DESIGN OF A MAKE AND BUY SOURCING STRATEGY TO MANAGE UNCERTAINTIES IN FOOD SUPPLY CHAINS

DESIGN OF A MAKE AND BUY SOURCING

STRATEGY TO MANAGE UNCERTAINTIES

IN FOOD SUPPLY CHAINS

MSc thesis

By

N.C. van der Wijden

MSc Technology & Operations Management

University of Groningen

MSc Supply Chain & Operations Management

Newcastle University Business School

Supervisors: J. Riezebos & Y. Yang

Student numbers: 1902695 (RUG) & 13059286 (NUBS)

Haarlemmerdijk 16

1013JC Amsterdam

2

A

BSTRACT

Uncertainties in food supply chains may take the form of high variability in supply, process, demand and price of ingredients that may significantly influence the delivery performance and sourcing costs for food processors. Therefore, an effective sourcing strategy is required that guidelines food processors in day-to-day sourcing decisions and that assures a minimal level of sourcing costs while still fulfilling demand.

Literature suggest to apply a flexible make and buy sourcing strategy that allows the food processor to flexibly switch among making and buying the ingredient to absorb these uncertainties. However, there is a clear gap defined in literature when exploring how food processors can apply such a sourcing strategy. Therefore, this dissertation contributes to both practical and scientific relevance by designing a policy that could assists food processors in applying a make and buy sourcing strategy to manage uncertainties in food supply chains.

The first phase of the research consists of a literature review and explorative case study that identify relevant variables that influence the design of the make and buy sourcing strategy. It is found that uncertainties in demand, supply, process and price are the main variables that should influence how food processors should apply a flexible make and buy sourcing strategy. Moreover, available capacity, competitive advantage and internal technical capability are as well relevant variables.

In the second phase the design of the policy that considers the relevant variables is introduced. At first, the policy suggests to group the products into two priority groups based on the actual effects of the uncertainties on the delivery performance of the ingredients. Second, separate dynamic lot-sizing models for both priority groups are introduced that determines the optimal ratio of making and buying. Also, it includes a flexibility range that allows the food processor to deviate for the optimal ratio to absorb uncertainties in demand, process, supply and price. It is shown that the application of the proposed policy in practice can result in cost reductions.

3

A

CKNOWLEDGEMENTS

By finishing this master thesis, a special moment has come in which I complete the Double Degree Master Operations Management at the University of Groningen and the Newcastle University Business School and say good bye to a great time at the university. I am very excited to present you this document, which could not have been finished without the inspiration and assistance of the following people.

First of all, I would like to thank Henk-Jan Meints as my supervisor during my internship for introducing me into such an inspiring organization and guiding me with his enthusiasm and intelligence to this final result. Next, I would like to thank Jan Riezebos as my leading supervisor during the whole research project. His critical notes, positive attitude and thoughtful insights gave me the boost that I needed every time I left his office. Moreover, I would like to thank Ying Yang as my co-assessor for providing helpful feedback. Next, I would like to thank everybody from the Milk Valorisation and Allocation department at Royal FrieslandCampina for their input and the pleasant stay during my internship. At last, I would really like to thank my family and friends for their support and interest, especially Alet Kamphuis who have been the best supporting friend I could have during my study.

4

T

ABLE OF CONTENT

1. Introduction ... 6

2. Theoretical background ... 9

2.1 Uncertainties in food processing supply chains ... 9

2.1.1 Sources of uncertainties ... 9

2.1.2 Effects of uncertainties ... 10

2.2 Decisions variables that influence the sourcing strategy design... 13

2.2.1 Uncertainties as variables ... 13

2.2.2 Other variables described in literature ... 14

2.3 Supplier selection & order allocation optimization models ... 16

3. The research context ... 19

3.1 The research methodology ... 19

3.2 The case company ... 19

3.3 Data collection and analysis ... 21

3.3.1 The explorative phase ... 21

3.3.2 The design phase ... 21

3.3.3 The validation phase ... 21

4. The findings of the explorative phase ... 22

4.1 Uncertainties in the case company ... 22

4.2 Strategies at the case company ... 23

4.1.1 Strategies to increase delivery performance ... 24

4.2.1 Strategies to decrease sourcing costs ... 25

4.3 The sourcing strategy decision at the case company ... 26

5. Design of the make and buy policy ... 28

5.1 Analyses of the variables ... 28

5.2 Process steps ... 30

5.3 Mathematical model ... 32

5.3.1 The sourcing costs optimization model ... 32

5

6. Numerical example & validation of the make and buy policy ... 40

6.1 Grouping the products ... 40

6.2 The sourcing costs optimization model ... 40

6.3 The delivery performance optimization model ... 47

7. Discussion ... 52

7.1 Results explorative phase ... 52

7.2 Results design phase ... 53

7.3 Results validation phase ... 54

8. Conclusion ... 56

9. References ... 58

10. Appendices ... 62

10.1 Grouping the make and buy products ... 63

10.2 Calculating the price parameter ... 66

10.3 Visualization of the mathematical models ... 67

10.4 Results sourcing cost optimization model ... 68

6

1. I

NTRODUCTION

Food processors face numerous challenges as a consequence of both the unique product and process characteristics of food supply chains and unsteady food markets (Banaszewska et al. 2013). These inherent characteristics of food supply chains mainly cause uncertainties in the supply, process, lead time and demand of ingredients requested by food processors (Vorst & Beulens 2002). Managing these uncertainties is one of the main challenges of food processors since the more or less predictable fluctuations in the supply, process, lead time and demand may result in high costs. For instance, inventory shortages may arise resulting in a costly downtime of production or even loss of goods sold. Or, on the other hand, uncertainties may result in high inventory costs, if the effects of uncertainties are simply managed by sourcing redundant amounts of ingredients (Heriot & Kulkarni 2001; Martinez & Simchi 2003).

The presence of these uncertainties in the supply of food ingredients triggers the question what quantities to order at what supplier (Guan & Philpott 2011; Rijpkema et al. 2014). However, in reality this decision becomes even more complex since the prices of most food ingredients are volatile and thus uncertain as well (Kouvelis, Li, & Ding, 2013). So, instead of solely question what quantities to order at what supplier, it is evenly important to consider the purchase timing to achieve the lowest purchase price (Keane & Connor 2009; Kouvelis et al. 2013).

To summarize, these uncertainties are main factors setting the sourcing environment of food processors. Uncertainties in supply chains need to be managed properly to achieve a minimal level of purchasing and inventory costs while still securing supply and thus fulfilling demand of the food processor (Martinez & Simchi 2003; Vorst & Beulens 2002). This highlights the clear need for an effective sourcing strategy that reduces the effects of uncertainty and that assists food processors in making day-to-day sourcing decisions.

7

sourcing environment characterized by uncertainties, several trade-offs may emerge in these decisions: e.g. single versus multiple sourcing (Costantino & Pellegrino 2010; Burke et al. 2007), local versus global sourcing (Christopher et al. 2011), or long term versus arms-length supplier relationships (Kalwani & Narayandas 1995).

However, no clear sourcing strategy is defined in literature that is designed to manage the uncertainties, while its vitality has just been highlighted. There are indeed other strategies described in literature that focus on managing either one type of uncertainty. For instance, financial hedging may be used to mitigate the effects of price uncertainty and is basically explained by purchasing insurance, but usually do not affect the day-to-day sourcing of raw materials (Berling 2011).

The answer might lay in operational hedging strategies where the concept of operational flexibility is applied: the organization’s ability to anticipate and flexibly respond to changes in market conditions by means of the firm’s operations (Boyabatli & Toktay 2004). Interestingly, Kouvelis & Milner (2002), who have studied the interplay of demand and supply uncertainty on outsourcing and capacity decisions in multi-stage supply chains, conclude that firms generally respond to uncertainties by simultaneously making and buying the same raw materials – a sourcing strategy called concurrent sourcing (Parmigiani 2007). Parmigiani (2007) has found strong support for the hypothesis that concurrent sourcing is a third, distinct sourcing strategy next to making or buying the raw material. It appears that firms first decide whether or not to produce internally, and thereafter determine the fraction of the total demand that is produced internally based on several variables.

8

However, these authors do not describe how a food processor nor a regular organization can apply this flexible make and buy sourcing strategy to deal with uncertainties. Therefore, this dissertation aims to assist food processors in applying a flexible make and buy sourcing strategy by designing a policy that answers the strategic questions (1) where to buy during a year (i.e. external or internal supplier) and (2) how much to buy where during a year. The research question that follows is

How can food processors apply a flexible make and buy sourcing strategy to manage uncertainties in food supply chains?

The sub-questions that are addressed in this dissertation to answer the research question are 1. What variables influence the design of a make and buy sourcing strategy according to

theory?

2. What variables influence the design of a make and buy sourcing strategy according to practice?

3. What policy could assist food processors in applying a flexible make and buy sourcing strategy, incorporating the relevant variables, to manage uncertainties?

4. How effective is the policy if tested on historical data?

9

2. T

HEORETICAL BACKGROUND

This section provides the theoretical background of this dissertation. It aims to answer the first sub-question “What variables influence the design of a make and buy sourcing strategy according to theory?” To sketch the sourcing environment of food processors, at first the main sources of uncertainties for food processors are discussed. Second, variables described in literature that influence the design of a make and buy sourcing strategy are introduced. Third, supplier selection and order allocation optimization models are discussed that show how these variables can be modelled in order to make the sourcing strategy decision.

It should be highlighted that in this dissertation the design of the make and buy sourcing strategy and thus the sourcing strategy decision is divided into two design problems based on the definition of a sourcing strategy by Costanino & Pellegrion (2010): (1) where to buy and (2) how much to buy where during a year. The definition of a sourcing strategy is thus limited into these two questions. Moreover, the concepts make, internal supplier, internal sourcing and buy, external supplier, external sourcing are used interchangeably.

2.1

U

2.1.1 SOURCES OF UNCERTAINTIES

Vorst & Beulens (2002), who have identified the sources of supply chain uncertainties, recognize that the specific product and process characteristics in food supply chains are the major sources of food supply chain uncertainties. These inherent characteristics cause uncertainties that may take the form of more or less predictable fluctuations in demand, process or supply of ingredients for food processors. Below, for each type of uncertainty the sources are shortly described.

10

economic conditions (Keane & Connor 2009). Process uncertainty is in general caused by variable process yield and scrap-rates of the supplier that produces the ingredient.

However, Vorst & Beulens (2002) ignore other crucial types of uncertainty for food processors. To explain, most ingredients of food processors are agricultural products (e.g. milk or potatoes) that are traded on commodity markets. In such markets, the prices of products are highly volatile, and thus uncertain. Price uncertainty can basically be explained by the inelastic demand of food products combined with the uncertainties in supply whereby even small changes in supply can cause very large changes in price (Keane & Connor 2009; Kouvelis & Milner 2002; Donk et al. 2008). Moreover, lead time uncertainty is also a type of uncertainty that has to be managed by food processors. Since most food processors rely on multiple suppliers from several global regions, they have to manage long, complex, and highly uncertain (i.e. variable) lead times (Kouvelis & Li 2008). Sources of lead time uncertainty might be supply network characteristics: e.g. the geographical spread of suppliers that are able to supply during different periods may lead to different lead times during the year. Another source might be the capabilities of the transportation network that may change due to changing weather conditions (Riezebos & Zhu 2014).

2.1.2 EFFECTS OF UNCERTAINTIES

11

Some studies demonstrate that the uncertainties in food supply chains directly influence the delivery performance and sourcing costs of ingredients. For instance, Vorst & Beulens (2002) demonstrate that uncertainties propagate throughout the supply chain causing inefficient processing and non-value adding activities. They state that “the more uncertainty related to a process, the more waste will be in the process” (p. 412). However, this may not always be the case. According to Tang & Tomlin (2008), food processors may apply strategies that reduce the negative effects of undesirable events. For instance, the presence of uncertainties stimulates the food processor to create costly buffers in time and inventory, but also in capacity to decrease sourcing costs and fulfil demand at all times (Carr & Smeltzer 1999; Guan & Philpott 2011; Serel 2007). These type of strategies weaken the positive relation between uncertainty and its effect on performance delivery and sourcing costs (figure 2.1). An overview of the strategies described in literature that reduce the negative effects of uncertainty is provided in table 2.1.

12

Literature stream Strategy Decreases the effect of … Literature Inventory management literature Building (safety, anticipation) stocks

All Kampen, Donk, & Zee (2009)

Having safety lead times Supply, lead time & process uncertainty

Kampen, Donk, & Zee (2009)

Reserve emergency capacity

Supply, lead time, process &demand uncertainty

Georgiadis, Vlachos, & Iakovou (2005); Johansen &

Thorstenson (1998) Financial literature Execute financial hedging Price uncertainty Ding et al., (2014);

Kouvelis et al., (2013) Supply chain management

literature

Using global sourcing Supply, process, demand & price uncertainty

(increases lead time uncertainty)

Christopher et al., (2011)

Using multiple sourcing Supply, process, demand & price uncertainty

(increases lead time uncertainty)

Costantino & Pellegrino (2010)

Having flexible supply contracts

Supply, demand & price uncertainty

Kouvelis & Li (2014) Standardize recipes of

ingredients

Supply & process uncertainty

Boyabatli & Toktay (2004)

Postpone the production decision

Supply & demand uncertainty

Boyabatli & Toktay (2004)

Table 2.1 Strategies described in literature that reduce the effects of supply chain uncertainties.

13

2.2

D

STRATEGY DESIGN

Aissaoui et al. (2007) argue that decision makers face different sourcing environments that lead to different sourcing decisions (p.3549). Consequently, in order to make the right decisions, it is vital to investigate the goal the decision maker aims to achieve when selecting a supplier. Since the sourcing environment of food processors is characterized by uncertainties and it is demonstrated in literature that firms respond to uncertainties by simultaneously and flexibly making and buying ingredients (Kouvelis & Milner 2002), the uncertainties described in section 2.1 are the main variables influencing the design of such a make and buy sourcing strategy (Kouvelis & Li, 2014; Lambrecht et al.,2011). Nevertheless, literature also describes other variables not related to uncertainty that may also have a significant influence. Based on a literature review, this section summarizes at first the uncertainties as variables and later on other relevant decision variables influencing the design of the make and buy strategy.

2.2.1 UNCERTAINTIES AS VARIABLES

The main effect of lead time, demand, supply and process uncertainty is a reduction in the delivery performance of the requested ingredients (Fisher, 1997; Vorst & Beulens 2002). The main idea behind simultaneously making and buying the ingredients is that a food processor can exploit the advantages of both sourcing strategies to reduce these effects. To explain, if a manufacturer faces uncertainty, some studies demonstrate that there might be an increased need in vertical integration or the make strategy (Kouvelis & Milner 2002). Applying the make strategy provides a manufacturer the opportunity to reserve emergency capacity (Kouvelis & Milner 2002; Donk et al. 2008; Vorst 2000) or to place emergency orders (Georgiadis et al. 2005) to absorb the fluctuations in lead time, demand, supply and process of the ingredients. In addition, basically it can be stated that the internal supplier is more reliable due to the advantages of a strategic partnership and aligned interests between buyer and supplier (e.g. shared benefits, trust, control) (Costantino & Pellegrino 2010).

14

demonstrate that a highly uncertain demand triggers manufacturers to rely more on outsourcing and thus buying the ingredients. For moments of high demand, the firm produces a quantity that allows the manufacturer to operate at optimal capacity utilizations and buys any quantity in excess at outside suppliers to absorb the fluctuations.

Furthermore, price uncertainty might have the result of paying too much for the ingredients of food processors, but on the other hand it provides food processors the opportunity to take advantage of its fluctuations (Ding et al. 2014). Taking the first perspective, price uncertainty may trigger a food processor to produce the ingredients internally and thus apply the make strategy. In that way, the purchasing price is equal to the production costs of producing a unit and thus relatively stable (Ding et al. 2014). In addition, the food processor can take advantage of the principles of economies of scale pressing down the purchasing price (Boyabatli & Toktay 2004). Taking the other perspective, the food processor should be able to benefit from the possible low prices at the commodity markets and thus be able to purchase the products at all times when the price is low. Therefore, the food processor should acquire multiple suppliers in their supplier base to be able to compare the prices and improve their negotiation position, and thus apply the buy strategy. (Costantino & Pellegrino 2010).

Nevertheless, the question when to use the make strategy and when to use the buy strategy, and thus the question how to use these uncertainties as variable in the design of a make and buy strategy is not answered in literature.

2.2.2 OTHER VARIABLES DESCRIBED IN LITERATURE

There is concurrent sourcing literature available that describes why an organization may choose to both make and buy the raw materials and how it affects a manufacturers’ performance (for a comprehensive review please refer to Mols (2010)). However, the introduced variables only influence the decision if a make and buy sourcing strategy should be applied, but not how.

15

based on the theory of complementarities and constraints and identifies two types of both concepts. These complementarities and constraints and the underlying mechanism they use to influence sourcing costs can be found in table 2.2.

Complementarities Variables Mechanism

Incentive

complementarities

Competition between suppliers Continuous improvement and

benchmarking between suppliers increase performance and in turn decrease average sourcing costs

Knowledge complementarities

Knowledge sharing (collaboration) Improvement of competence of suppliers increases performance and in turn decreases average sourcing costs

Constraints Variables Mechanism

Limits to scale Diseconomies of scale, excess capacity (due

to demand uncertainty)

The presence of diseconomies of scale and excess capacity increase average sourcing costs

Barriers to exit Constraints to exit internal or external

sourcing (e.g. reputation or commitment lock-ins, pressure from stakeholders, employment contracts etc.)

Not able to end a sourcing mode increase average sourcing costs

Table 2.2 Complementarities and constraints influencing the design of a make and buy sourcing strategy in the study of Puranam et al (2013).

16

Although there is literature available that argue that the make and buy strategy lays at a certain point of the make or buy continuum (Heriot & Kulkarni 2001) and thus similar variables may influence the design of the make and buy sourcing strategy, more recent literature (Puranam et al. 2013; Parmigiani & Mitchel 2009) argue that concurrent sourcing is a distinct strategy. The variables in the make or buy literature are thus irrelevant for this research.

2.3

S

&

MODELS

This dissertation eventually attempts to propose a policy that models the sourcing strategy decision problem answering the questions where to buy (i.e. intern or extern) and how much to buy during the year in an environment characterized by uncertainties. According to Aissaoui et al. (2007), these questions are similar to the questions asked in the supplier selection and order quantity allocation problem in uncertain environments. Therefore, this section reviews models available in literature that consider these problems in uncertain sourcing environments.

Aissaoui et al (2007), who reviewed the literature about supplier selection and order lot sizing models, argue that these models consists of three main stages: decision criteria formulation, pre-selection of suppliers, and final pre-selection. During the decision criteria formulation, the relevant decision variables dependent on the sourcing situation are identified. Generally, two basic types of variables are described: objective and subjective ones. The objective variables are quantitative in nature, e.g. cost. In contrary, the subjective variables are harder to measure, e.g. reliability of supplier. In some cases, these variables are conflicting and a trade-off is required to make: e.g. an unreliable supplier offering a low price versus a reliable supplier offering a high price. During the pre-selection of suppliers, it is aimed to dismiss the inefficient suppliers and to reduce the supplier base to a small amount of acceptable suppliers. Several evaluation methods (e.g. AHP) are used to assess the performance of the suppliers. In the final selection stage, orders are allocated to the selected suppliers taking into account the system constraints and several objective and subjective variables. This stage has had great attention in literature and numerous supplier selection and order allocation models are developed.

17

relationships or multiple sourcing where the orders are split among the available suppliers to ensure reliability (for instance Burke et al. 2007 and Costantino & Pellegrino 2010). These models can include a multi- or single-period problem. In contrary to the single-period problem, most multi-period models consider inventory management and consequently balance ordering, purchasing and holding costs. According to Aissaoui et al (2007), multi-period lot sizing has been one of the most studied problems in production and inventory management literature (p. 3527). The models may use the Economic Order Quantity concept to select suppliers and allocate orders among them (for instance Ghodsypour & Brien 2001), or uses a multi-period horizon where variables are defined to determine the quantity purchased in each separate period (for instance Basnet & Leung 2005). Indeed, almost all multi-period models consider any time depending (dynamic) parameter as deterministic and known, while these parameters are often uncertain (e.g. price, demand, etcetera).

The models that do insert stochastic parameters to model uncertainties are mostly conducted in a newsvendor setting (Ray & Jenamani 2014). These models determine the optimal order quantity by trading-off holding excess inventory and having stock outs (Dada et al. 2006). Recent examples are Yang, Yang, & Abdel-Malek (2007), Xanthopoulos, Vlachos, & Iakovou (2012) andRay & Jenamani (2014). However, there is no such model that considers more than two types of uncertainties, as considered in this research context.

18

uncertain and modelled by a known demand distribution and fraction of orders placed at suppliers that will not arrive. The suppliers have different yield distributions, procurement costs and capacity levels.

For a comprehensive literature review of (multi-period) supplier selection and order allocation models, please refer to the study of Aissaoui et al (2007). Generally, these models aim to minimize costs: not only the purchasing price and inventory costs are included, but the objective function may as well include penalties related to poor quality, shortage or inefficient utilization of capacity. The objective function is often restricted by system’s constraints as supplier capacities, buyer’s storage capacity, minimum order quantities by suppliers, and etcetera. Most models use mathematical programming techniques to formulate this decision problem for decision makers.

19

3. T

HE RESEARCH CONTEXT

This section describes the methodology of this research; discusses how data will be collected and analysed; and an introduction of the selected case will be presented.

3.1

T

To design a policy for food processors, a design research is executed. As the starting point of the design research, it is chosen to select a single case company to design, diagnose and analyse the problem in practice according to the first phases of the regulative cycle of Strien (1997). The single case study is used to analyse the variables described in literature and introduce other variables. The choice for a single case study rather than a multiple case study is based on the reasoning of Eisenhardt & Graebner (2007) who state that a single case study is most appropriate when a new phenomenon will be investigated and detailed data and insights should be extracted. Since there is a clear gap demonstrated in literature and thus a clear need for theory building, a single case study is the most appropriate research method.

In order to select a suitable case study for the research purpose, selection criteria are defined. As the research problem suggests the food processing industry, the first selection criterion is that the case company is a food processor. Second, the case company is facing the introduced uncertainties in the supply of their ingredients. Third, the case company should be able to apply the make and buy sourcing strategy and thus has to some extent an integrated supply chain.

3.2

T

Based on the selection criteria listed in the research methodology, a dairy processor located in the Netherlands is chosen as the case company. The company processes their ingredients into consumer products as dairy-based beverages, toddler nutrition and cheese, but also into semi-finished products for manufacturers of infant nutrition, the food industry and the pharmaceutical sector. Thus both consumer products and semi-finished products are finished products of the dairy processors’ supply chain.

20

processor. Another main ingredient is the so-called dairy raw material, a semi-finished product made from milk (e.g. milk powders, lactose). The dairy processor can both make the dairy raw materials but as well buy them externally, since most dairy raw materials are related to agricultural commodity markets. In this dissertation, only the dairy raw materials fit in the research scope based on the predefined selection criteria since it can be assumed that the liquid milk is solely ‘made’ by the dairy cooperation.

The dairy processor operates with a centralized sourcing department that is responsible to source and supply the required ingredients to all production plants. They collect the requested quantities and arrival times of all productions plants that form the total demand of a ingredient for a specified arrival time. The sourcing department then decides how much is sourced internally and thus produced by another production plant, or sourced externally. In sourcing these ingredients, the sourcing department especially faces uncertainties in supply, process, demand and price. Lead time uncertainty is considered as irrelevant in this research, since it is mentioned by the case company that the effect of lead time uncertainty on the delivery performance is negligible.

Figure 3.1 The research scope

21

department, that is significantly influenced by price, supply and process uncertainty, as already been demonstrated in section 2.2.1. It should be mentioned that based on this research scope the demand -of the production plant, not customer demand- is fulfilled when the delivery performance of the ingredients is optimal.

3.3

D

The research project was performed at the case company during a period of five months. The project was compromised in three phases: the explorative phase, the design phase and the validation phase.

3.3.1 THE EXPLORATIVE PHASE

The explorative phase is used to answer the second research sub-question. It covers the first two phases of the regulative cycle of Strien (1997) aiming to design the problem and perform the required analysis. To do this, qualitative and quantitative approaches were executed to gain insight in the challenges and opportunities of the case company. The richest information was gathered by several interviews (36), especially with the sourcing managers and category manager of the sourcing department. Notes were taken during each interview and stored in a central database. The insights gathered from the interviews were deepened by analysing quantitative data.

3.3.2 THE DESIGN PHASE

The design phase uses the analysed data of the explorative phase and the literature review described in chapter 2 as main input. To achieve triangulation of the research, the results from the explorative phase and literature review were presented to and discussed with the category manager and sourcing managers of the sourcing department. It was discussed if any crucial data was missing to design the suitable policy. After designing the policy, the results were again presented and discussed with the category manager and the senior manager of the sourcing department to assure its applicability.

3.3.3 THE VALIDATION PHASE

22

4. T

HE FINDINGS OF THE EXPLORATIVE PHASE

This section attempts to answer the second sub-question “What variables influence the design of a make and buy sourcing strategy according to practice?” At first, the main sources of uncertainties in the supply chain of the ingredients of the case company are discussed. Next, it is described how the case company applies different strategies to deal with the effects of uncertainties. At last, it is described what sourcing strategy is currently used for their input products to introduce other relevant variables.

4.1

U

Supply uncertainty of the dairy ingredients is basically explained by the seasonal production of milk. Due to calving patterns, changing weather conditions and changing cow feeding patterns, the milk production follows different seasonal patterns worldwide. Especially the grass feeding countries as Ireland, New Zealand and Australia have milk productions with clear seasonal patterns. Since the dairy ingredients are produced by processing liquid milk, they may as well follow a seasonal production pattern. Figure 4.1 visualizes the milk production and production of a dairy ingredient (Skimmed Milk Powder) in Ireland and Australia in 2013 and 2014. If both patterns are compared, similar seasonal peaks are identified. As can be seen in figure 4.1, in some regions the dairy ingredients are even not available during specific months. Moreover, some dairy ingredients require specific milk compositions (i.e. percentages of fat, protein and dry mass) in order to meet the quality standards and, consequently, can only be made during the peak seasons of the milk production. Furthermore, the sources of supply uncertainty may as well lay in the occurrence of random events as changing weather conditions or political and economic actions.

23 0 200 400 600 800 1000 1200

Jan Feb Mrt Apr May Jun Jul Aug Sept Oct Nov Dec

L x m ln Milk production 2013-2014 Ireland Australia 0 5 10 15 20 25 30 35

Jan Feb Mrt Apr May Jun Jul Aug Sept Oct Nov Dec

kt

on

x

ml

n

Skimmed Milk Powder production 2013-2014

Ireland Australia

Figure 4.1. Seasonal production patterns in milk and dairy ingredients.

Next to supply and demand uncertainty, process uncertainty emerges due the general reasons as machine breakdowns, production yield and unacceptable product quality. However, a typical source of process uncertainty is the occurrence of animal diseases, which may result in random fluctuations in the supply of ingredients.

At last, the sources of price uncertainty at the case company are similar to the sources established in the economics literature and primarily related to a combination of the somewhat unique characteristics for food demand combined with unexpected variation in food supply (4.1.1). In addition, the price development in commodity markets is extremely vulnerable to several other factors (e.g. the occurrence of a recession). In general, the prices of dairy ingredients differ between regions since all regions have different supply and demand patterns.

4.2

S

24

uncertainty may have a direct effect on the performance delivery and price uncertainty only directly on the sourcing costs (inventory & purchasing costs), the strategies are divided in strategies that focus on increasing performance delivery and decreasing sourcing costs.

4.1.1 STRATEGIES TO INCREASE DELIVERY PERFORMANCE

At first, the sourcing department strives to have at least five available suppliers year round for each product to assure supply. Therefore, the department has appointed suppliers from different regions to manage fluctuations in supply (e.g. due to seasonal production) and process. Moreover, the sourcing department attempts to equally spread the order quantities over different geographic regions, to avoid complete dependency on one region (e.g. a total region can be disrupted by drought).

Second, the sourcing department attempts to build in buffers in time and capacity. For instance, they have time buffers in their sourcing planning that enables the sourcing department to switch among supply contracts to absorb short-term fluctuations in demand, process and supply. Moreover, the department carries out a flexible capacity planning at the internal supplier to absorb both short- and long-term fluctuations in demand. The planning is made 15 months ahead before the capacity is actually used and updated every quarter. In addition, the company reserves buffers in their capacity planning to absorb emergency orders that emerge due to uncertainty in supply and process.

Third, the sourcing department has agreements with some suppliers in which they have agreed that the external supplier carries out anticipation inventory to be able to supply the product year round. The external supplier produces more than requested in seasonal peaks to be able to supply the products to the sourcing department during the months that normally the product is not available due to seasonality.

Fourth, the production plants maintain a specified level of safety stock to absorb the fluctuations in supply, process and demand.

25

rolling forecast of 18 months ahead from the production plants. Each forecast that is received by the sourcing department requests a specific arrival date. The sourcing department then sums up the forecasts for each ingredient for all production plants with the same requested arrival date to determine what where they are willing to source and when.

4.2.1 STRATEGIES TO DECREASE SOURCING COSTS

First, the concept of multiple and global sourcing is used to have multiple suppliers year round assuring an optimal negotiation position. In this way, the sourcing department can compare the prices between different regions and negotiate between different suppliers to assure the lowest purchasing price. Of course, the company tries to have suppliers in different regions to be able to source in the region where the purchasing price is the lowest.

Second, the sourcing department has appointed a group of experts that are responsible to forecast the prices of the commodity products. The experts are using advanced econometric methods to forecast the price as accurately as possible. However, even they can only forecast the price 6 months ahead due to its extreme volatility. After 6 months, the forecast accuracy is unacceptable. Based on the acceptable price forecast, the sourcing department decides where it is optimal to buy. For each dairy ingredient, it can be assumed that the uncertainty in price and thus forecast accuracy is similar.

Third, the sourcing department uses different types of supply contracts. They use year contracts with a fixed price if supply security is vital, short-term contracts if the price is currently decreasing, long term contracts based on a commodity trading index or they can buy at spot markets. These contracts can be seen as different tools that can be applied to assure the lowest possible purchasing price in each scenario.

Fourth, the sourcing department has the possibility to carry out stock at the internal supplier or at the production plants. The sourcing department is then able to buy more than required when the price is low to secure a low price and exploit the advantages of price volatility. However, the ingredients cannot be stored for an infinite time due to its perishable nature.

26

4.3

T

The sourcing department has 13 products in their portfolio that can be made internally and bought externally. The key variable that influences the decision where to source is the pure difference in purchasing price from the internal and external supplier. Since both prices are market prices and thus follow the variable price development in the commodity markets, flexibility in making this choice is crucial. However, the case company might be an exception in using this comparison mechanism instead of production costs (i.e. costs of raw material, costs of capacity etc.) versus acquisition costs (i.e. transport costs, negotiation costs, purchasing price etc.) as been often described in literature. When comparing production and acquisition costs, it is demonstrated that it might be favourable to produce the product internally to reduce the effects of price uncertainty and exploit advantages of economies of scale. This mechanism does thus not apply at the case company, since the sourcing department is seen as a general customer and capacity at the internal supplier can always be sold to other customers.

However, the sourcing department is constrained in their sourcing flexibility by maximum amounts that are allowed to source at one supplier in order to be not completely dependent on a single supplier. Of course, the reasoning behind this rule lies at the possibility to mitigate the effects of supply, process and demand uncertainty on delivery performance. Furthermore, for some products a minimum fraction of the requested amount should be sourced internally because of strategic reasons (i.e. marketing issues) or to maintain production skills (i.e. for products with high product complexity).

The sourcing strategy decision significantly influence the capacity planning of the internal supplier. Since the capacity planning is made 15 months before the scheduled capacity is actually used, this sourcing strategy decision should be made already 15 months ahead. The case company is therefore interested in the optimal ratio between making and buying the product per year. This ratio can guide the sourcing department in making their day-to-day sourcing decisions (i.e. make or buy).

27

based on the so-called valorisation value of the product. To explain, since the dairy processor is required to process all the milk that is supplied by its dairy cooperative members, they strive to extract the most value out of their milk. Based on an equivalency table that calculates the value of the finished product per processed kilogram milk (i.e. valorisation value), the milk is allocated to the most profitable products. In case of a peak supply in milk, more milk is allocated to the products with the highest valorisation value, the second highest etc. In the low season, first less is allocated to the lowest profitable end product, the second lowest etc. Thus, for these supply driven products the capacity planning is that flexible that the total yearly requested volume can be sourced internally, but the available capacity might be constrained by products with higher valorisation values.

28

5. D

ESIGN OF THE MAKE AND BUY POLICY

In this section the third sub-question “What policy could assist food processors in applying a flexible make and buy sourcing strategy, incorporating the relevant variables, to manage uncertainties?” is answered. First, the variables that influence the design of the make and buy strategy described in literature and practice are analysed based on their relevance. Next, the process steps that need to be made in order to design a flexible make and buy strategy are described. At last, the mathematical model is described that defines the optimal make and buy strategy for each ‘make and buy’ ingredient.

5.1

A

To start, in both theory and practice it is demonstrated that supply, price, demand and process uncertainty are the main triggers for applying a flexible make and buy strategy and thus significantly influence the design of such a strategy. Both theory and practice describe the same sources of these uncertainties. Furthermore, it is demonstrated in practice and theory that the effects of these uncertainties on the delivery performance and costs can be reduced by the use of different strategies applied in planning, scheduling and control. Both theory and practice name the use of global sourcing, multiple sourcing, buffers in time and capacity and advanced forecasting methods as effective strategies. Financial hedging is not considered as relevant in this dissertation, since it does not affect the day-to-day sourcing policy. Additionally, the carried safety stocks at the production plant falls beyond the scope of this research since it does not have effect on the delivery performance of the ingredients. Moreover, the supply contracts that are used by the sourcing department are seen as irrelevant, since the contracts are closed after it is decided where and how much to buy during the year.

29

Costantino & Pellegrino 2010) that optimal flexibility in sourcing is crucial. Global and multiple sourcing can be used to assure and optimal negotiation position. The incentive complementarities of Puranam et al (2013) do apply here, because multiple suppliers are competing and driving the sourcing costs down.

According to both theory (Aissaoui et al. 2007) and practice, in an uncertain environment, a single source should be avoided to assure high delivery performance. To increase the delivery performance on the short them, the sourcing department may rely on both emergency internal capacity and flexible volumes at suppliers. Both mechanisms are seen as equally effective in practice, since the internal and external supplier are equals in the case company. Both mechanisms are thus used in the make and buy policy.

The barriers to exit constraints of Puranam et al (2013) do also apply in practice, since the case company is required to source a minimum amount of some of their ingredients intern, due to strategic or technical reasons (e.g. maintain technical capability). This may result in an increase in sourcing costs, if the internal purchasing price is higher than the external purchasing price. Put differently, if a product is important to a company’s competitive advantage or a certain level of technical capability should be maintained, the product should be (partly) made internally. Moreover, the maximum amount of products to buy at one supplier is constrained by the available capacity intern and extern that may also result in a higher sourcing costs. The available capacity intern depends on the valorisation value of the product. The available capacity extern depends on the demand of competitors and possible effects of supply and process uncertainty.

At last, the constraint limits to scale of Puranam et al (2013) does not apply to the case company. To explain, if it is chosen to not source internally, the available capacity is scheduled to another product or is reserved for another customer. The capacity utilization remains the same, so there are no costs of diseconomies of scale or excess capacity.

30

Figure 5.1 The relevant variables that should be considered in the design of the policy.

5.2

P

31

To do so, the sources of uncertainties and strategies that reduce the effects of the uncertainties (figure 5.1) should be measured for each product. The measures are listed in table 5.1. The measures are designed to group the dairy ingredients into two priority groups: a group that solely prioritizes sourcing costs optimization and a group that focuses as well on performance delivery optimization. For both groups, price uncertainty and its effects on the sourcing costs can be seen as significant but equal. Therefore, the ingredients are grouped based on the actual effect supply, demand and process uncertainty have or might have on the performance delivery of the ingredient. So, only the sources of supply, process and demand uncertainty and the strategies that reduce the effects are considered here. In sum, the group with a high delivery performance – i.e. the supply, demand and process uncertainties and or their effects are negligible – prioritizes sourcing costs optimization, and the group with a low delivery performance – i.e. the supply, demand and process uncertainties and or their effects are significant – focuses both on sourcing costs and delivery performance optimization.

After the products are grouped in the two priority groups, it should be investigated if the product has strategic (e.g. competitive advantage) or technical capability requirements to produce the product internally. If so, the next step is to assume a reasonable minimal amount that should be sourced at the internal supplier based on the product characteristics. After these steps are taken, the mathematical model can be used to design the optimal make and buy sourcing strategy.

Measure Outcome

The number of available suppliers year round

… The number of available sourcing regions year round

… The extent of forecast accuracy of demand

% Probability of seasonal production, changing weather conditions, animal diseases

High/medium/low Flexibility range in available

internal capacity (short- & long-term)

% Flexibility range in available external capacity (short- & long-term)

% Possibility to build anticipation stock at external supplier

Yes/No

32

5.3

M

The mathematical model designs the make and buy strategy by identifying an optimal ratio of making and buying and a flexibility range wherein the sourcing department can deviate from this ratio in case of changing conditions. Since the capacity planning and in turn the sourcing strategy decision is already made 15 months before the capacity is used and before it is decided for each order where exactly to buy and how much to buy, the mathematical model optimizes the ratio between making and buying for each year. In other words, the sourcing strategy decision is thus a strategic decision and aims to guide the sourcing department in making day-to-day sourcing decisions. Consequently, this dissertation does not answer the questions where to buy and how much to buy when for each specific order, but for the sum of orders during the year.

For both groups of products separate mathematical models are designed. Since the model that solely optimizes sourcing costs (group with high delivery performance) can be seen as a building block for the model that also optimizes delivery performance (group with low delivery performance), it is at first described how the sourcing costs optimization model calculates the optimal ratio between making and buying per year and the flexibility range. Thereafter, the extensions in the delivery performance optimization model are introduced.

5.3.1 THE SOURCING COSTS OPTIMIZATION MODEL

The building block of this model is derived from the dynamic lot sizing models introduced in section 2.3. Because the model does not decide for each specific order where to buy and how much to buy in order to achieve a certain service level, the stochastic models in newsvendor settings are less suitable to the design problem considered in this dissertation. Thus, the model does not include uncertain or stochastic parameters, but is built to fit in an uncertain environment.

33

effective capacity of the suppliers can thus vary per quarter. Therefore, it is chosen to calculate the optimal order quantity for each quarter (multi-period).

Basically, the model determines for a representative set of historic years the optimal (e.g. minimal purchasing and inventory costs) order quantities to source at the internal and external supplier in each quarter by comparing the internal and external purchasing price. Both prices are derived from historic prices set in the related commodity markets per region. The average of the optimal ratios from the representative set of years can be seen as the optimal ratio for future years to source internally or externally.

However, ideally, the model should calculate the optimal ratios for future years and thus forecasting the price is required. However, in such high volatile markets, it is almost undoable to forecast the price for commodity products. This statement has not only been confirmed by theory (Kouvelis & Ding, 2009) but also by practice. Additionally, the sourcing strategy decision should be made 15 months ahead: it is then certainly undoable to forecast the price. Therefore, it is chosen to not exactly forecast the price, but use the average of the price of representative historic years (e.g. in most cases the most recent years) as main parameter to determine the optimal ratios for future years. It does not make sense to use as much years as possible to assure statistical significance because the development in the price depends on the conditions of the current dairy market. The average of the price is not fixed for all future years, but moved along the most recent values of the price by using time series forecasting techniques. In appendix II, it is shown how this parameter is calculated. The average of the price includes all relevant patterns (e.g. it may be on average cheaper to source in the first quarter) in the price development during a year that may influence the sourcing strategy decision. Next to the price parameter, the other main input to the model is demand. To calculate the average optimal ratio, historic demand is used. For future years, demand forecasts are inserted to the model.

34

external supplier, but at the production plant that demands the ingredient), if it is chosen to source more externally. However, the flexibility in sourcing is not infinite in that the amount that is sourced more than the actual demand and thus stored as inventory has a shelf life constraint to avoid deterioration. Moreover, building up inventory results in inventory costs, so it may not be that favourable at all. Furthermore, since these products are proven to have a high delivery performance, it is assumed that no emergency capacity is required.

Eventually, the sourcing costs optimization model tries to find the optimal order quantities for each supplier at each quarter by minimizing inventory and purchasing costs while considering the constraints. The model has three outputs

1) The average optimal ratio based on a representative set of historic years

2) The average optimal flexibility range based on a representative set of historic years 3) The optimal ratio for future years using average prices as parameter

The maximum flexibility range for all years is defined as the minimum and maximum amounts that are historically chosen by the model to source internally and externally. Beyond these ranges it can be assured that it is not cheaper to make the other choice. For example, if historically the minimum and maximum amount sourced internally was 30% and 70% respectively of the total yearly demand, it thus have never occurred that it was optimal to source less or more than 30% or 70% at the internal supplier. So, the probability that it is optimal to source less or more than 30% and 70% in future years is ignorable. Thus, it will not make any sense to source beyond these ranges. The flexibility range can be used to deviate from the optimal ratio to absorb fluctuations in demand, supply, process and price on the short- and long-term.

For both models the same assumptions do apply:

35

2. The external purchasing price is based on the smallest price of the related commodity markets of each external supplier where the product can be sourced from. Thus, if sourced from region x and y and price x<price y, the external price is x.

3. The transportation costs are assumed to be equal for each supplier. 4. The negotiation costs and import costs can be ignored (e.g. order costs). 5. The capacity of the inventory warehouses is assumed to be infinite.

6. It is assumed that there is a sufficient availability of ingredients (e.g. milk) to produce the product internally.

7. It is assumed that the sourcing department can rely on the total available (effective) capacity of the suppliers. Shared capacity is thus not taken into account.

8. It is assumed that if a supplier or region is disrupted, the consequence of the disruption is that the total available capacity is 0. In this way, the sourcing department is always prepared for the worst case scenario.

9. Other events than seasonal production and changing weather conditions that may disrupt a supplier from using its available capacity are not taking into account because the probabilities are assumed to be equal for each supplier.

It is assumed that the following inputs of the model are known:

Dt : Demand forecast in tons in quarter t (t=1,2,3,4).

cit : Purchasing cost per ton of supplier i (i=1,2 with 1 is the internal supplier and 2 is the

external supplier) in quarter t

hit: Holding costs per ton at the supplier i in quarter t

Cait: Available capacity at supplier i in quarter t

pit: The probability that the Cait of supplier i in quarter t is disrupted

yit: The remaining fraction of Cait in percentiles due to pit at supplier i in quarter t

s: The shelf-life of the product in quarters

36 The following inputs have to be calculated

Qit: The order quantity at supplier i in quarter t

Iit: Inventory in tons at supplier i in quarter t

Ceit: The effective available capacity at supplier i in quarter t

By

Iit = [Iit-1 + Qit - Dt]+ where [x]+ = x if x ≥ 0 and [x]+ = 0 if x<0.

Ceit = Cait (pit yit + 1 - pit ) where the available capacity is reduced by the average expected

value of the disruption effect on the capacity at supplier i in quarter t. For example, if the average probability pit is 0.20 that a supplier

is disrupted and the remaining fraction of the capacity yit is 0, then

the effective capacity Ceit is 0.80Cait.

And the objective function is to minimize

Z = ∑4_𝑡=1∑ (2_𝑖=1 Qit cit +Iit hit) where Qit≥ 0.

With the constraints of

Ii t ≤ Dts where Dtsis the maximum amount of orders that can be stored at

the internal supplier or production plant. In case of a shelf life s of

1

3quarter (1 month), the maximum inventory is 1

3Dt, if s is 2

3 quarter

then 2

3Dt, if s is 1 quarter then Dt, etcetera.

∑ (4

𝑡=1 D1t) ≥ f ∑4𝑡=1∑ (2_𝑖=1 Dit) if the product has strategic or technical requirements to source a

37

At last, the sum of all order quantities per quartile intern and extern should be equal to the total yearly demand.

∑ ∑ (2

𝑖=1 4

𝑡=1 Qit) = ∑4_𝑡=1∑ (2_𝑖=1 Dit)

The model is inserted in Microsoft Excel and uses a Linear Programming Tool to calculate Qit

and minimize Z.

5.3.2 THE DELIVERY PERFORMANCE OPTIMIZATION MODEL

The delivery performance optimization model can be seen as an extension of the sourcing cost optimization model. The same assumptions and constraints do apply, and the same input parameters are required to calculate the optimal ratios and flexibility ranges for the ingredients. However, two main differences can be distinguished.

First, the objective function is extended by penalty costs Uit (uncertainty costs)for orders that do

not arrive on time, in the right quantity or quality. The objective function is then to minimize

Z = ∑4_𝑡=1∑ (2_𝑖=1 Qit cit +Iit hit + Uit) where Qit≥ 0.

The penalty costs can be seen as a constraint in the model of Puranam et al (2013) that marginally increases the average sourcing costs. In this way, it might be optimal to source more at the reliable supplier, even if the unreliable supplier is cheaper. The similar ways of arguing as Federgruen & Yang (2009) and Mafakheri et al. (2011) are used, who source higher volumes at the more reliable suppliers. The value of the total penalty costs Uit at supplier i in quarter t is

dependent on the average expected value of the fraction of the requested volume Rit that does

arrive on time, in the right quantity or quality and 1-Rit is thus the expected value of the fraction

of the requested volume that does not arrive according to the specifications. Only for the 1-Rit

fraction of sourced volume Qit penalty costs uit should be paid. The formula is then

38 Rit is calculated by

Rit = ptitytit + paityait + pqityqit + (1-ptit - pait - pqit)

Where ptit is the probability that orders do not arrive on time with the remaining average fraction

of the requested volume ytit that does arrive, pait the probability that orders do not arrive in the

right quantity with the remaining average fraction of the requested volume yait that does arrive,

and pqit the probability that orders do not arrive in the right quality with the remaining average

fraction of the requested volume yqit that does arrive. It is then assumed that if ptit, pait or pqit≥ 0

then the other probabilities are 0. It is assumed that the probabilities and consequences for each historical order are known, so the average probability and consequence in quarter t can be calculated.

To give a brief example, consider the case that from 100 orders requested by supplier i=1 in quarter t=1 on average 20 orders do not arrive on time (ptit =0.20) and thus the average fraction

of the requested amount that arrive is 0 (ytit). Next, 5 orders on average do not arrive in the right

quantity (pait=0.05) with yait = 0.8 of the volume that is on average delivered in the right

quantity. At last, 5 orders on average do not arrive in the right quality (pqit = 0.05) and yqit =

0.60 of the volume that is on average delivered in the right quality. The average expected value of the fraction of the requested volume R11 that does arrive according to the specifications is 0.7 ∙

1 + 0.2 ∙ 0 + 0.05 ∙ 0.8 + 0.05 ∙ 0.60 = 0.77.

The orders that do not arrive are backordered. The penalty costs uit are estimated based on the

costs for holding extra units of inventory and extra transport costs (i.e. in such a short term the requested amounts needs to be transported by flight which is quite expensive) to assure that demand is still fulfilled.

39

probability of not on time, in quantity or in quality arrival. Therefore, capacity should at all times be available at the internal and external supplier to absorb fluctuations in demand, supply and process.

The following does then apply. Consider the case that the internal supplier is disrupted. It is assumed that if the supplier is disrupted, the capacity is 0, then the required emergence capacity Cn2t at the external supplier 2 in quarter i is

Cn2t = Q1t + Q2t

The Cn2t should then be lower than the actual available capacity at supplier 2 in quarter t, thus

Cn2t≤ Ca2t≤ Ce2t

Thus,

Q1t + Q2t = Cnit≤ Cait≤ Ceit

40

6. N

UMERICAL EXAMPLE

&

VALIDATION OF THE

MAKE AND BUY POLICY

This section applies the make and buy policy proposed in chapter 5 in the case company. It attempts to answer the question “How effective is the policy if tested on historical data?” taking a cost perspective.

6.1

G

The first step of the make and buy policy is to group the products into two priority groups. For each product the measures of table 5.1 are applied. To validate the content of the groups, the analyses are checked with the category manager of the sourcing department. An overview of the results can be found in appendix I and the definitive distinction in table 6.1. The cursive products require a minimum amount to be sourced internally.

Group with high delivery performance

1 SMP MH

2 SMP MH for UHT 3 SMP HHHS30+ 4 Lactose Edible

Group with low delivery performance

5 AMF 6 BMP HS 7 DEM. WP25%

8 DEM.WP90%

9 Gum based MP

10 Lactose Edible 200 mesh

11 SMP Agglo

12 Whey permeate

13 WMP Instant 28/37,5

Table 6.1 The ingredients of the case company grouped into the priority groups

6.2

T

Product 2 (SMP MH for UHT) is taken as an example to show how the sourcing costs optimization model can be implemented in practice. Product 2 can be sourced in the USA, Ireland, New Zealand and Australia. The purchasing cost per ton of the external supplier c2t in

41 2700,00 2800,00 2900,00 3000,00 3100,00 3200,00 3300,00 3400,00 3500,00 3600,00 1 2 3 4 $ t

Average prices per quarter t for product 2 (2009-2013)

AUS US NL IR NZ

export subsidies in the dairy markets were just been abolished that had a significant influence on the price development in the dairy commodity markets. In addition, an acceptable amount of data points is now considered in the model (5 years of 12 data points).

Figure 6.1 shows that in the development of purchasing prices a trend can be recognized. It is clear that it is on average cheaper to source in the first quarter than in other quarters. The reasoning behind the lower prices in the first quarter of the year is quite obvious: in the aftermath of the milk production in New Zealand (December until February), a peak in the production of the product 2 takes place, resulting in lower prices for product 2.

6.1 The average price development of SMP per month.

42 6% 11% _{0% 0%} 6% 6% 11% 6% 11% 22% 6% 6% 83% 72% 78% 72% 1 2 3 4 t AUS IR NZ US $1.600,00 $2.100,00 $2.600,00 $3.100,00 $3.600,00 $4.100,00 $4.600,00 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 $ t

AUS US NL IR NZ Minum price extern

Figure 6.2 The price development of product 2 per quarter t for years 2009-2013.

Furthermore, figure 6.3 uses monthly data to show what fraction of time it is the cheapest to source in a certain region. If the sourcing department decides to source externally, it is optimal to use this distribution to spread the order quantities among the regions.

Figure 6.3 Fraction of time that a given region is the cheapest to source for product 2 per quarter t.

43

Besides the purchasing prices, the following inputs to the model are known:

1) The demand forecast Dt. Since the contract date is one quarter before the requested

amount actually arrives, the demand Dt in the sourcing cost optimization model is the

demand at the contract date. For example, in the first quarter t=1, the demand is actually D2.

2) hit is determined by the sum of inventory costs per ton per quarter ($4,56) and the

financing costs of $0,0025c1t that depends on purchasing price c1t. h2t = $75,00

3) ∑ (4_𝑡=1 Ca1t) = 42750 ton and ∑ (4𝑡=1 Ca2t) = 199700 ton.

4) Ce2t differs per quarter due to seasonal production in Ireland and New Zealand and a

probability of drought in New Zealand. Based on historical data derived from the case company, it can be assumed that 1 out of 2 summers (December till February) drought occurs in New Zealand. It is assumed that that no capacity of the external suppliers is then available. The same assumption does apply for reduction in capacity due to the low season months of New Zealand (May till June) and Ireland (December and January). In table 6.2, pit and yit for each region are summarized.

5) s= 11

3 quarter, thus the maximum inventory is Iit ≤ Dt + 1 3𝐷t+1

6) It can be assumed that a sufficient amount of milk is available to produce the product internally. The forecasts milk availability for 2014 is 9480 million kilogram milk. For 1 ton product 2, 10.25 ton milk is required.

t=1 t=2 t=3 t=4 P21 y21 p22 y22 p23 y23 p24 y24 EU 1 1 1 1 1 1 1 1 USA 1 1 1 1 1 1 1 1 SA 1 1 1 1 1 1 1 1 AUS 1 1 1 1 1 1 1 1 NZ 0,333333333 0 0,666666667 0 1 1 0,166667 0 IR 0,333333333 0 1 1 1 1 0,333333 0