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.