be one Product that moves along the network according to Ciii. i’s demand.
Each Civ. i is supplied of Product X exclusively by the Supplier S.
For each Cv. i, we have the recent historical Monthly Demand (sales curve). Demand
behavior is similar for every Ci.
Supplier S supplies uniquely its Product X to the Customers Cvi. i’s. S has to work
out the Production and Distribution Plan for Product X according to its customers
needs.
Suppose that transportation time is relatively short, so it is possible to vii.
approximate the Global Demand for S (in terms of time and quantity), as the
sum of the individual demands in each Ci.
Suppose that transportation cost is high, so transportation cost is very sensitive viii.
to freight consolidation.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
64
Now, which strategy must the Supplier S follow in order to create a Supply Policy for
its Customers Ci? Are all its Ci Customers asking for more product than needed? Can S
trust this current sales data in order to make a global prediction? Is Supplier S pushing
the product to its Ci’s so there will be big chances that global demand decreases because
of overstocking at each Ci’s warehouse? How can the Manufacturer produce in order to
supply according to the Distribution needs?
DISTRIBUTION FORECASTING
Within the cooperation frame between enterprises categorized as S and C, it is
very important to foster the mutual collaboration when building the Operations Plans:
Demand Plan, Production Plan and the Distribution Plan. Relationships between non-
collaborating enterprises show supply problems such as: product stockouts, overstocking,
considerable forecasting errors, etc.
Many of these problems come from some companies’ lack of cooperation and
the differences in Negotiating Forces that exists among the actors in the network;
differences that we will comment at the end of this publication. Based on a policy of
mutual collaboration between S and C, how can a forecast be calculated in order to
have a distribution plan (time and quantity) through a Basic Supply and Distribution
Network? We propose to do this using Quantitative Forecasting Tools. Historical Demand
data for Product X and four C’s (four Customers) is showed in Figure #2.
Sales; Liters
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Year 2003
Item
Real Real Real Real Real Real Real Real R eal Real Real Real
Customer A
766 1,279.68 1,363 1,784.70 1,646 1,641.74 1,460 2,005.26 1,610 1,437.45 1,399 1,204.60
Customer B
575 863.04 1,153 1,189.80 1,213 1,106.39 904 2,005.26 1,449 1,273.17 1,166 919.30
Customer C
192 238.08 419 436.26 650 499.66 661 791.55 644 739.26 433 538.90
Customer D
383 595.20 559 555.24 823 321.21 452 474.93 322 657.12 333 507.20
total 1,916 2,976 3,495 3,966 4,332 3,569 3,477 5,277 4,024 4,107 3,332 3,170
Year 2004
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Real Real Real Real Real Real Real Real R eal Real Real Real
Customer A
1,444 1,534 2,253 2,351 1,755 2,439 2,087 2,050 2,675 1,685 2,370 1,807
Customer B
1,083 1,035 1,907 1,568 1,293 1,644 1,292 2,050 2,408 1,492 1,975 1,379
Customer C
361 285 693 575 693 742 944 809 1,070 867 733 809
Customer D
722 714 924 732 877 477 646 486 535 770 564 761
total 3,611 3,568 5,778 5,225 4,618 5,303 4,968 5,395 6,688 4,814 5,642 4,756
Year 2005
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Real Real Real Real Real Real Real
Customer A
1,470 2,536 3,208 3,166 3,062 3,057 2,898
Customer B
1,103 1,710 2,714 2,111 2,256 2,060 1,794
Customer C
368 472 987 774 1,209 930 1,311
Customer D
735 1,180 1,316 985 1,531 598 897
total 3,676 5,898 8,225 7,035 8,058 6,646 6,900
Sales History
Figure 2. Basic Supply and Distribution Network. Ci Sales/Demand History for S
In order to forecast demand, we can use several well-known Quantitative Forecasting
Methods: Moving Average, Simple Exponential Smoothing, Trend Corrected Exponential
Smoothing (Holt’s Model), Trend and Seasonality Corrected Exponential Smoothing
(Winter’s Model) and the Static Method, among the most popular forecasting methods
according to Chopra et al. (2004). It is not an objective of this article to explain the
algorithm of each forecasting method, but to set a guideline of an application of these
methods in a Distribution context.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
65
Following the Model showed in Figure #1, and calculating the Forecast for the Global
Demand according to data showed in Figure #2, we can see the results in the following
comparative chart. For each of the Forecasting Methods we have compared the most
common Forecast Evaluating Measures (Error, Absolute Error, MAD (Mean Absolute
Deviation), and MAPE (Mean Absolute Percentage Error)). Based on this, is possible to
evaluate the convenience of choosing one method (see Figure #3).
Summary Table
Forecasting Effectiven ess Indicators
MAD MAPE
Method Mean Absolute Deviati on
Mean
Absolute
Percentage
Moving Aver age 896 17 -11,29 -0,22
Simple Exponenti al Smoothing 1063 24 -9,37 11,07
Trend Corrected Exponential Smoothing (Holt's Model) 760 17 -3,78 5, 19
Trend and Seasonality Corrected Ex ponential Smoothing (Winter 's Model) 411 9 -3,62 5,90
Static Method 372 8 -3,91 5,27
Tracking Signal
TSt
Figure 3. Forecast Evaluating Measures for the Global Demand Forecast.
The best MAPE (8% in this case) is related to the Static Method. The second best
MAPE is related to the Winter’s Method which shows 9%. When analyzing the MAD, the
best values are related to the Static and the Winter Method with 372 and 411 units.
Simple Exponential Smoothing method yields a variation (1063 units) that exceeds the
double of the variation related to the Winter’s Model. MAD is related to the random
component of the demand, so, the bigger the MAD, the forecast for the real demand
becomes more variable. According to Chopra et al. (2004) “the MAD can be used to
estimate the standard deviation of the random component assuming that the random
component is normally distributed”.
The Holt’s, the Winter’s and the Static methods show the steadiest Tracking Signal
values. Tracking Signals measure the consistency of the method according to its
capacity to not to bias its predictions. One biased prediction can consistently over
or underestimate demand; the normal bias will fl uctuate around zero since it will be
random; please refer to Chopra et al. (2004).
In this case, either the Static Method or the Winter’s Method would be chosen over
the others methods. The convenience of using the Winter’s Method rather than the
Static Model is that Winter’s has a dynamic characteristic, since this method takes into
account the evolution of new demand and changes the Method’s parameters (Level,
Tendency and Seasonality Factors). On the other hand, the Static Method does not
change; the parameters of the initial calculations are used until the initial calculation is
run once again. Winter’s Method (because of its self-changing properties) is convenient
for multiproducts environments (since many different products can be forecasted
without having to recalculate the parameters each time).
It is also possible to calculate an individual forecast for each network’s node; the
same Winter’s analysis could be done for each Ci node. It is up to the analyst to set the
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
66
convenience of the aggregation level for the forecasting, since for many cases it would
be important to calculate the forecast for all the nodes as a big node, and in other cases
it would be important to calculate each node’s forecast (for example, if it is the case
of a Distribution Center that supplies all nodes, it is useful to calculate the forecast
for the four nodes as a big node since we want to forecast the demand that will be
allocated to this Distribution Center; later on we will distribute product to each node).
Decision must be based on the real network’s features and the possibility to postpone
Distribution based on pull requirements and transportation feasibility.
Please refer to the next fi gure #4 Winter’s Method Evaluation for Ca node showing the
Winter’s calculations for this node. Same calculations should be done when forecasting
Cb, Cc and Cd demands.
Trend and Sea sonality Corrected Exponential Smoothing (Winter`s Model)
Alfa= 0.05 Beta= 0.1 Gama= 0.1
Tracking
Signal
year month period
demand
Dt
Level Lt Trend Tt
Seasonal
Factor, St
Forecast,
Ft
Error, Et
Absolute
error, At
Mean
Squared
Error, MS Et
MADt %Error MAPEt TSt
0
1090 52
2003 1 1 766. 4 1140 52 0.70 800 34 34 1125 34 4 4.38 1.00
2003 2 2 1279.68 1198 53 0. 98 1172 -108 108 6386 71 8 6.41 -1.05
2003 3 3 1363.05 1245 52 1. 19 1491 128 128 9713 90 9 7. 40 0.60
2003 4 4 1784.7 1302 53 1.28 1657 -127 127 11341 99 7 7.33 -0.74
2003 5 5 1646.16 1362 53 1. 09 1481 -165 165 14545 112 10 7.88 -2. 13
2003 6 6 1641.74 1415 53 1. 17 1661 19 19 12182 97 1 6.76 -2.27
2003 7 7 1460.34 1466 53 1. 02 1504 43 43 10710 89 3 6.22 -1.98
2003 8 8 2005.26 1530 54 1. 14 1738 -267 267 18281 111 13 7.10 -3. 98
2003 9 9 1609.6 1577 53 1.13 1786 176 176 19696 119 11 7.53 -2.26
2003 10 10 1437.45 1636 54 0. 82 1339 -98 98 18689 117 7 7.46 -3. 14
2003 11 11 1399. 44 1680 53 0.94 1584 184 184 20072 123 13 7.98 -1.48
2003 12 12 1204.6 1728 53 0.74 1276 71 71 18821 118 6 7.81 -0. 93
2004 1 13 1444.4 1795 54 0.70 1242 -203 203 20530 125 14 8.28 -2.51
2004 2 14 1534.24 1834 52 0.99 1833 299 299 25458 137 20 9.08 -0.10
2004 3 15 2253.42 1888 53 1.18 2232 -22 22 23792 130 1 8.54 -0. 28
2004 4 16 2351.25 1935 52 1.29 2497 146 146 23639 131 6 8.40 0.84
2004 5 17 1754.84 1967 50 1.10 2195 440 440 33624 149 25 9.38 3.69
2004 6 18 2439.38 2020 50 1.17 2364 -76 76 32073 145 3 9.03 3.28
2004 7 19 2086.56 2069 50 1.02 2115 28 28 30428 139 1 8.63 3.63
2004 8 20 2050.1 2101 48 1.16 2461 410 410 37329 152 20 9.20 6. 00
2004 9 21 2675.2 2162 50 1.12 2400 -275 275 39161 158 10 9.25 4.04
2004 10 22 1684.9 2203 49 0.83 1830 145 145 38335 158 9 9.22 4. 97
2004 11 23 2369. 64 2267 50 0.93 2086 -283 283 40156 163 12 9.34 3.07
2004 12 24 1807. 28 2325 51 0.73 1697 -111 111 38993 161 6 9.20 2. 42
2005 1 25 1470.4 2361 50 0.71 1682 212 212 39228 163 14 9.41 3. 69
2005 2 26 2536.14 2420 50 0.98 2352 -184 184 39018 164 7 9.33 2.55
2005 3 27 3207.75 2482 52 1.18 2925 -283 283 40544 168 9 9.31 0.80
2005 4 28 3165.75 2531 51 1.28 3243 78 78 39311 165 2 9.06 1.28
2005 5 29 3062.04 2595 53 1.08 2798 -264 264 40359 168 9 9.05 -0.31
2005 6 30 3057.16 2645 52 1.18 3112 55 55 39116 164 2 8.81 0.02
2005 7 31 2898 2704 53 1. 02 2752 -146 146 38540 164 5 8. 69 -0.87
Forecast E quation Ft+l = (Lt + lTt) * S t+l
1 8 32 1.14 3151
2 9 33 1.13 3172
3 10 34 0.82 2351
4 11 35 0.94 2737
5 12 36 0.74 2188
Coeficients
Interception 1090.32
Variable X 1 52. 266
-500
0
500
100 0
150 0
2000
2500
3000
3500
4000
12345678910111213141516171819202122232425262728293031
Demand Dt
Forec ast , Ft
Error, Et
MADt
Forec as t
plus MAD
Error MAD
included
Figure 4. Winter’s Method Evaluation for Ca node
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
67
Some remarks related to fi gure #4 are:
We propose the use of the graphic tool as a way to display and, therefore, •
understand the effectiveness of the forecasting method along recent historic
data.
Negative Forecast Error represents stockouts (when the forecasted line is below •
the demand line this represent a stockout); positive Forecast Error is related to
overstocking.
Using MAD, it is possible to estimate the standard deviation of the demand’s •
random component. Using this criterion, it is possible to set a policy of Safety
Inventory, due to the fact that if we add a MAD factor to the Forecast, it is
possible to reduce the possible stockouts using higher inventory level at each
node; also a global Safety Inventory can be set in S and “Pull” according to each
node’s requirements.
The • Forecast plus MAD line (semi-continuous green line) is exactly the same line
that the yellow one (Forecast Line); note that the difference is that Forecast
Plus MAD line has been moved up by adding a 1.0 MAD factor to the forecasted
values. The standard deviation on the demand’s random component is considered
to be 1.25MAD, so this Safety Inventory is related to a protection of less than a
standard deviation. We propose that this level should be set qualitatively by the
analyst according to the Supply Chain’s inherent characteristics.
Using this criterion, it is possible to set the Safety Inventory of the Distribution Plan.
Now, it is important to defi ne policies regarding where to keep this Safety Inventory:
Should we keep it at each node? Should we aggregate it in a strategic node and pull it
according to current demand evolution? The answer to these questions lies within each
Strategic Network case and the postponement possibilities.
The following fi gure shows a summary of the Distribution Forecast for August,
September, October and November. The Winter’s Forecast shown is not altered with
any MAD protection factor. This forecast application allows the company’s analyst to
calculate the Supply Chain’s forecast for all the items that must be Distributed along
the network’s nodes.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
68
0
500
1,000
1,500
2,000
2,500
3,000
3,500
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
2003
2004
2005
Winter's
Forecast
Figure 5. Basic Distribution Network. Winter’s Method Forecast for Ca node.
Another important remark regarding the Simple Distribution Network, is that
Forecast can help to defi ne the Pull-Push Distribution boundary. In this case, the Push
Method can be used to send product to each node according to the forecasted needs
(since this demand has some degree of certainty and this allows to profi t from the
Transportation Economy of Scale). Pull Methods can be used to handle the uncertainty
demand (MAD) and pull stock from other nodes. Increasing the forecasting effectiveness
for each node minimizes overstocking in certain nodes and stockouts in others, since
product allocation within the Basic Distribution Network will be more effective.
Using the Basic Distribution Network as a basis, we can jump into conclusions when
analyzing Multiproduct Distribution Networks. Multiproduct Distribution Networks are
similar to the Basic Network but its confi gurations change since S supplies different
products (x,y,z,...n) to each one of its Ci’s, which makes Networks much more complex.
Please refer to fi gure #6. When forecasting the Multiproduct Network, it is possible
to use the same forecasting procedure already presented; but when planning the
Transportation Plan, it is important to take into account that Transportation now
should consolidate different products fostering the economy of scale of the trip.
Ca
Cd
Cc
Cb
S
X
Y
Z
Cj
Ci
Ce
Cf
Ch
Cg
Figure 6. Multiproduct Distribution Network.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
69
The Strategic Planning for a Multiproduct Distribution Network is much more
complicated since this Network has to take into account Multi-Relationships among the
multiples Si and Ci and different products (x,y,z,…n). At the same time, these Si actors
play the Ci roll for other actors and vice-versa. These relations will be discussed later.
Quantitative Forecasting methods are not enough for Multiproduct Networks.
Qualitative Methods can improve the Forecast effi ciency since they include predictions
based on expected future facts (as per Carranza (2004), it is necessary to use forward
information) not included in the historic information, for example: new markets or
customers. The Qualitative and Quantitative Methods interaction will be presented
in the Manufacturing application that follows; a future study branch will be how to
integrate the Qualitative Methods in the Distribution context.
MANUFACTURING FORECASTING
Quantitative Methods can be automated, since it is possible to use computers to work
out the Forecast for many products. Qualitative Methods are more diffi cult to implement
since the expert criterion should be heard and this is a time-intensive process. Another
important factor to consider when forecasting is the aggregation level, since it is easy
to work out a Quantitative Method for a SKU (Stock Keeping Unit) level, but it is
almost impossible to do so using a Qualitative Method (because of the large quantity of
SKU’s in the multiproduct scenario, which results very diffi cult for humans to manage).
Nevertheless, the expert criterion is easy to take into account for a higher aggregation
level (family level, market level, etc…). According to Bowersox et al (2002) and Frazelle
(2002), it is important to integrate and rationalize top-down and bottom-up forecasts
with human intelligence. During our application in the Distribution context, we have
realized the importance of considering “qualitative input”. For this reason, we have
included this consideration within the Manufacturing context.
Manufacturing is the Supply Chain’s source; it feeds product to the chain and
makes possible the Distribution process afterwards. Manufacturing increasingly faces
the product proliferation phenomenon in terms of demand and product diversity. This
proliferation has made diffi cult to match the product’s supply and demand, especially
since factors such as strict customer needs, lead time reductions requirements, life cycle
reductions, globalization and obsolescence risk increase due to emerging technologies
and competition proliferation that have made this match harder than ever. Since
Manufacturing is the Supply Chain’s source, it has to be strictly planned in order to
guarantee product availability along the Chain.
Some techniques have already been developed in order to counterbalance this
proliferation phenomenon; among the most popular we have “manufacturing postponement”
and “logistics postponement”, as per Bowersox et al. (2002). These techniques are based
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
70
on “Pull” principles. Nevertheless, most of the companies feel the environment’s pressure
in terms of a great dependence of Push Manufacturing Strategies (MTP/MTS for example;
please note that all Manufacturing Strategies, even MTO or ATO1 have certain degree of Push
Manufacturing; as is the case of components procuring) since there is a need to promptly
fulfi ll customer needs and therefore, speculative (forecasted) needs have to be considered
in advance in order to manufacture products prior to customer orders (when it is not
possible to attain a fl exible and capable manufacturing system). This Push Manufacturing
dependence makes the precision of the Forecasting Process even more critical.
MULTIPRODUCT FORECASTING CALCULATION COMPLEXITY
The complexity of Multiproduct Forecasting calculations relies on internal and
external factors. Among such internal factors we could highlight: large quantity of
items (SKU’s), a big pool of clients, a lot of different family products, new products
coming out everyday, products with correlated demand, complementary products, high
obsolescence rate due to product characteristics and nature, and so on. All these factors
and other manufacturing dynamics must be taken into consideration when the analysts
make forecast calculations for each SKU.
At the same time, analysts must take into consideration external factors such as:
changing markets, sales risk increases, market expansions, demand oscillations, higher
product obsolescence rate due to new technologies, the proliferation of competition,
etc. All these factors must be taken into account in the Forecasting Process, especially
through the use of Qualitative Methods.
We propose to integrate internal factors and external factors at the same time. A
rich source with basic information could be historical sales data; in this data we can
fi nd the historic internal factors’ interaction and the real demand that the company
has faced. Through the study of this data, it is possible to calculate (for each SKU)
the demand components such as Level, Seasonality and Trend; using the same analysis
that we have already done in the Distribution case. This analysis or technique, also
known as “back-casting”, as stated by Frazelle (2002), allows us to calculate forecasts
using several quantitative methods and then compare its capabilities to predict the
demand’s pattern in order to choose the best method. Usually this technique is easy to
automate since quantitative methods are composed by mathematics calculations. This
feature makes “back-casting” a possible method to be used in a multiproduct context
since it is easy to calculate forecasts for a lot of SKU’s using computational resources.
Nevertheless, this method is based on the assumption that future sales behavior could
be predicted based on historic sales; in this case, it is understood that the internal
and external conditions will be the same in the future, so they will be likely to repeat
1 MTO or Make-to-Order; ATO or Assembled-to-Order.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
71
themselves and so, we could forecast future relying on the past. This assumption is not
totally valid, since it is very likely that conditions will change because of the market
and company’s dynamics; as is the case of political variables as per Carranza (2004).
This constraint has made us consider the need to incorporate to the forecast
calculations factors that could change future demand. Several authors agree with this
and state that “in order to improve the forecasts, it is important to obtain forward
information” as could read the translation of what Carranza (2004) has stated. In order
to attain this integration, we propose to integrate the Quantitative and the Qualitative
Methods in their convenient aggregation level.
QUANTITATIVE METHODS
As previously commented, Quantitative Methods decrease calculations times and
their complexity in a Multiproduct context. As presented in the Distribution case, we
will use fi ve of the most used methods, and we will judge their prediction capability
for each product demand pattern based on the measures of forecasting error already
presented; please refer to section 2.1 Distribution Forecasting.
We propose to use the Quantitative Methods in a Low Aggregation Level. Low
Aggregation Level has to be defi ne for each SKU; we propose to aggregate the SKU’s
in the lower but convenient aggregation level; for example we can aggregate SKU’s in
small families that includes similar or related SKU’s, or we could aggregate single SKU’s
(which would be seen as one-member family). We propose to use Quantitative Methods to
profi t from the historic data related to each low aggregation level and the possibility of
individual calculations; we propose to chose the best method that forecasts the product’s
demand; as is the case of the following fi gure which present a coordinate for Winter’s
or Holt’s Method as the chosen method for a certain SKU aggregation level. After all
calculations have been completed for all SKU’s, all these results are considered together
as one coordinate (SKU aggregated). Note that Quantitative Methods could be used in
Higher Aggregation Levels, but we propose to use them in Low Levels; see next fi gure.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
72
Qualitative
Methods
Quantitative
Methods
Low Aggregation
Level
High Aggregation
Level
W inter 's Met hod or
Holt's, etc.
SKU
Aggregation
Figure 7. Quantitative Methods and Low Aggregation Level in the “Method Category-Aggregation
Level Matrix”.
QUALITATIVE METHODS
Within the “Method Category-Aggregation Level Matrix”, we propose to incorporate
subjective variables to the forecast calculations at Higher Aggregation Levels. The
subjective factors that are incorporated through Qualitative Methods include the
manager’s intuition (intuition developed based on the manager’s experience and “know-
how”), previous knowledge of variables that will affect the demand’s level (for example:
temporary offers, temporary product importation that will compete with the company’s
products, future market conditions, etc.), and others.
Expert criterion allows the analyst to incorporate his intuition into the forecast in a
subjective manner for future demand. This criteria incorporates factors that will affect
the future and that perhaps have not impacted sales during historic sales, therefore it
permits to consider trends that would not be taken into account by the Quantitative
Methods, which base their decision only on historic data. Among the most popular
Qualitative Methods, we can highlight the following: a. Opinion Jury, b. Commercial
Personnel Proposition, c. Delphi’s Method, d. Market Research, and others as presented
in Heizer et al. (2001).
Within a Multiproduct Manufacturing frame, it is more feasible to consider the
expert’s criteria in Higher Aggregation Levels and in monetary terms (revenues). It
is very diffi cult for a Sales Department or for a Manager to estimate a forecast with
certainty for every single SKU. Nevertheless, when forecasting SKU groups or even
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
73
product families, qualitative forecasting is easier and precise. For example, it is easier
for a Sales Manager to estimate global sales of 4 million dollars and it is very likely that
this forecast become precise since the expert knows his company’s sales behavior; his
expert knowledge allows him to jump into subjective predictions related to multiples
variables and factors. These predictions are truly diffi cult to obtain via mathematical
models and its numerous relations that are hard to represent and justify mathematically
speaking (especially since it represents complicated and time-consuming tasks). As
commented by Silver (1985) there is a relationship between the method that has been
used and the aggregation level; expert criteria is essential for the aggregated midterm
forecasts. The following fi gure shows the two coordinates presented at the moment,
as is the case for Qualitative Methods which are showed in High Aggregation Levels
(Global Aggregation).
W inter 's Met hod or
Holt's, etc.
Global
Aggr egation
Delphi's Method
or Opinion Jury
Low Aggregation
Level
High Aggregation
Level
SKU
Aggregation
Qualitative
Methods
Quantitative
Methods
Figure 8. Qualitative Methods and High Aggregation Level in the “Method Category-Aggregation Level
Matrix”.
INTEGRATION OF QUANTITATIVE AND QUALITATIVE METHODS
In fi gures 7 and 8, we can recognize the coordinate “method category-aggregation
level” concept within the matrix. Each of these coordinates suggests that each method
is convenient to be used at a certain aggregation level; convenience that we have
already discussed in terms of precision and calculation feasibility. This concept allows
us consider the possibility of playing with several coordinates and integrate its results
in order to achieve better forecasts. In this case, we propose to profi t from the different
advantages regarding each one of the coordinates and integrate them.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
74
In order to conceptualize this integration, we propose to create the “Integration
Constant Axis (Φ)” in the matrix; this axis integrates the two coordinates. The
“Integration Constant Axis (Φ)” presents the infi nite possible integration combinations
between these two coordinates; please refer to the next fi gure.
W inter 's Met hod or
Holt's, etc.
Global
Aggr egation
Delphi's Method
or Opinion Jury
Low Aggregation
Level
High Aggregation
Level
SKU
Aggregation
Qualitative
Methods
Quantitative
Methods
Integration Constant
Axis (Φ)
Figure 9. “Method Category-Aggregation Level Matrix” and the “Integration Constant Axis (Φ)”.
Once the “Integration Constant Axis (Φ)” is drawn, it is necessary then to determine
the constant value that better integrates both coordinates. When defi ning Φ we
propose to use qualitative criteria using the expert opinion regarding the economic
context where the company lies; the more stable the market is (this is the more stable
the historic data is and the more it is expected to be in the future), the more reliable
the model should be to the quantitative coordinate, since quantitative is based on the
historic; the more unstable the market is, the more reliable the model should be to the
qualitative coordinate; Φ should be biased accordingly. We also propose to qualitatively
modulate Φ based on the results of the calculated forecast; In other words, based
on the calculations result, we propose to validate the chosen Φ’s value. We can see a
potential Φ’s value in the next fi gure.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
75
Winter's Method
or Holt' s, etc.
Global
Aggregation
Delphi's
Method or
Opinion Jury
Low Aggregation
Level
High Aggregation
Level
SKU
Aggregation
Qualitative
Methods
Quantitative
Methods
Integration Constant
Axis (Φ)
Cons tant Φ
“Integration
Coordinate”
Figure 10. “Method Category-Aggregation Level Matrix” and the Constant Value (Φ)”.
Note that this integration ends up with a new coordinate, the “Integration
Coordinate”. This new coordinate represents a new forecast that is formed by a new
component along the Aggregation Level Axis and a new component along the Category
Method Axis. In practice, this concept is quite interesting since it is possible to profi t
from the Qualitative Methods calculation easiness (in low aggregation levels) and to
integrate the results of Qualitative Methods (in high aggregation levels and in terms
of revenues/sales). This concept allows us to integrate the Top-Down and Bottom-
Up concepts as Frazelle (2002) suggests. As stated by Bowersox (2002), Bottom-Up
methods develops SKU forecasts and then builds them into an aggregation demand
projection; The Top-Down approach develops a global forecast and then spreads the
volume at a SKU level based on historical patterns.
PROPOSED CALCULATION ALGORITHM
Based on the conceptual frame already presented, we present a calculation algorithm.
BOTTOM UP CALCULATION
As already discussed, the low aggregation levels will be defi ned in terms of “SKU
families” or Fsku’s; Fsku’s should be chosen based on criteria such as complementary
products (products that complements each other in terms of demand), demand
correlation, demand substitution, and the convenience of aggregating products in
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
76
order to improve forecast’s precision and calculation easiness. Fsku’s could include
several SKU’s or even be composed of a single SKU. The general idea is to determine
little families, that because of product similarities, it is convenient to aggregate in a
single family. For example, in the case of a forecasting process for a Supermarket, it
is convenient to calculate as a family the forecast for products with similar behavior
such as is the example of sodas; in this case we can get a global forecast and later on
decompose it according to the historic data sales percentage of each soda brand. In
this case, we use aggregated forecast to counterbalance the fact that certain customers
search to buy one soda, and that it could be any of his preferred brands (product
substitution; complementary products). Determining Fsku’s is still considered as a low
aggregation level, since if we compare an Fsku’s within the multiproduct context, we
realize that this aggregation is small if we compare it with the total SKU’s quantity in
the multiproduct context.
Once we have defi ned the Fsku’s we “back-cast” the forecast (as we did in the
Distribution case showed in this article). We will have the Quantitative Method that
adjusts the best to each Fsku demand’s pattern, this will allow us to fi nd the best forecast
for each Fsku (we propose to call this forecast FORFsku ). Once FORFsku has been calculated
we will decompose it accordingly for each SKU. In this case, this decomposition will be
based on the SKU’s historic weight or historic percentage within the SKU family (Fsku).
In order to do this calculation we propose the following formula:
[1]
where:
wfsku: SKU’s demand weight factor within the SKU family (Fsku).
Dsku: SKU’s historic demand.
Dt: SKU family’s (Fsku) total demand.
Once we have calculated every wfsku we calculate the forecast for each SKU (we
propose to call it FORsku) with the following formula:
[2]
where:
FORsku: individual SKU forecast.
FORFsku: SKU family’s (Fsku) forecast.
wfsku: SKU’s demand weight factor within the SKU family (Fsku).
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
77
In this moment we have the forecast (FORsku) for each SKU that composes the SKU
family (Fsku). This procedure that we have already presented has to be done for each
SKU that is manufactured in the company. In the case of new products (new SKU’s), in
which no historical data is available, FORsku will be calculated using the most convenient
method (for example: through a qualitative method or simply extrapolate a forecast
from an existing similar product). This calculation will be not presented here and will
be considered as a future investigation to incorporate in our model.
Once we have all the FORsku’s we calculate the monetary aggregation of these
forecasts; we will call this aggregation AFORsku (SKU’s Forecasts Aggregation). AFORsku’s
calculation follows this formula:
[3]
where:
AFORsku: SKU’s Forecasts Aggregation in monetary terms.
ρsku: SKU’s selling price for each one of the n SKU’s.
AFOR
sku
(SKU’s Forecasts
Aggregation)
Qualitative
Methods
Quantitati ve
Methods
Low Aggregation Level;
Bottom-Up
High Aggregation Level;
Top-Down
Winter's Method
or Holt' s, Moving
Average, etc.
F
sku
Aggregation
Figure 11. Bottom-Up Calculation and AFORsku. “Method Category-Aggregation Level Matrix”
TOP-DOWN CALCULATION
Once we have set AFORsku, we proceed with the Top-Down Calculation. As stated
before, this calculation will consider the use of Qualitative Methods. In our application
case, the methods chosen to estimate global forecast were: Opinion Jury, Commercial
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
78
Personnel Proposition and Market Research. This estimation was done in terms of
revenue and globally speaking (all SKU’s aggregated); note that in our application case,
managers and experts had the expert criteria to estimate forecast, globally aggregated
and expressed in monetary terms (since they have built their know-how during years
analyzing global revenues, not product units).
The more decomposed the Qualitative estimations, the more precise the forecast
could be; but, a higher amount of macro-families result in a more expensive forecast,
since Qualitative Forecasting is an intensive time consuming activity (in man-hours).
Qualitative Forecasts go with higher aggregation levels, so this is another constraint to
consider, since executives feel comfortable guessing for higher aggregation levels and
not in lower levels; this is what we call aggregation level trade-off. It is not an objective
of this article to present how to calculate forecast with Qualitative Methods, but to
show how to apply them.
Once the global estimation has been made (we will call it Global Forecast or GFOR),
we will include it in the Method Category-Aggregation Level Matrix; please refer to the
next fi gure.
Opinion Jury,
Commerci al
Pe rsonne l
Proposition
and Market
Research
Qualitative
Methods
Quantitative
Methods
Low Aggregation Level;
Bottom-Up
High Aggregation Level;
Top-Down
Winter's Method
or Holt' s, Moving
Average, etc.
F
sku
Aggregation
Global
Aggregation
G
FOR
(Global
Forecast)
AFOR
sku
(SKU’s Forecasts
Aggregation)
Figure 12. Top-Down Calculation and GFOR. “ Method Category-Aggregation Level Matrix”
BOTTOM-UP AND TOP DOWN INTEGRATION
Once we have the AFORsku and GFOR coordinates, we proceed to integrate these two
along the “Integration Constant Axis (Φ)”. This integrated coordinate will be called
Global Integrated Forecast or GIFOR. In order to integrate these coordinates we propose
the next formula:
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
79
[4]
We will present a GIFOR using an Φ=0.95 in the following fi gure:
Opinion Jury,
Commerci al
Pe rsonne l
Proposition
and Market
Research
Qualitative
Methods
Quantitative
Methods
Low Aggregation Level;
Bottom-Up
High Aggregation Level;
Top-Down
Winter's Method
or Holt' s, Moving
Average, etc.
F
sku
Aggregation
Global
Aggregation
G
IFOR
(Global
Integrated Forecast)
Integration Constant
Axis (Φ)
Cons tant
Φ=0.95
AFOR
sku
(SKU’s Forecasts
Aggregation)
G
FOR
(Global
Forecast)
Figure 13. Bottom-Up and Top-Down: GIFOR coordinate. “Method Category-Aggregation Level Matrix”
Now, the Global Integrated Forecast (GIFOR) must be decomposed in individuals SKU
Integrated Forecasts (or IFORsku’s). In order to have this GIFOR decomposed we propose
the following formula:
[5]
ALGORITHM APLICATION
Once we had the algorithm and its equations, we proceeded to apply it to a
Manufacturing Enterprise that produces over 250 different fi nished products (SKU’s).
The model was run using the historical data related to 3 years of sales history and
proceeded to forecast a six month period (from October 2005 to march 2006).
We started to set the different SKU’s families (Fsku’s) and to “back-cast” its future
sales with Quantitative Methods; according to equation (1) and (2) we calculated FORsku
for all of the 250 SKU’s. Later on, using (3) and a global level we calculated AFORsku.
Using Qualitatives Methods and higher aggregation levels, we defi ned GFOR. Using (4) we
calculated GIFOR and then applying (5) we obtained IFORsku.
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
80
When using (4) it is necessary to defi ne the Φ’s value. In our case we did set it
as Φ=0.95 since expert criteria led us there because of the economical context of
the company and its economical expectative; since the economical context has been
changing (unstable), managers think that historical data should have little impact in
the global prediction and qualitative methods should have more impact; note that even
when Φ’s value gives AFORsku light weight, AFORsku dictates the IFORsku’s sales curve
form when using (5).
Among the global results of the algorithm applied to the enterprise we have an
average error decrease from 80% for the AFORsku to 6.2% for the GIFOR. On the other
hand, it is logical to think that when comparing GIFOR error to GFOR error (the 6.2%
error for the GIFOR to 2.9% error for the GFOR), GFOR has a smaller error due to its global
aggregation level. Nevertheless, note that GIFOR allows us to smooth the possible error
related to the qualitative forecasting since it considers sales’ history weight; even if
error is slightly bigger than GFOR, GIFOR allows us to “hear” the historic demand pattern
and include it in the forecast’s calculation.
In the Manufacturing context, the forecast decomposition plays an important roll
since it is critical for the planner to know the forecasted or estimated quantities for
each SKU. So, in this sense, decomposing GIFOR into IFORsku is of great value, since this is
useful data for the planner. Even if we loose forecast precision, due to a decrease of our
aggregation level, decomposing is a must for the Manufacturing operation.
Regarding the algorithm’s results at a decomposed SKU level (this is comparing
IFORsku level vrs FORsku level), we propose to analyze both forecasts in monetary terms.
In order to evaluate the convenience of this algorithm we will contrast the cost of
using the IFORsku’s with the cost of using the FORsku’s. In order to contrast these two
forecasting methods, we propose to quantify the value of each method in terms of
cost. Each method will be compared to the real sales for the forecasted periods; please
note that at this moment we know the exact sales quantities for the forecasted period
(October 2005-March 2006). To quantify the cost of each forecasting method we propose
to consider the over-forecasting cost (overstocking) and the under-forecasting cost
(stockout). We propose to consider the overforecasting cost as the monthly carrying
cost, and the underforecasting cost as the stockout cost related to the monthly lost
sales (in terms of the lost earnings or lost margin related to the products not sold). This
procedure helps us to evaluate forecasting methods considering the Manufacturer’s real
situation (in terms of inventory carrying costs and sales loss).
So, assigning the positive forecast error to MCCR and the negative error to the SoCR
we have:
[6]
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
81
where:
FMC: Forecasting Method Cost.
pesku: positive error in units for a certain SKU.
MCCR: Monthly Carrying Cost Rate.
nesku: negative error in units for a certain SKU.
SoCR: Stockout Cost Rate
ρsku: SKU’s selling price.
After comparing both costs, we discovered that the algorithm yields an average cost
reduction comparable to a 6.6% of the earnings margin of the product (6.6% out of 18%
as the earnings margin), which is quite attractive. Note that the company’s MCCR and
SoCR values used were around MCCR= 1% and SoCR=18%.
As presented, this algorithm allows the Manufacturer’s analyst to calculate the
Supply Chain’s forecast for all the items that must be fabricated in order to be, later
on, Distributed along the Supply Chain in order to be available to fi nal customer.
As presented, this case considers a global aggregation for the whole enterprise when
applying Qualitative Methods; it is evident that the same algorithm can be applied
to Multiproduct Environments but at a lower global aggregation level. In our case,
250 SKU’s permitted us to aggregate them in a global prediction. When using this
same principle, but in a company with a larger quantity of SKU’s, we could decompose
the totality of products into strategic ensembles that could be treated as targets to
calculate GFOR and AFORsku and later on get a GIFOR; in this sense, we would use the
“Method Category-Aggregation Level Matrix” concept to each of the strategic ensembles
within the enterprise, and later on aggregate its results; we could see it as a “company
within a company” treatment.
BUSINESS STRATEGY AND FORECASTING
Through the Distribution and Manufacturing Process, companies should materialize
its Business Strategy, since product availability (in terms of quantity and place) is
essential to satisfy Customers. Distribution is seen as the latest step, supplied by the
Manufacturing step.
Forecasts can be used as a tool to produce and allocate product to each Customer.
According to Carranza (2004), forecasting processes are much more effective when
they are performed in collaboration with the entire Distribution Network than when
they are individually calculated by each S and C actors; they can be used to strengthen
the Supplier and Customer relationships. This collaboration is not natural between
members, since it consumes time and energy to do it. Although it is diffi cult, some
businesses have realized about its importance, since improvements in Forecasting and
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
82
Planning have had signifi cant success, as stated by Chopra et al (2004).
We propose that the implementation of a collaborating forecast is related to the
Negotiating Force of the S and C actors. This force difference will also determine Supply
Policies. The following are three types of possible relationships based on the different
Negotiating Force between Suppliers and Customer when negotiating Supply Policies:
Supplier Negotiating Force Superiority over its Customer.i.
Customer Negotiating Force Superiority over its Supplier.ii.
Supplier and Customer Negotiating Force Parity.iii.
NEGOTIAGING FORCE SUPERIORITY OVER ITS CUSTOMER
In this case, the Supplier will set the guidelines according to what is convenient for
him, for example:
Supplier will control • Lead Times by pushing his customers to place purchase
orders with as much possible time in advance. Doing so, Supplier will increase the
precision of his forecast (since he will produce “make-to-order”). This practice
will help him reduce his operative costs.
M.O.Q’s policies (Minimum Order Quantity) will be implemented so that the •
Supplier could profi t from the production and transportation economies of
scale. Suppliers sometimes pay for the transportation cost as a Customer Service
Policy, but their main objective is to force the customer to place M.O.Q’s Purchase
Orders. All these policies should be tacitly accepted by the market and customers;
otherwise they become counterproductive as a risk of potential market loss.
Supplier will try to push to its Customer the Economical Inventory Risk related •
to Forecasted Sales; S will try to push the product to C at the earlier possible
moment.
Supplier will not be worried to develop and to train its Customers with Forecasting •
Tools and Supply Policies in order to optimize the Chain. The interest is unilateral
and S makes decision aiming his local optimal point. Sometimes, this policy
could yield short term profi ts but later on long term losses (so is the case when
the Supply Chain gets saturated due to supplier and customer communication
problem; the Beer Game is a parody related to this problem as evoked by Carranza
(2004). These communication problems could be very expensive for Suppliers
since its Production Capacity has to be changed accordingly.
Supplier S will offer a slightly better Customer Service Level in terms of his •
competitor’s Service Level. The Strategy would be to differentiate from
competitors but not completely exceed them. This is how S will avoid his
Customers to place purchase orders to the competition. For example, a Customer
will prefer a Supplier that offers him the possibility to demand partial and
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
83
immediate shipments, with shorter lead times and the same quality (this is an
example of a differentiating strategy). Please see Figure #14.
C
a
C
d
C
c
C
b
S
X (
in lots)
PUSH
Notes:
1. The lenghts of the so lids arrows are
according to the del ivery Lead Time.
2. The lenght and direction of the doted
arrows indicate th e size and direction of the
Stock Holding and Forecastig risks.
PUSH
PUSH
PUSH
Figure 14. Basic Supply and Distribution Network: Supplier Negotiating Force Superiority over its
Customer
CUSTOMER NEGOCIATING FORCE SUPERIORITY OVER ITS SUPPLIER
In this case, the Customer will set the guidelines according to what is convenient
for him. For example:
Customer will prefer his Suppliers to follow • Just in Time supply policies. Since its
commercial advantage allows him to exploit the equation service, the customer
will aim to have the product at the Right Time, in the Right Place, and in the
Right Quantity (an example could be the supermarket sector and its relationship
with its suppliers). Since Suppliers should react immediately, this makes them
deal with all the Forecasting and Planning burden. This practice pushes the risks
towards Suppliers. Just in Time orders are characterized by its small sizes and
high frequencies due to short lead times.
Customer will foster his Supplier proximity in order to guarantee its product •
supply and fl exibility even under strong demand changes. In some cases, C will
foster S physic proximity in order to minimize the transportation time (classic
example of the automobile industry).
The economical inventory holding risk will be pushed toward S. It is a frequent •
practice for the biggest C’s, to make its Suppliers to carry a fi xed physical Safety
Inventory in order to guarantee an agreed Service Level (this is usually done
under economic penalty conditions for not fulfi llment cases). This penalty
pressure makes the Supplier to have a bigger need for Forecast accuracy or
higher Safety Inventory levels.
Another Suppliers strategy is to guarantee a Customer Portfolio that allows S to •
supply many other customers with reasonable size (as the Cd, Cc and Cb case in
fi gure #15). This allows S to equilibrate the higher economic pressure that the
biggest C puts on him (see below fi gure).
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
84
Ca
C
d
C
c
C
b
S
PUSH
PULL
PUSH
PUSH
X´s in small and
frequent lots
Notes:
1. The lenght of the solids arrows are
according to the del ivery Lead Time.
2. The lenght and direction of the doted
arrows indicate th e size and direction of the
Stock Holdin g and Forecasti g risks.
Figure 15. Basic Supply and Distribution Network: Customer Negotiating Force Superiority over
its Supplier
SUPPLIER AND CUSTOMER NEGOTIATING FORCE PARITY
In the case of Negotiating Force Parity, both actors will try to set the guidelines
according to what is convenient for them, for example:
Both actors will be interested in mutual growth. •
Mutual coordination will be aimed in order to set the Supply Policies that works •
the best.
The Negotiating Force Parity condition could come from many possible sources, for
example:
Negotiating Force Evolution through time for one of the actors.i. For example:
aggressive Customer requirements (costs reductions, shipping conditions, etc.)
sometimes make small and medium suppliers go bankrupt, or to “merge” with
a stronger actor (or even to sell the company). Later on, the market that these
competitors used to own, is absorbed by the strongest “survivor” Supplier who
now gains Negotiating Force toward Customers.
Negotiating Force gains due to a Strategic Advantage. ii. For example: a big
Customer wants to develop a strategic Supplier in order to guarantee his
requirements supply such as: quality level, physical proximity, supply fl exibility,
technological advantage, etc. In this case the Supplier gains Negotiating Force.
FORECASTING AS A COUNTERBALANCE FOR NEGOTIATION FORCE DIFERENCES
These three scenarios highlight the pressure that each actor has. We can compare
this pressure to the Implied Demand Uncertainty concept since, according to Chopra
et al. (2004), it “is the resulting uncertainty for only the portion of the demand that
the supply chain must handle and the attributes the customer desires”. This pressure
based on the Implied Demand Uncertainty could also come from the differences in
Negotiating Forces.
Nowadays we can hear from collaborative planning techniques such as CPFR
(Collaborative Planning, Forecasting, and Replenishment); these techniques have been
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
85
successfully implemented in Negotiating Force Parity situations, since both actors are
truly interested in mutual benefi ts, which motivate them to allocate their resources to
this project. Chopra et al. (2004) shows some of this examples.
In many cases, Force Superiority can not be exploited by stronger actors in a
sustainable way without considering the long term impact over the weaker actor
(especially if the weaker actor can fi nd an advantage in order to be considered by
the stronger as a critical strategically speaking actor). The weaker actors could profi t
from this fact and use it as an argument in order to negotiate and foster teamwork to
improve the Supply and Distribution Network.
Our model proposes the importance of using simpler collaborative techniques
in the Negotiating Force Non- Parity environments; in this sense, the weaker actor
requires to improve its products supply management through a forecasting process
improvement, as is the case of the Method we are proposing, and therefore reduce its
pressure or Implied Demand Uncertainty. This improvement can help the weaker actor
to counterbalance its Negotiating Force by being proactive with the stronger actor and
fostering a collaborative environment to improve the service the weaker offers. We
propose that this initiative must come from the weaker actor; a possible tool for weaker
actors to reach this is through the use collaborative Forecasting Process.
Within the reality of the Supply Chain, since companies usually have different
suppliers and customers, companies play different Negotiating Force rolls; in this
sense, companies could play the weaker or stronger actor roll depending on each
case. When Planning the Multiproduct Supply Chain, it is evident that each company
has to concentrate in the most important of these relationships; a Paretto analysis is
recommended in this situation.
CONCLUSION
Distribution and Manufacturing Strategic Planning is critical for companies that
deal with Manufacturing and Distribution processes. Both processes have to be planned
in accordance one with another.
Forecasting processes can be implemented in the Distribution network in order
to guarantee product availability by improving the product allocation process within
the Supply Chain (within each Supply Chain’s node). Forecasting processes can be
implemented in the Manufacturing process in order to improve the availability of
product to supply the custumer’s needs in terms of quantity and place.
When planning the Manufacturing and Distribution processes it is critical to consider
the company’s position within the proposed Negotiating Force frame; the company must
understand its position and try to improve it strategically. Customer’s or Supplier’s
pressure can be handled and reduced through forecast as a step to reach collaborative
forecast. Since in the Multiproduct context there are multiples customer-to-supplier
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
86
and supplier-to-customer relationships, each company has to understand which of
these relationships represents the critical ones in order to strategically improve them.
REFERENCES
Bowersox, D. Closs, J. Cooper, M. “Supply Chain Logistics Management”. Mc Graw Hill
Irwin. Edition - 2002.
Carranza, O. “Logística: Mejores Practicas en Latinoamérica”. Thomson. Edition-2004.
Frazelle, H. “World-Class Warehousing and Material Handling”. Logistics Resources
International. Logistics Management Library, Edition - 2002.
Kaplan, R; Norton, D. “The Balanced Scorecard”. Gestión 2000, Edition - 2002.
S. Meindl, P. “Supply Chain Management, Strategy, Planning and Operations”. Pearson
Prentice Hall. Edition - 2004.
J. Heizer, B. Render. “Dirección de la Producción, Decisiones Estratégicas”. Pearson
Education, 2001.
E. Silver, R. Peterson. “Decision Systems for Inventory Management and Production
Planning”. John Wiley and Sons, 1985.
Biography
Jose F. Roig Zamora is an associate professor of the Industrial Engineering School
of the University of Costa Rica, also partner of the Logistics Studies and Applications
Center at the National University of Cuyo, Argentina. He received Master degree in
Logistics from the National University of Cuyo (Argentina) and his Master of Science
degree in Industrial Systems Innovation Engineering from the Ecole Supérieure
National de Génie de Systèmes Industriels- INPL (France), both in 2006. His research
interests include Supply Chain Management and Logistics applied to industry
problems.
Raymundo Q. Forradellas is a professor of Information Systems at the Industrial
E ng i ne er in g S c ho ol (E ng i ne er i ng Fa c ul ty, at th e N at io na l U niv ers it y o f Cu yo ), Di r ec t or
of the Masters in Logistics and Director of. Logistics Studies and Applications Centre.
He received his PhD degree on Artifi cial Intelligence from the Polythecnic University
of Madrid, Spain. His research interests include applied artifi cial intelligence systems,
planning & scheduling, logistics and industrial systems.
Mauricio Camargo is associate professor on Management of technology and
innovation at the Ecole Nationale en Génie des Systèmes Industriels of Nancy (The
Industrial Engineering School of the National Polytechnic Institute of Lorraine
Brazilian Journal of Operations & Production Management
Volume 4, Number 1, 2007, pp. 61-87
87
-France). He is a B.S. Chemical Engineering graduate of the Universidad Nacional
de Colombia and held a PhD. on Automatics of industrial and human systems at the
Universite de Valenciennes et de Hainaut Cambresis. His main research interests
are in the fi eld of new product development, cost estimation models, design-to-cost
and technology forecasting. His late research concerns application of Multi objectif
Evolutionary techniques to evaluate product performances and decision making at
the design stage.
