¿²¼ Ñ°»®¿¬·±²-
Ü»³¿²¼ ½¸¿®¿½¬»®·¦¿¬·±² ¿¬
Ú®·»-´¿²¼ Ú±±¼- Í«°°´§ б·²¬
Ô»»«©¿®¼»²
Ô»²²»®¬ ¼» Ù®¿¿º
˲·ª»®-·¬§ ±º Ù®±²·²¹»²
Organization: Friesland Foods Supply Point Leeuwarden Company supervisor: Karst Groothoff CPIM
RUG supervisor: Dr. D.P. van Donk
The link between Sales and Operations
Demand characterization at Friesland Foods Supply Point Leeuwarden
- Public version -
Author:
Lennert de Graaf University of Groningen (UG) Faculty Economics & Business
MSc. Business Administration, specialization Operations & Supply Chains
August 2008
Hoendiep 51 9718 TC Groningen g.l.de.graaf@student.rug.nl
Student number: 1660799
I PREFACE
This thesis is the final part of my education as a Master of Science in Business Administration, with the specialization Operations & Supply Chains. Furthermore, it is not only the last part of my Master education; this thesis means also the finish line of a total of nineteen years of education (at least for now). In some way, finishing my Master is a pity, whereas my grades on exams were following an upward trend! Who knows what grades would have followed up after my last exam…!
The finish of this master thesis is more than the end of my education. It is also the end of the first part of my life. The part of my life of studying, holidays, student jobs and more comes to an end and will be followed up by a working life.
Performing this thesis project at Friesland Foods - Supply Point Leeuwarden has been a very interesting and educational time. Not only have I been able to learn more about the Frisian culture (Skûtsjesilen and Frisian Flag coffee enrichers), I can now also understand some Frisian language! Furthermore, it was great to experience the friendly and informal company culture at Friesland Foods, which should be a great example of how a company culture should be.
It was also a great experience to be able to recognize and apply theories at Friesland Foods, which I learned at the University of Groningen. The research gained me a lot of insight on the structure and processes of Friesland Foods, one of the largest original Dutch companies (number 36 on the FEM Business Magazine top 500).
I would like to show appreciation to my supervisors of the RUG, Dirk-Pieter Van Donk and Tim van Kampen for their critical but positive and useful feedback. Also great thanks go to my supervisors of Friesland Foods, Karst Groothoff and Koen Vos for their information, help and support during this project.
I would also like to thank Jouke de Vries (the one and only Excel expert) for his advanced Excel-lessons. In addition, my thanks go out to my Belgian ‘colleagues’ of the consultancy company International Business Consultants (IBC); Veerle Stoffelen and Guy van Zeelen.
I would like to thank them for keeping my fluid level on a proper level (serving coffee) but moreover for their support and tips.
Finally yet importantly, most thankfulness goes out to my girlfriend Eva, who is the most important person in my life and was of great support and inspiration during this period.
Lennert de Graaf August 2008, Groningen
EXECUTIVE SUMMARY
___________________________________________________________________________________________________________________________________________________________________________________
II EXECUTIVE SUMMARY
This research aims at characterizing demand of Friesland Foods Supply Point Leeuwarden.
After characterizing demand, a policy is developed how the company should handle the different types of demand.
The research question on which this research is based is as follows:
‘How can the demand of Friesland Foods Supply Point Leeuwarden be characterized and which production strategy should be applied to each product, taking its demand characteristics into account?’
The model created for characterizing SPL’s demand uses the three variables ‘total shipped volume’, ‘shipped in n months’ (a measure for regularity) and ‘variability in shipped volumes’ (a measure for stability).
The model categorizes the products into the three production strategies SPL applies. These production strategies are make to order (MTO), make to stock (MTS) and make to forecast (MTF). The MTO production strategy is applied to products with unstable and irregular demand, the MTS production strategy is applied to products with predictable demand, but small volumes (clustered to make an efficient production run), and the MTF production strategy is applied to products with predictable demand and large volumes. The figure below (figure 1) shows the theoretical decision tree for production strategies.
SKU
Stable
Regular Yes
Large volume
MTO Yes
No
No
MTS MTF
Yes No
Figure 1. Theoretical decision tree production strategies
As the model above shows, stable, regular and large volume products are categorized as MTF products. Products that show a stable and regular demand pattern, but are low volume products, are categorized as MTS articles. Products with unstable and/or irregular demand patterns are categorized as MTO articles. After this categorization, all products that fall into the MTO category with shipped volumes of more than four million kilograms are reassessed by checking their forecasts. This to make sure that new products, which are not marked as regular, can also be categorized as MTF products. The final categorization is shown in the table below (figure 2).
Strategy Number of products
MTO X*
MTS X*
MTF X*
As the table shows, (…)* production strategy. These products show too irregular and/or unstable demand patterns to be produced to stock or to forecast. *) Classified
Finally, several recommendations came up during and after this research. The recommendations are as follows:
• The advice to SPL is to use the new categorization model on annual base, in order to control the decisions on production strategies more consequently. The model provides a basis for the supply chain department of SPL to base decisions on production strategies in a structured way.
• The order policies that come out of the categorization should be compared with the current policies on each SKU, to check whether each SKU has the appropriate production strategy.
• This demand characterization and SKU categorization can also be used to inform customers on their past demand, and try to move them into the direction that is ideal for SPL, for example by smoothening demand.
• The supply chain department should investigate whether this categorization can be implemented in the new SAP environment, to make future categorizations easier.
Furthermore, SPL should try to save the information about initially requested amounts of products, with the use of the new SAP software.
• Finally, the supply chain department of SPL cannot easily change production strategies on products. This because they try to maintain a high degree of customer satisfaction. Therefore, the company should try to cooperate with customers to improve demand patterns. If the demand pattern does not improve, SPL can take action by changing the production strategy.
GLOSSARY
___________________________________________________________________________________________________________________________________________________________________________________
III GLOSSARY Abbreviation Explanation
BU Business Unit. The structure of Royal Friesland Foods exists of a number of business units. These business units can be production locations and sales locations. The production locations that are discussed in this research are Supply Point Beilen and Supply Point Leeuwarden. The sales locations discussed in this research are FRL Export, FRL Hellas, FRL Hong Kong, FRL Middle East, FRL Wamco and FRL West Africa (FRL is the abbreviation of Friesland Foods).
COV Coefficient Of Variation. This measure is used in this research to indicate stability. It is calculated by dividing the mean of shipped volumes per year by the standard deviation of these shipped volumes.
EVAP Evaporated Milk is one of the products that Supply Point Leeuwarden produces.
MTF Make-to-forecast is a production strategy based on forecasting.
Products are produced on forecast and later on matched with actual customer demand to ensure short lead times.
MTO Make-to-order is a production strategy based on customer orders.
Production will only start when actual orders on the product arrive at the sales department of the company.
MTS Make-to-stock is a production strategy based on efficiency. Large batches are produced in one production run to eliminate start-up and cleaning time.
SCM Sweetened Condensed Milk is a derivation of Evaporated Milk.
Sweetening ingredients are added to the milk to give it a sweet taste.
SKU Stock Keeping Unit is a unique product with a unique article number.
SPB Supply Point Beilen is one of the two production business units of Friesland Foods Consumer Products, located in Beilen (The
Netherlands). The business unit produces milk products in powder form.
SPL Supply Point Leeuwarden is the other production business unit of Friesland Foods Consumer Products, located in Leeuwarden (The Netherlands). It produces evaporated milk and evaporated milk related products.
IV TABLE OF CONTENTS
I PREFACE 2
II EXECUTIVE SUMMARY 3
III GLOSSARY 5
IV TABLE OF CONTENTS 6
1. INTRODUCTION 9
1.1 Company description 9
Supply Point Leeuwarden. 9
Production process at SPL. 9
Product range. 10
Industry characteristics. 10
1.2 Problem statement 10
Amount of SKUs. 10
Delivery to business units. 10
Production strategies. 11
Initial problem statement. 12
1.3 Research objective 13
Objective. 13
Scope. 13
1.4 Research question 13
Theoretical part. 13
Practical part. 14
Analysis part. 14
1.5 Research methodology 14
Literature research. 14
Interviews. 15
Data search. 16
Data analysis. 16
2. DEMAND CHARACTERIZING METHODS 17
2.1 Methods 17
2.2 First analysis 18
Variation of demand sizes. 19
Volume. 19
Demand pattern. 19
2.3 Selection criteria 19
Simplicity. 19
Overview. 19
Aggregation level. 20
MS Excel compatible. 20
Statistically based. 20
Production strategy decision tool. 20
2.5 Further analysis 20
Unusable methods. 20
Usable methods. 21
TABLE OF CONTENTS
___________________________________________________________________________________________________________________________________________________________________________________
Important criteria. 22
2.6 Requirements of SPL 23
2.7 Final method 23
Demand pattern. 23
Volume. 25
Decision tree. 25
3. DEMAND CHARACTERIZATION 26
3.1 Dataset 26
Unit of measure. 26
Scope. 26
3.2 Demand characteristics and trends 26
General characteristics. 26
Trends. 28
Bullwhip effect. 28
3.3 Determination of limits on categorization 29
Regularity and stability of demand. 29
Demand volumes. 30
Final categorization method. 32
4. ANALYSIS ON PRODUCTION STRATEGIES 33
4.1 Production strategies 33
First categorization. 33
Exceptions on categorization. 33
Final categorization. 34
4.2 Unstable products 35
5. CONCLUSION 36
5.1 Conclusion 36
5.2 Recommendations 37
Future categorizations. 37
Comparison with order policies. 37
Awareness at customers. 38
SAP implementation. 38
Production strategy decisions. 38
5.3 Discussion 38
Shipped volumes representing customer demand. 38
Shortcomings on the model. 38
6. REFERENCES 39
6.1 Articles 39
6.2 Other references 41
APPENDICES 42
1. Categorization methods 43
2A. Trend of total and per sort 44
2B. Trend per BU 45
2C. Trend per product line 46
3A. Cumulative kilograms EVAP 47
3B. Cumulative kilograms SCM 48
4. Decision tree production strategies 49 5A. Final categorization of all ‘active’ SKUs (part 1/2) 50 5B. Final categorization of all ‘active’ SKUs (part 2/2) 51
6. Manual for future categorizations 52
1.INTRODUCTION
___________________________________________________________________________________________________________________________________________________________________________________
1. INTRODUCTION
This part of the research deals with the research methodology used for this thesis. First a description of the company is given, after that follows the problem statement, the research objectives, the research question and the research methodology.
1.1 Company description
Friesland Foods is the Dutch part of Royal Friesland Foods, which developed itself in the past 125 years into a leading producer and supplier of dairy products, fruit-based drinks and ingredients. The company has strong market and brand positions in Western and Central Europe, Southeast Asia, West Africa and the Middle East. The company has 14,600 employees worldwide and revenue of 5.1 billion euro’s in 2007. Royal Friesland Foods has business units (BUs) located all over the world.
Supply Point Leeuwarden. One of Royal Friesland Foods’ BUs is Friesland Foods Supply Point Leeuwarden (SPL). This BU is specialized in long-life dairy products like evaporated milk and coffee and tea enrichers. The products of the BU are sold all over the world, mainly in cans, but also in cups, bottles and bags-in-boxes.
SPL is Friesland Foods’ largest manufacturing site, which annually produces over 1.5 billion consumer units out of more than 800 million kilograms of milk. The site has over 700 employees working in day- and nightshifts, producing approximately X* different Stock Keeping Units (SKUs). *) Classified
Production process at SPL. In the figure below (figure 1.1), the production process of SPL is shown. Underneath the figure follows a description of the process.
Figure 1.1 Production process of SPL
The production process of SPL starts with the daily delivery of approximately X* cubic meters of milk, which is collected at a large number of farmers. After testing every batch, the milk is processed into the required products as specified in one of the many receipts (propack processing). When the recipe is ready, it is put into the required package (for example can, bottle or cup). The process above can be seen as the primary process of the company; processing milk into different end products like evaporated milk. *) Classified
Next to the primary process, the company also has a large-scale production line of cans to be used for packing the end products. Here, large rolls of steel are processed into different cans varying in size from 78 gram cans up to 410 gram cans.
Product range. The two largest sorts of products SPL produces are Evaporated Milk (EVAP) and Sweetened Condensed Milk (SCM). These two product ranges are responsible for approximately X* percent of the total sold products. The figure below (figure 1.2) shows the packing volumes per sort. *) Classified
397g 410g
78g 170g
BULK
SCM EVAP
1000g
Figure 1.2. Packing volumes per sort
As the figure shows, SCM products are mainly packed into 78, 397 and 1,000 gram cans.
EVAP products are mainly packed into 170 and 410 gram cans. Both SCM and EVAP products are also sold in large (bulk) volumes. Examples of packaging for large volumes are containers containing 1,000 kilograms of milk and jerrycans with 25 kilograms of milk. The EVAP and SCM products are produced in separate plants at Leeuwarden, and have separated planning and control processes.
Industry characteristics. The processing industry in which SPL operates has to cope with many difficult characteristics. As literature states (Soman, Van Donk and Gaalman, 2002), plant characteristics can be described as expensive capacity and sequence-dependent set- up and cleaning types between product types. Product characteristics can be described as variation in supply and quality of raw material, limited shelf life for raw material, intermediate products and end-products and many different end-products. Finally, the process characteristics can be described as variable yield and processing times.
1.2 Problem statement
In this chapter, the problem statement is described. First, the main difficulties are described, after that follows the initial problem statement.
Amount of SKUs. The company of SPL has to deal with many different Stock Keeping Units (SKUs). Whereas the company delivers it products to over 100 different countries, the company delivers in total approximately X* different products. All those products follow different demand patterns and have largely varying order sizes, which causes a large amount of demand data. *) Classified
The planning and forecasting of those X* SKUs is mostly based on historical sales data. Each BU generates its own forecast, which they deliver to the supply chain department of SPL on a monthly base. *) Classified
Delivery to business units. One of the main complexities of SPL is the structure of the company. SPL sells and delivers its products not directly to its end-users, but to different BUs of Friesland Foods. The BUs that SPL serves are Friesland Foods (FRL) Wamco, FRL Hellas, FRL Middle East, FRL Hong Kong, FRL West Africa and FRL Export. Table 1.3 shows the markets that the different BUs serve.
1.INTRODUCTION
___________________________________________________________________________________________________________________________________________________________________________________
Business Unit Serving market(s)
Friesland Foods Wamco Nigeria Friesland Foods Hellas Greece
Friesland Foods Middle East Middle East countries (like Iraq and Egypt) Friesland Foods Hong Kong Hong Kong
Friesland Foods West Africa West African countries (like Ivory Coast and Ghana Friesland Foods Export Remaining customers
Table 1.3. Business units SPL serves
As can be seen in the table above, SPL has to deal with six different BUs. Whereas the five upper BUs have a rather straightforward market to serve, the BU FRL Export serves markets all over the world, which makes it more complicated. This BU has to deal with problems like cultural problems, payment problems, shipping problems and more.
(…)*. *) Classified
Production strategies. SPL applies several production strategies, which are Make-to- order (MTO), Make-to-forecast (MTF) and Make-to-stock (MTS). The figure below (figure 1.4) shows the three production strategies and their general characteristics (derived from Meredith & Akinc, 2007).
Make to order
Make to forecast
Make to stock
Responsiveness and risk Customization
Low High
High
Figure 1.4. Production strategies
The figure shows the three production strategies and their general characteristics, they are explained below.
In the MTO situation, the company produces only when a customer orders the product (Meredith & Akinc, 2007). In this situation, the customer has to deal with long lead times. A long lead time is a low responsiveness, because it takes a lot of time for the company to react on the order of the customer, from production to delivery. In the case of SPL, the company its customers have a production lead time of approximately eight weeks. Next to the production lead time, some customer have to deal with long transportation times.
However, these transportation times are depending on the location of the customer.
Customers far away from the Netherlands can have a transportation time of a few weeks because a ship with the goods takes quite some time to cross the ocean(s). The advantage for SPL on applying a MTO production strategy is the low degree of risk the company has to deal with. SPL has no risk of stock keeping, because products are produced and delivered
immediately. The company applies the MTO production strategy to the products it sells, of which demand cannot be predicted adequately to be produced on MTF or MTS strategy.
The MTF production strategy uses a forecast to start the production. Later on, during production, customer orders are matched with the forecasted production. The customization degree is somewhat lower than in the MTO situation, because products are partially produced based on forecasts instead of pure customer orders. As the picture above indicates, responsiveness is higher than in the MTO situation. Responsiveness is higher because lead times are shorter in the MTF situation, in the case of SPL to a maximum of four weeks production lead time. However, the risk of a MTF production strategy is higher than in the MTO situation, because production starts based on a forecast. Whereas Nahmias (1997) states in his article that the first rule of forecasting is that a forecast is always wrong, SPL always runs the risk of a forecast which does not fit with the customer orders. In that situation, the company has to deal with so-called orphans (Meredith & Akinc, 2007). The orphans are the products, which are produced, based on forecast, but cannot be matched with the actual customer orders. Friesland Foods runs the risk on obsolete stock for these orphans. SPL applies the MTF production strategy to the products it sells, of which demand can be predicted adequately. On the products which are made to forecast, SPL does not run the risk of ‘orphan-products’. Furthermore, products made to stock are products which are sold in larger volumes and can be periodically produced in large and efficient volumes.
In the MTS situation, one produces products to stock, to ensure short lead times (high responsiveness) when customers order products. The risk in this production strategy is that stock becomes obsolete. The most important reason for SPL to apply a MTS production strategy is to cluster the demand for small volumes, to ensure an efficient production run.
Applying a MTS production strategy to the low volume SKUs, SPL can avoid a number of high setup times, because several runs of a SKU are clustered. This also avoids a part of the changeover costs. Therefore, the efficiency rate for the MTS product is higher. Another consideration for SPL to produce products on stock are customer agreements. With a number of customers, the company agrees to produce the products on stock to ensure short lead times for them.
Within the production planning of SPL, the application of production strategies has some consequences on how customer orders are treated. SPL has a forecast of all customers on how much of each product they predict to buy from SPL. These forecasts are adapted often by SPL’s customers as time passes. Furthermore, products with a MTF strategy are planned as the amount in the forecasts states. For MTS products, amounts on the forecasts for the coming periods are clustered and planned in one large batch. SPL also has forecasts of MTO products. However, these products are only planned for production when SPL receives an order for the product.
At this point, there is not a clear policy for applying one of the three production strategies.
The decisions about the strategies per SKU are at this time mainly based on common sense, instead of fixed policies.
Initial problem statement. The company of SPL has problems with keeping overview of all its SKUs. All products show different demand characteristics, which are not all well known by the supply chain department of the company. Because the supply chain department does not know all different demand patterns of the products, it cannot anticipate very adequately on these, when transforming the forecasts into a production
1.INTRODUCTION
___________________________________________________________________________________________________________________________________________________________________________________
planning. Furthermore, SPL now applies a production strategy (MTO/MTF/MTS) mainly based on common sense instead of a decision based on a clear policy.
1.3 Research objective
This chapter discusses the research objectives. After discussing the research objectives, the scope of this research is explained.
Objective. The main objective of this research project is developing a method to characterize and categorize the demand of SPL. The method has to categorize SPL’s demand into different categories. On each of these demand categories, a production strategy is applied which fits best to the demand characteristics of that category. The methods has to be a tool which the supply chain department of SPL can re-use in future.
Scope. This research will focus on the two largest product ranges of the company. These two product ranges are evaporated milk (EVAP) and sweetened condensed milk (SCM).
These two product ranges form approximately X* to X* percent of SPL’s sales. The decision on focusing on EVAP and SCM products is not only based on the fact that these two product groups form X* of SPL’s sales and production, but also (…). *) Classified
Furthermore, the method is only applied to the sales data of 2007. Only sales data of 2007 is used because older sales data is not relevant enough due to the quickly changing demand of SPL’s ‘fast moving consumer products’. The products with which SPL serves its markets, have short life cycles and can therefore be replaced relatively fast by improved and/or renewed products.
1.4 Research question
Based on the research objectives as stated in the previous chapter, the research question and sub research questions are created.
The research question for this research is as follows:
´How can the demand of Friesland Foods Supply Point Leeuwarden be characterized and which production strategy should be applied to each product, taking its demand characteristics into account?’
To conduct this research and create a proper answer for this research, a number of sub research questions are stated, clustered into three groups:
• Theoretical part; Discussion of papers and criteria;
• Practical part; Collecting data and implement the chosen method(s);
• Analysis part; Discussing the outcomes of the implemented models and advice on these outcomes.
The sub research questions and their outcomes are discussed below, divided into the three parts as mentioned above.
Theoretical part.
• Which are the different methods for characterizing demand?
Outcome: Spreadsheet with the found methods (short description and application).
• Which of those methods are applicable for the situation of SPL?
Outcome: Discussion on why methods can or cannot be used.
• Which is or are the best method(s) for characterizing demand of SPL?
Outcome: One or more methods that are best suitable for characterizing the demand of SPL and a detailed description why those methods are chosen.
• What does this method exactly look like?
Outcome: Detailed description of the methods, which are chosen to characterize the demand of SPL (which data is needed, exact purpose etc).
Practical part.
• How can the demand of the Stock Keeping Units (SKUs) be characterized?
Outcome: A characterization of SPL’s demand and an analysis how the categorization method should be used and implemented.
Analysis part.
• How should SPL handle the different demand categories?
Outcome: Implementation of the categorization method followed by the categorization of products into production strategies.
1.5 Research methodology
As stated in the previous chapter, this research consists of three main parts (theoretical, practical and analysis part). This chapter deals with the research methodology of the research.
The following table (table 1.5) shows the data collection method and the place for collecting the data for each sub question. Below the table, the methods of collecting data are explained in more detail.
Sub question Method of
collecting data Place of collecting data Which are the different methods
for characterizing demand? Literature research Books, articles (electronic databases library RUG)
Which of those methods are
applicable for the situation of SPL? Literature research,
interviews Books, articles, persons (as specified below)
Which is or are the best method(s)
for characterizing demand of SPL? Literature research, interviews, data analysis
All collected data from above mentioned inquiries
What does this method exactly
look like? Literature research Articles
How can the demand of the SKUs
be characterized? Data search, data
analysis Forecast spreadsheet, historical sales data of SPL
How should SPL handle the different demand categories?
Literature research, interviews
Literature about production strategies and general literature about handling demand
Table 1.5. Research methodology and planning
Literature research. The literature research consists of searching articles with theories, which can be applied in this research. During this research, the databases EconLit (EL), Academic Search Premier (ASP), and Business Source Premier (BSP) are used. These databases contain most relevant journals in the research field of operations, logistics and (supply chain) management, whereas the EL database contains approximately 785,000 publications, the ASP database contains 4,500 journals and the BSP database contains approximately 2,300 journals. Some examples of the journals included are Interfaces,
1.INTRODUCTION
___________________________________________________________________________________________________________________________________________________________________________________
Journal of the Operational Research Society, International Journal of Physical Distribution &
Logistics Management and International Journal of Production Economics.
During the literature research, several search keywords are used, varying in number from one to a maximum of three keywords per search query. To keep enough options open when a search is performed, the third keyword was mainly not used. The keywords are shown in the following table (table 1.6).
First keyword Second keyword
ABC Categorization, classification
Consumption Categories, categorization, characterization
Demand Case study, categories, categorization, characterization, chart, management, mapping, model, order-size, patterns, type, quadrants Method Categorization, characterization
Product Categorization, characterization, chart, mapping, order-size SKU Categories, categorization, characterization, chart, mapping Table 1.6. Search keywords
Articles are not only searched by using the databases of journals. In addition, reference lists of articles are used for forward and backward search. The principle of forward and backward search is depicted in the figure below (figure 1.7).
Figure 1.7. Forward and backward search
As the picture shows, the starting point is article A. This article has cited references B, C and D. Searching to those articles is backward search. The principle of forward search is searching for the articles which cite starting point article A. This method has been applied to the largest part of the articles which are discussed in the chapter of demand characterizing methods.
Interviews. Interviews are defined as planned meetings in which information is gathered on a structured way. The interviews are prepared adequately, to ensure that the right information is gathered at the end of the interview. A conversation is not the same as an interview, because it is unplanned and not prepared in advance. In the beginning of this research, several interviews have taken place, to gather general information about the company and the objectives of this research. The interviews are conducted with different people within SPL:
• Higher management: An interview with the Supply Chain Manager of SPL has been conducted to get a wide view of the organization and its objectives. In addition, an
Time Starting article A
Cited references:
B, C, D
B C D
Backward search
Article E
Cited reference:
Article A Forward search
interview with the Supply Chain Manager of BU FRL Export has been conducted, to learn more about the most complicated customer of SPL.
• Middle management: Interviews with the Supply Chain Master Planner and Manager Logistics gained a more detailed view on the planning and logistics situation of SPL.
• Other: Several other interviews have been conducted, with for example the Supply Chain Management Accountant, the Manager Customer Service and the Sales Planner of BU FRL Export. These functions represent several other viewpoints of the organization, which can be important for this research.
The interviews are conducted at different layers of the organization, as well as at different functions of the organization, to ensure that the research will not be too focused on one department, and only be in the interest of a small group. Whereas this research is focused on the supply chain department, it is important to know how other departments and people of the organization think about certain issues and processes. Another important factor to choose for the different layers and functions is the reliability of the information. Whereas higher management can have an ideal and strategic view, the lower management and other functions can be more realistic about certain issues and have a tactical view.
Data search. Data search represents the search for appropriate data, which can be used for this research. This data, on individual SKU level, consists of information about sales of SPL like order dates and order sizes from the past years. The data will be inquired with the help of the following software applications; Prism Logistics, Business Objectives (data analysis tool) and MFG/Pro (used by the order processors).
Data analysis. The data analysis is mainly about applying the chosen models on the datasets. To do so, the data is first structured into a clear and consistent spreadsheet. After structuring the spreadsheet, the chosen models will be applied to it.
2.DEMAND CHARACTERIZING METHODS
___________________________________________________________________________________________________________________________________________________________________________________
2. DEMAND CHARACTERIZING METHODS
This part is the theoretical part of the research. It describes all found methods which can be used to characterize demand. Most of these methods are based on papers, published in academic journals. After all methods are described, selection criteria are discussed and finally, one or more methods are chosen to characterize the demand of SPL.
2.1 Methods
This chapter deals with the different methods of characterizing demand. The twenty found characterization methods are shortly described below, in alphabetical order of the author(s) name(s). For a more detailed overview see appendix 1. The appendix shows the focus, business, technique and applied criteria per method.
1. Akkerman (2007); The method described in the thesis of Akkerman is used for characterizing demand of a case study dairy plant.
2. Chen, Chen, Lin (2004); The paper of Chen et al. describes a method to categorize and prioritize product features demanded by customers.
3. Chen, Li, Kilgour and Hipel (2008); The article of Chen et al. describes a model for a multiple criteria inventory policy with the use of a case based distance model. They use an illustrative case to show the model.
4. D’Allessandro & Baveja (2000); The paper of D’Allessandro & Baveja describes a method to analyze demand variability for a chemical processing plant.
5. Eaves & Kingsman (2004); The paper of Eaves & Kingsman describes a method for characterizing demand patterns of product. They use the case of spare parts in inventory of the Royal Air Force as an illustrative case.
6. Ernst & Cohen (1990); The research described in the article of Ernst & Cohen had the objective to cluster different SKUs into categories with the same characteristics.
Their operations related groups method (ORG method) supports strategic planning for the operations function of a company.
7. Fisher, Raman, and McClelland (2000); Fisher et al. describe their research in which they compare forecasts of specialists with forecasts based on historical sales data.
They plot products to show the forecast errors of both compared groups.
8. Flores & Whybark (1987); The researchers Flores & Whybark describe their method in which they extended the ABC-analysis to manage the inventory at a service organization and a consumer goods manufacturer.
9. Garcia-Flores, Wang and Burgess (2003); The demand characterization method described in the article of Garcia-Flores et al. is developed to improve the inventory management at a small chemical company in the UK.
10. Guvenir & Erel (1998); The method described by Guvenir & Erel is a new application of genetic algorithms which they developed to use in inventory management. They tested the method on two small inventory cases.
11. Huiskonen (2001); The article of Huiskonen describes the observation that many inventory managers do not understand difficult inventory management systems and that they therefore use their own assessment. Huiskonen emphasizes the need for an inventory policy to be split in different categories of products. These different categories are created with a categorization model.
12. Huiskonen, Niemi and Pirttilä (2003); The method described in the paper of Huiskonen et al. combines the production based information with customer characteristics of products to make decisions on inventory policies.
13. Johnston, Boylan and Shale (2003); The article of Johnston et al. describes the examination of half a million orders on their size at an electrical wholesaler. They characterize the products and develop a policy for slow moving inventory items.
14. Ng (2006); The inventory classification model developed by Ng is created to extend single criterion inventory classification to multiple criteria inventory classification by using linear programming.
15. Partovi & Anandarajan (2002); The artificial neural network approach (ANN), as described in the paper of Partovi & Anandarajan, is a method to classify inventory into ABC categories using multiple criteria.
16. Ramanathan (2006); The article of Ramanathan stresses the need for multiple criteria inventory classification instead of the widely employed ABC-categorization.
They describe a simple method of linear programming to apply as a multiple criteria inventory classification.
17. Schomer (1965); One of the first methods to manage a company’s inventory is the ABC-analysis, as described in the article of Schomer. Schomer stresses the need for a clever inventory policy for relatively small companies. This because these companies do not have the resources to invest in large inventories.
18. Syntetos, Nikolopoulos, Boylan and Fildes (2008); The article of Syntetos et al.
describes the research on forecast adjustments. They plot demand to distinct different types of demand in their pharmaceutical research case.
19. Talluri, Cetin and Gardner (2004); The method for characterizing demand, described in the paper of Talluri et al. tries to help managers making policies for inventory management. The demand and supply variability are plotted to decide on safety stock policies for a pharmaceutical case-company.
20. Williams (1984); The method described in the article of Williams helps managers to decide on a method for stock control. Deciding on a method for stock control depends on the demand distribution, which they therefore categorize in their method.
2.2 First analysis
A first remark which can be made is that the theory about the characterization of demand is not very extensive. Indeed, half the amount of the twenty articles mentioned above, are more inventory focused rather than demand focused. The choice to also mention the inventory-focused methods is based on the fact that inventory management is mainly about demand issues. Inventory management and policies are for a great part determined by demand patterns.
Whereas the inventory-based methods are mostly quite sophisticated and vast methods, the methods for characterizing demand are rather straightforward. All of the demand characterizing methods use two or three criteria to characterize it, which makes it not that difficult to plot all SKUs into a graph.
A first study shows that there are some criteria which are discussed more than once in the demand characterizing methods. These recurring criteria indicate that those criteria have some kind of importance. The recurring criteria are discussed below.
2.DEMAND CHARACTERIZING METHODS
___________________________________________________________________________________________________________________________________________________________________________________
Variation of demand sizes. First, the variation in demand sizes is used in three methods (D’Allessandro & Baveja (2000), Eaves & Kingsman (2004), Syntetos et al. (2008)). The variation in demand sizes is indicated with the coefficient of variation (COV) on the demand sizes. The higher the COV, the more demand sizes vary and the more difficult they are to predict adequately.
Volume. The second criterion, which is used several times, is volume. In the articles of Akkerman (2007), D’Allessandro & Baveja (2000) and Johnston et al. (2008), the volume criterion is mentioned as average demand (or order-) size. The article of Huiskonen et al.
(2003) can also be mentioned here, because they apply the annual sales volume. The annual sales volume also gives a good indication on demand volumes of a certain SKU, however not on average base, but total annual base.
Demand pattern. The third criterion which is mentioned several times in the articles is demand pattern. Several other authors (Huiskonen et al. (2003), Talluri et al. (2004) and Williams (1984)) use the criterion demand pattern. These methods indicate demand patterns on a scale of continuous demand pattern to discrete demand pattern. Whereas in continuous demand patterns, products are ordered for example every two weeks, in discrete demand patterns, products are ordered for example only once or twice per year.
The less continuous the demand pattern is (and the more discrete it is), the less it can be predicted accurately.
In two articles the demand pattern is indicated as the variability in time between orders (Eaves & Kingsman (2004), Williams (1984)), whereas two other articles use the average time between orders to indicate the demand pattern (Akkerman (2007), Syntetos et al.
(2008)).
2.3 Selection criteria
Some of the methods described in the previous two chapters (chapter 2.2 and 2.3) are quite difficult and sophisticated methods. Clearly, not all of them can easily be used for the purpose of this research. Several interviews and conversations with people from the supply chain department, as well as observations within SPL led to a number of preconditions which the models should meet. In cooperation with the supply chain department, the following selection criteria are used to select the method(s) for characterizing SPL’s demand.
Simplicity. The first selection criteria the demand characterizing model should meet, is simplicity. A number of the researches, discussed in the chapter above (chapter 2.1;
‘Methods’), explain methods that are way too complicated for the scope of this research.
One of the aims of this research is developing a characterization method, which the supply chain department of SPL can reuse again on the demand data later in time. Therefore, the method, to be used should not be too complicated. If it would be too complicated, the method would not be attractive to use, which can be a great boundary for the supply chain department to use this method again. Simplicity (and thus complexity) is a rather vague term which is hard to measure. However, some of the methods are clearly very complicated.
For example the methods using the Artificial Neural Network Approach and the Operations Related Groups approach are evidently too complicated.
Overview. During several interviews with the people of the supply chain department, it became clear that they wanted a simple overview on demand of all the SKUs they are working with. Therefore, the method to be implemented at SPL should give a clear overview on the demand characteristics (to be chosen). When the overview is created, the people of
the supply chain department can easily look at it to check the characteristics of the SKU before planning it.
Aggregation level. As stated above, the supply chain department needs an overview of the characteristics of the SKUs it works with. This means that the method should deal with the SKUs individually, instead of using an aggregation level like for example product group or product family.
MS Excel compatible. As already discussed in the description of some of the methods, a number of them need sophisticated software to make a categorization. Using sophisticated software would be a boundary for the supply chain department to use the method. Not only would it could become too complicated, also the availability of the software can be a problem. Hence, the software to be used for the characterization method has to be MS Excel. The people of the supply chain department know how to use the program, which makes it more attractive to reuse the method. In addition, cost are a great barrier to use a new program. Not only purchasing the program is a barrier, linking it with the SAP software at SPL can be a problem.
Statistically based. The method to characterize SPL’s demand should be statistically based. It should be statistically based because the size of the dataset is rather big, which means that if all SKUs should be treated and assessed individually, it would take a large amount of time. In addition, the SPL department will not perform a reassessment later, because of their valuable and limited time. Another important reason for choosing a statistically based method rather than a method based on inquiries and judgment lies in the fact that one can assess products into its own or its departments favor. Sales people will try to ensure larger inventories to guarantee short lead times towards their customers, while the finance department favors small inventories to cut holding cost and other financial risks and production favors efficient batches and smooth demand.
Production strategy decision tool. As discussed in the aim of this research will be used both for characterizing demand as well as for a categorization into production strategies.
Therefore, a usable method should use criteria which can be used to make a categorization for production strategies. Whereas the method as described in the article of Huiskonen et al.
(2003) is a clear method to decide on production strategies, a number of the articles, like Ramanathan (2006) and Ng (2006) do not give any kind of support on these strategy decisions.
2.5 Further analysis
When applying the preconditions and the ability to support on production strategy decisions to the initial twenty methods (as mentioned in chapter 2.1 ‘Methods’), a number of articles turns out to be unusable for this research.
Unusable methods. The methods which are not usable for this research purpose are shown in the following table (table 2.1).
2.DEMAND CHARACTERIZING METHODS
___________________________________________________________________________________________________________________________________________________________________________________
Author(s) Year Reason
Chen et al. 2004 Not statistically based and purpose does not fit
Chen et al. 2008 Too complicated method
Ernst & Cohen 1990 Too complicated method
Fisher et al. 2000 Purpose of showing forecasting error does not fit Flores & Whybark 1987 Inventory categorization, no useful criteria Garcia-Flores et al. 2003 Demand characterization based on recentness Guvenir & Erel 1998 Too complicated method
Huiskonen 2001 Inventory categorization, no useful criteria
Ng 2006 Too complicated method
Partovi & Anandarajan 2002 Too complicated method
Ramanathan 2006 Inventory categorization and too complicated Schomer 1965 Inventory categorization, no useful criteria Talluri et al. 2004 Method used for safety stock considerations Table 2.1. Unusable characterization methods
The table above shows thirteen of the twenty methods, which do not fit into the purpose of this research. A great part of these methods is unusable because they are rather complicated. For example, one of the methods is based on statistical clustering, and another one uses an artificial neural network approach. Using such a method would take a immense study in understanding and using the method. Considering the limited amount of time, that is not a realistic option. Moreover, there is a great chance that the supply chain department is not going to use it for reassessment because of its complexity.
Another great part of the unusable methods is not usable because of the criteria they use.
Criteria like ‘forecasting error’ and ‘perceived value of a part’ do not fit the purpose of this research and are therefore not usable.
Usable methods. The remaining seven methods do fit the purpose of this research. They all have the purpose to characterize demand, and a part of the models also takes into account the decisions on production strategies. Because all seven remaining methods share the same kind of purpose, they partly show overlap in the criteria they use to characterize the demand. The remaining methods and the criteria each method uses, are shown in the following table (table 2.2).
Author(s) Year Criteria
Average order size
Akkerman 2007
Inter-arrival time of orders COV demand sizes
D’Allessandro & Baveja 2000
Average order size
Transaction variability rate Eaves & Kingman 2004
Order size variation Demand pattern
Huiskonen et al. 2003
Annual sales volume Average order size
Johnston et al. 2003
Number of orders per SKU COV demand sizes
Syntetos et al. 2008
Inter-arrival time of orders Transaction variability rate
Williams 1984
Demand pattern Table 2.2. Remaining methods
The table above shows the remaining methods with the criteria they applied. The different colors show the overlap in applied criteria. The blue colored criteria are all related to demand patterns, the yellow criteria are all related to volumes.
Important criteria. As indicated in the table of the previous paragraph, two categories of criteria can be distinguished. These two categories of demand pattern and volumes and their relation with production strategies are discussed below.
• Demand pattern: This criterion indicates how demand of a certain SKU behaves during a period. The methods (see table 2.2) use this criterion in different ways, some methods use the variability in order or demand sizes (calculated by the coefficient of variation), while other methods classify a demand pattern for instance as continuous or lumpy. Also the inter-arrival time, transaction variability rate and order size variation are used in the upper methods to classify the demand pattern.
As described in chapter 2.1, the demand pattern is an important measure, needed for the categorization of SPL’s products into the appropriate production strategies.
• Volume: This criterion is used in several ways in the methods (see table 2.2). Some methods specify volume by using the average order size, while the annual sales volume and the number of orders per SKU are also used in other methods. To make a decision on production strategies for SPL, information about volumes should be known.
As a conclusion for this analysis, it can be said that the demand characterizing literature (as used in this research) in most cases apply both a demand pattern indicating criterion and a volume specifying criterion. Therefore, the demand characterizing method to be developed for SPL should also use both types of criteria, to be able to categorize products into the appropriate production strategy categories.
One remark which should be made here, is that this research only uses the demand characteristics to categorize products into a production strategy. Researchers as for example Soman (2002) state that decisions on production strategies should also be based on criteria like holding costs, irregular workload, setup times and perishability. Therefore, the categorization which is the result of this research only indicates which production
2.DEMAND CHARACTERIZING METHODS
___________________________________________________________________________________________________________________________________________________________________________________
strategy should be applied to the products based on the demand characteristics. As a consequence, SPL should also take the other criteria (as for example stated in Soman, 2002) take into consideration before applying a production strategy on a certain SKU.
2.6 Requirements of SPL
During several interviews and conversations, it became clear that the supply chain department of SPL had several requirements on the criteria, which should be fitted into the method of demand characterization and categorization.
Firstly, products which apply for a MTS or MTF production strategy, should be sold on a regular base. This means they should be sold in a large number of the months of a year. In the more months of a year the product is sold, the more regular it is sold. The regularity criterion as SPL prefers to apply, fits in one of the two categories of criteria as defined in the previous chapter (chapter 2.5). The regularity criterion clearly fits in the category of criteria indicating the demand pattern, and therefore fits well with the demand characterizing literature as described in the previous chapter. Because the criterion fits well with the demand characterizing literature, it can evidently be used in this research.
The second criterion which the supply chain department of SPL wants to be applied in the method of characterization and categorization is the split into two product categories.
Whereas the two most important products of SPL (EVAP products and SCM products) are produced in separate production environments and processes of planning and control are also split, the method should also be split. Moreover, the sold volumes of EVAP and SCM products differ a lot from each other, which makes it even more reasonable to split the two sorts of products in this method.
2.7 Final method
The final method which can be applied in the situation of SPL, cannot be directly distracted from one of the seven methods (as described in chapter 2.5). Firstly because most of these models only have the purpose of characterizing demand. Whereas this research has the additional purpose of categorizing SKUs into production strategy categories, these methods cannot be directly used as they are. In addition, the only method which uses demand characteristics to apply production strategies to products (Huiskonen et al. (2003)) also cannot be used in this research. The method of Huiskonen et al. (2003) cannot be used because the criteria it uses do not fit with the requirements which are made in cooperation with the supply chain department of SPL. The method classifies a demand pattern from discrete to continuous by calculating target inventory turnover rates, while SPL wants the products to be checked on regularity (by counting the number of months the product is sold). Because of these reasons, none of the seven usable methods is chosen directly to characterize SPL’s demand and categorize it into production strategies. However, both categories of criteria (demand pattern and volume) will be used to create a new model for the situation of SPL. Moreover, the requirements of SPL will be used in creating the new model for SPL.
Demand pattern. The first measure to be used in the new method is the demand pattern. The demand pattern is measured by using two criteria, being regularity and stability.
As stated in chapter 2.6, regularity is measured by counting in how many months of a year a certain SKU is sold. In the more months of a year it is sold, the more regular demand is. This
method uses the regularity measure, because SPL pointed out the importance of using this criterion for selecting the appropriate production strategy.
The second criteria used to indicate the demand pattern is stability. Stability is measured by using the COV of demand volumes. The COV of demand volumes is calculated as shown in the formula below (figure 2.3).
Figure 2.3. Formula for COV calculation
As the figure depicts, the COV of demand volumes is calculated by dividing the standard deviation of demand volumes by the mean of demand volumes (the standard deviation is the standard deviation of the whole population, and the mean is the arithmetic mean). The COV of demand volumes is a decent measure for checking stability of demand. The measure indicates how much the demand of a SKU varies during a period. The higher the COV of demand volumes is, the more variation in demand volumes and thus the less stable demand is. The following figure (figure 2.4) depicts both measures of the demand pattern and their relation with production strategies.
Figure 2.6. Stability and regularity
As the figure above shows, the lower stability and regularity are, the more a MTO production strategy fits to the demand characteristics of the product. The higher stability and regularity are, the more a MTF or MTS production strategy fits to the demand characteristics of the product. The decisions on which characteristics are regular, irregular, stable and unstable are made in the next part of this thesis (part 3. Demand characterization).
COV =
MTF/MTS
MTO
Stability Regularity
Low Low High
High
σ
µ
2.DEMAND CHARACTERIZING METHODS
___________________________________________________________________________________________________________________________________________________________________________________
Volume. The other important measure for deciding on production strategies for a product at SPL is the volume measure. Volume is measured by calculating the total demand for a product during a certain year. The volume measure is needed to make the decision between a MTS and a MTF production strategy. Products with a stable and regular demand pattern apply for one of these two production strategies. To decide on one of these two, the total demand volume is taken into consideration. The higher the total demand volume, the more a MTF production strategy fits the product its demand characteristics. Products with lower demand, and a stable and regular demand pattern best fit with a MTS production strategy. SPL bundles predicted small demand volumes over more periods, and makes one efficient larger production run. Decisions on which volumes are high and which are low are made in the next part of this thesis (part 3. Demand characterization).
Decision tree. To summarize this chapter and this part of the research a decision tree is created, to illustrate the decision-taking process for categorizing SPL’s products based on its individual demand characteristics. The decision tree is shown in the figure below (figure 2.7).
SKU
Stable
Regular Yes
Large volume
MTO Yes
No
No
MTS MTF
Yes No
Figure 2.8. Theoretical decision tree production strategies
The figure above shows the decision pattern for categorizing all SKUs into the appropriate production strategy. The first step in the decision tree is to look at the stability of demand.
This is done by calculating the COV of demand volumes. Unstable demand volumes are directly put into the MTO category. Secondly, the demand pattern of the SKU is evaluated, by counting the number of months in which the product is demanded by customers. SKUs with a stable volume, but irregular demand pattern are also categorized into the MTO category. Finally, the annual shipped volumes are examined. SKUs ordered in stable demand volumes, showing a regular demand pattern and total demand volumes are large, are categorized into the MTF category. The SKUs with stable demand volumes, showing a regular demand pattern, but have a small total shipped volume, are categorized into the MTS category. The extended version of this decision tree, made up after the data analysis, is shown in appendix 4.
3. DEMAND CHARACTERIZATION
This is the practical part of the thesis. It describes the method for characterizing demand in more detail. First, the general characteristics about the dataset and demand are described, followed by a description of the trends in demand. After that follows the method for characterizing demand, with the application of borders for the decision tree.
3.1 Dataset
The dataset, which will be used in this data analysis, is provided by the Supply Chain Management Accountant. The dataset is a download of MFG Pro, the program used by the order processors of the company.
It contains the amounts of boxes delivered (shipped goods) to the customers of SPL, with the dates requested by the customers to ship the goods. This means that a customer in Hong Kong, who ordered a requested date of the first of February, wants the goods to be on the boat leaving at that day. He has to wait some additional weeks for the goods to arrive at the port and to be delivered at his address.
Unit of measure. The amounts of shipped goods (for example boxes) are used, because those are the only amounts available in the database. Those amounts can differ from the original requested amounts of goods by the customer, but are overwritten as soon as the amount for a certain order changes because of customer requests or constraints of SPL. To be able to use the data, the assumption is made that the shipped amounts of products do not differ much from the quantities requested by the customers. Whereas amounts of shipped goods do not give a proper view on actual shipped volumes, another variable is used. The amounts do not give a proper view because a box with cans of EVAP contains a lot less kilograms (approximately twenty kilograms) than a ‘bag in box’ item, which is filled with a total of 1,300 kilograms of EVAP or SCM. Therefore, the shipped amounts are multiplied with the kilograms of fluid they carry. In short, instead of amounts of items, the shipped volumes in kilograms are used.
Scope. By narrowing the spreadsheet down to the scope of this research (only orders for SPL, only the year 2007 and EVAP and SCM products), the dataset eventually contains X*
order rows. These order rows represent the orders of X* different SKUs, which consist of X*
SCM SKUs and X* EVAP SKUs. *) Classified
A part of these all SKUs ordered in 2007 now have the status of being expired. These products will not be sold anymore, but the data of the expired SKUs are used in this analysis, to determine the right categories. The final categorization of production strategies however, does not take the expired SKUs into account because that information would be irrelevant for SPL.
3.2 Demand characteristics and trends
This chapter deals with the characterization of SPL’s demand. First, the results of a general analysis are shown, followed by a description of the trends in SPL’s demand.
General characteristics. The total demand volume of SPL for its EVAP and SCM products was nearly X* million kilograms in the year 2007. *) Classified
The graph below (figure 3.1) shows the (fluctuating) trend of the demand over the year 2007.
