Reducing demand uncertainty

Reducing demand uncertainty

in a fabless semiconductor company

University of Groningen

Faculty of Economics & Business

MSC BA Operations & Supply Chains

S.M. Schoot Uiterkamp

Reducing the impact of demand uncertainty at SiTel semiconductor B.V. by

Stephan SchootUiterkamp S1465317

University of Groningen Faculty of Economics & Business

MSc Business Administration Operations & Supply Chains

First supervisor: Dr. ir. H. van de Water Second supervisor: Drs. R.A. Rozier

Company: SiTel (Dialog) Semiconductor Location: s’-Hertogenbosch Company supervisor: Marcel Dinnissen

Management summary

The objective of this thesis is to provide in insight in how SiTel can reduce the negative impact of demand uncertainty, while the inventory costs and delivery performance remain the same or are improved.

To achieve this, the current processes and performance regarding inventory costs and delivery performance have been examined, as well as the impact of demand uncertainty and the inventory costs and delivery performance. Next, the factors causing demand uncertainty have been

determined through interviews and analysis of data and the manner in which each of these factors influence demand uncertainty. After this measures have been researched to reduce the impact of the causes of demand uncertainty;

1. The predictability of the demand pattern. 2. The introduction of new products. 3. The accuracy of shared forecasts. 4. The forecasting bias.

5. Order changes.

Finally the effect of each researched measure on the inventory costs and delivery performance has been determined. The following measures are proposed to reduce the impact of demand

uncertainty:

The predictability of the demand pattern

By reducing the replenishment lead time it becomes possible to react faster to changes in demand. This can be achieved by moving the customer order decoupling point for a selected number of products with high component commonality forward, through the use of

postponement.

The introduction of new products

By effectively communicating any delays in product introduction the inventory for the delayed product can be limited. However because the causes of these delays often lay outside the control of the parties in the supply chain its impact cannot be structurally reduced.

The accuracy of shared forecasts

Improving the trust that exists between SiTel and third parties in the supply chain to increase the amount and value of information that is shared.

Collecting additional information regarding the market developments for end products and derive future demand and trends from this information.

The forecasting bias

Order changes

Freezing part of the production schedule will lower the system nervousness, yet the positive effect of freezing should be weighed off against the result of lower responsiveness.

Holding safety stock increases the responsiveness by decreasing the delivery lead time.

The most important factor in combatting demand uncertainty and its impact is more accurate and timely demand information. Gathering more demand information can improve the delivery performance while the inventory costs will remain the same. However the costs of gathering this information falls outside the scope of this thesis and are therefore not taken into account. However, some data and information is easily accessible and can be stored against little costs.

Preface

Six months of research at SiTel semiconductor have resulted in the master thesis lying before you and the completion of the Master program Business administration - Operations & Supply chains. I would like to thank the following people for the help and advice they provided.

First I would like to thank my university supervisor Hen van de Water, for his feedback and support. His guidance helped me to provide structure and focus in the first stages of writing the thesis and the subsequent chapters. I would also like to thank my second university supervisor Robert Rozier for his feedback.

Secondly I would to thank Marcel Dinnissen for the opportunity to write my master thesis at SiTel and for the all information and feedback he provided. I would also like to thank the colleagues in office 204 for their support and their diligence in answering all my questions.

s’-Hertogenbosch 2011

Table of Contents

1. Introduction 6

1.1 Company background 6

1.2 Operations 7

1.3 The fabless semiconductor market 8

2. Research design 10

2.1 Methodology 13

3. Business processes 17

3.1 Forecasting& planning 17

3.2 The manufacturing process 18

3.3Customers 20

4. Current performance level 22

4.1 Inventory costs 22

4.2 Delivery performance 25

5. Demand uncertainty 28

5.1 Measuring demand uncertainty 28

5.2 The impact of demand uncertainty on current performance 30

5.3 Causes of demand uncertainty 33

5.4 Influence of the causes of demand uncertainty 35

6. Reducing the impact of the factors causing demand uncertainty 41

6.1 Predicting the demand pattern 41

6.2 New product introduction 46

1. Introduction

Semiconductor chips are used in a myriad of electronic devices, ranging from phones to computers to cars. Because the semiconductor market is such a large and varied market some more information about SiTel, the company where the research project is carried out and the market in which they operate is therefore in order. In this chapter an overview is provided of SiTel, the processes which are used to produce a semiconductor chip, and of the market in which SiTel operates.

1.1 Company background

SiTel Semiconductor BV is a company that specializes in the designs and sale of semiconductor chips which are used in wireless applications. They outsource the fabrication of the devices to

manufacturers and are what is known as a ‘fabless’ semiconductor company. Their chips can be used in different wireless applications such as telephones, game controllers, wireless sensors, headsets, baby monitors and other devices. The chips are designed in the Netherlands and Greece.

Once a chip is designed it is produced by one of several manufacturing companies to which SiTel has outsourced production. After that the chip gets assembled. Next to standard assembly, SiTel’s network includes a subcontractor who adds additional modules for specific functions to the chips produced by the manufacturers. Thereby a more specialised version of the standard chip is created. After production the chips are stored in Singapore with JSI, a distribution company to which the warehousing and distribution are outsourced. Finally the chips are sold through the sales offices of SiTel. An overview of the locations and their purpose is shown in figure 1. The SiTel owned sales and design offices are shown in purple, the manufacturing and assembly companies are shown in blue and the distributor in green.

1.2 Operations

SiTel has outsourced all production and assembly processes. A description of each process step required to produce a semiconductor chip is given below.

Manufacturing

The production process is outsourced to four different parties. SiTel has one primary supplier and three secondary suppliers. This is part of their secondary source strategy, which entails that a second source should be available for production when the primary supplier is not able to meet the demand in time. Since 2008,this second source can be either a second factory of the main supplier or a different factory of a third party. The primary supplier is the Taiwan Semiconductor Manufacturing Company or TSMC. TSMC has production plants in the USA and China and supply SiTel with over 80% of their total demand. The secondary suppliers are used as a back-up and for flexibility reasons, they are: Towerjazz semiconductor which has a plant in Israel and National Semiconductor (NSC) with a plant in the USA. Telefunken in Germany supplies SiTel with dies that are used to manufacture the modules later in the process.

Assembly

From the manufacturing locations the products are shipped to UTAC in either Singapore or Shanghai for sorting. SiTel is a medium to large customer to UTAC, and has a reasonable amount of influence to get their wishes seen to. In the sorting process chips are checked and non-functioning chips are discarded. While a second source strategy is followed at the manufacturing stage, SiTel has chosen not to follow such a strategy for the sorting process. They have chosen for this because of the

economies of scale it provides, the sharing of hardware and the improved engineering support that is achieved when only one factory is used. These factors result in lower costs for the single source approach while the improved engineering support ensures that quality is kept high.

Next the chips are assembled by the main supplier UTAC or by Carsem. In the assembly process the same secondary source strategy as with manufacturing is followed. Meaning that UTAC factories are used as backup for other UTAC factories.

Distribution

After the chips are packaged, they are shipped to either the warehouse in Singapore, or to In-tech in Shenzhen to undergo further assembly, such as the placing of the packaged chip on a printed circuit board with additional functionality. After that they are shipped to the warehouse in Singapore. From here it is send to the customer. The warehousing and transport are outsourced to JSI. An overview of the companies to which SiTel has outsourced their processes and the process flow is shown in figure 2. NSC Manufacturing (Maine) Towerjazz Manufacturing (Migdal Haemek) TSMC Manufacturing (Washington, Shanghai) Telefunken Manufacturing (Heilbronn) UTAC Sort (Singapore, Shanghai) Carsem Assembly (Shuozhou) UTAC Assembly (Singapore, Shanghai) UTAC Assembly (Dongguan) UTAC Final test (Singapore, Shanghai) UTAC Final test (Dongguan) JSI Warehouse (Singapore) Customer In-tech Modules (Shenzhen) Delta Modules (Taoyuan)

Figure 2, overview outsourced processes

1.3 The fabless semiconductor market

During the past 30 years, the global semiconductor industry has experienced rapid rates of technological change, rising costs for production capacity, and declining prices for final products (Macher et. al, 2002). This period has also seen an increase in vertical specialisation in semiconductor design and manufacturing.

The specialisation is illustrated by the ‘fabless’ design and marketing companies and the

manufacturing foundry companies that contract for the production of new product designs. The fabless companies are able to offer more innovative designs and shorter delivery times than so-called merchant semiconductor companies. Merchant semiconductor companies are manufacturers that sell most of the semiconductors they manufacture to unrelated buyers (Macher, 2002). These more innovative designs and reduced lead times are achieved by reducing the design cycles through the reuse of design components (known as intellectual property blocks) and the subcontracting of design services (Macher, 2001).

50% and 90%. In the IAD and VOIP segment, SiTel is a small player. Demand in these segments is rather unpredictable, even more so when new products are implemented. Since the average lifecycle of semiconductor products are about 2 years, new products are introduced regularly. These short product life cycles increase the risk of obsolescence and the cost of excess supplies (Fisher, 1997). Because manufacturing is outsourced even more insecurity is added to the market due to the increased risk of lost control. The most important factors contributing to this risk in the fabless semiconductor market are the lack of demand visibility, uncertainty and assurity of supply (capacity) (Brandell, 2007). In the market in which SiTel operates, retailers with direct access to the customer generally have about 7 days of demand visibility, which makes it difficult to predict demand. This demand uncertainty increases farther upstream of the supply chain as visibility decreases. To be able to deal with the lack of visibility and uncertainty, flexible manufacturing capacity is required.

However, because manufacturing has been outsourced this flexibility is limited. When the

manufacturing capacity has to be changed this must be negotiated several months in advance. Even then only limited changes are possible. Due to this lack of control and limited visibility as well as the short life cycles and technological changes, forecasting is a key area in achieving supply chain success (Macher& Mowery, 2009). While it is clear that most of these factors limit SiTel’s ability to meet customer demand, it is not yet clear how great their impact is. However, the recognition of these problem areas provides a starting point for the framework of the research design.

2. Research design

A major complicating factor in the management of supply chains performance is uncertainty in the supply chain. This is even more so in the fabless semiconductor market, since outsourcing has only increased uncertainty. However for most semiconductor companies it is simply not possible to produce in-house because a plant to manufacture and assemble chips costs over a billion dollars. Most companies therefore have no choice but to deal with the additional uncertainty in the supply chain created by outsourcing. According to Acar et al. (2010) uncertainty in the supply chain is composed of supply uncertainty, demand uncertainty and transportation lead time uncertainty. SiTel sells their product in a market that has short product lifecycles and rapid technological changes, making demand unpredictable (Brandell, 2007). Manufacturing is outsourced which makes it more difficult to quickly change the available level of production capacity, thereby creating supply

uncertainty. These supply and demand uncertainties make planning harder; in turn result in changing transportation lead times. However SiTel has outsourced transportation to large transport companies such as FedEx. These companies have enough capacity to be flexible and semiconductor chips do not take up much space and as such any changes in transport dates provided no problems. Of demand and supply uncertainty, demand uncertainty has unquestionably the most negative impact on both cost and customer service performance (Acar et al., 2010).

According to Epstein (1999), uncertainty refers to situations where the information available to the decision-maker is too imprecise to be summarized by a probability measure. While it would be interesting to research all three types of uncertainty, due to time constraints it would be more effective to focus in depth on one type of uncertainty. In several interviews SiTel has indicated that they consider costs and delivery performance as important performance factors in their market. Delivery performance defined as the combination of the number of products delivered on the customers desired delivery date divided by the number of products ordered, the number of products delivered on the suppliers confirmed delivery date divided by the number of confirmed products and the average weighted lateness of the delivered orders (Schönsleben, 2007) (Acar et al., 2010). Because demand uncertainty has the most negative impact on cost and the customer service

performance of the different types of uncertainty (Acar et al., 2010), this thesis will focus on demand uncertainty. Demand uncertainty is defined by Ambrose et al. (2006) as a lack of information

concerning demand levels or specifications for goods or services. The uncertainty can arise from lack of data, or from data not being available soon enough to allow for effective planning of operations. According to Speh & Wagenheim (1978) demand uncertainty is related to sales or demand which materialized randomly either in excess or short of forecasted sales for a specified time period. Information or data about the expected demand becomes more important when demand has a tendency to fluctuate and becomes less predictable. From these definitions we can conclude that demand uncertainty has two aspects: the fluctuations in demand and the ability to detect these fluctuations timely and accurately.

increasing the safety lead time. Safety lead time is defined by Hariharan and Zipkin (1995) as the difference between the release time and the due date minus the supply lead time of the product, where supply lead time is defined as the time that is required to produce the order. This will lead to (artificial) increased lead times, or safety lead time, which will increase the likelihood that an order is delivered on time. That in turn will lead to improved delivery performance (van Kampen et al., 2010), but it can also drive away customers who might not be willing to accept the increased delivery lead time.

However, when looking at costs, not all costs need to be taken into account. Only the costs that are affected by demand uncertainty fall within the scope of this research. The total costs are made up of administration costs, manufacturing costs, logistics costs and inventory costs (Simchi-Levi, 2008). The administration costs are the costs made to release and complete an order (Schönsleben, 2007). SiTel releases orders according to the ‘first come, first serve’ rule. This means that order rescheduling is limited. The administration costs are therefore unlikely to change under demand uncertainty because of a lack of demand information will not cause changes in the time needed to release and complete orders. Therefore these will not be taken into account in this thesis.

Manufacturing costs are the costs made to produce, assemble and test the semiconductor chips. The manufacturing costs are dependent upon the capacity that SiTel (can) buy(s) from their suppliers. However capacity availability and their costs are defined as supply uncertainty (Acar et al., 2010) and as such are not part of the scope of this thesis. However the effect that the availability of capacity has on the inventory costs, such as the holding of additional safety stock, will be taken into account. The transportation costs depend upon the number of orders that need to be shipped. Because the current outsourced transportation capacity is flexible enough to deal with any fluctuations in

demand, no extra costs are incurred to cope with demand uncertainty. Therefore the transportation costs will not be taken into account.

The inventory costs are made up of the carrying costs, depreciation costs and storage infrastructure costs (Schönsleben, 2007). However because inventory handling is outsourced the infrastructure costs that are incurred are included into the carrying costs and will not be taken into account separately.

Inventory costs performance and delivery performance will be included as performance indicators in this research. In order to determine how best to deal with the situation of demand uncertainty and both inventory costs and delivery performance the following research objective has been

formulated:

Provide insight in how SiTel can reduce the negative impact of demand uncertainty while the inventory costs and the delivery performance remain the same or are improved.

any possible actions might have on the cost and delivery performance will be measured as well. Based on these objectives and limitations the following research question is formulated:

Which factors cause demand uncertainty and how can SiTel reduce the influence that these factors have on demand uncertainty while maintaining or improving their inventory costs and delivery performance?

To answer this research question, several sub questions are needed. First an overview of the current situation is required to provide a starting point from which possible improvements can be made.

1.

To understand SiTel’s current ability to meet customer demand and the effect that any future changes might have on the inventory costs and delivery performance the current processes and the current inventory costs and delivery performance level will need to be explored. More knowledge about the operations will make it easier to discover the impact of possible causes of demand uncertainty and possible solutions. This leads to the following sub question:

2.

To be able to reduce the demand uncertainty it must first become clear which factors are causing the uncertainty and what the impact of these factors on the inventory costs and delivery performance is. Once the factors causing of demand uncertainty are identified, the strength of the relationship between these factors and demand uncertainty needs to be researched. An appropriate approach to deal with each factor causing demand uncertainty must also be found. This leads to the third sub question.

3.

2.1 Methodology

In this section the methods that are used to gather data are described and the reliability and validity of this study will be discussed.

Data has been collected from several resources. The research project started with the reading of scientific literature. The literature provided more knowledge about factors related with demand uncertainty and helped to highlight areas on which more information was needed. Secondly the available data stored by SiTel in their internal data warehouse, SAP business warehouse, regarding delivery reliability, forecasting and other factors was analysed. Finally interviews were performed to both confirm data from the data warehouse and to fill in data gaps.

The most important source of data for this thesis has been the quantitative data from the data warehouse. Because the majority of the available data has been quantitative and because it provides an objective source of data, it has been used as the basis to determine the current performance and impact of demand uncertainty. The quantitative data has subsequently been crosschecked against (subjective) qualitative data provided by the interviews.

In order to be able to propose improvements to either decrease the impact of demand uncertainty or increase SiTel’s ability to deal with the impact of demand uncertainty, it is important to understand which processes are important for SiTel to meet customer demand. Therefore interviews have been conducted to create an overview of SiTel’s current processes. The processes that have been

described are selected based on factors that have been identified in scientific literature and from the before mentioned interviews as factors that could affect SiTel’s ability to deal with the impact of demand uncertainty such as the flexibility of production capacity, the length of the delivery lead time and the sharing of information between parties in the supply chain.

The current demand uncertainty, inventory costs and delivery performance have been explored by going through booking, costs and delivery data of the last three years stored in the data warehouse. Based on the definitions of these concepts provided by scientific literature, data has been selected to calculate their performance. Finally the results of these calculations have been used to determine whether there are causal relationships between the demand uncertainty, inventory costs and delivery performance and how changes in these factors might affect each other. This has been done by calculating the statistical correlation between these factors.

To be able to reduce the impact of demand uncertainty the factors which are causing demand uncertainty first have to be identified. For this both interviews and a questionnaire have been used. The questionnaire that is used is based on Ho et al. (2005) and can be found in appendix A. The questionnaire was send to sales and marketing personnel, the deputy direction of supply chain and manufacturing and the director of the short range wireless business line. Based on an average score of the received answers and information gathered from interviews the factors causing demand uncertainty have been identified.

Finally this knowledge was applied to fit SiTel’s specific situation. This in turn has led to proposals to reduce the impact of demand uncertainty for SiTel.

Besides these proposals, alternative options to increase SiTel’s ability to deal with demand uncertainty have been suggested. This has been done because it is not always possible for SiTel to reduce the impact of demand uncertainty as some of the causes are outside their control. An increase in SiTel’s ability to deal with demand uncertainty will also, just as reducing the impact of demand uncertainty does, improve SiTel’s performance. The options that are proposed have been selected based on scientific literature and on the descriptions of SiTel’s processes. An overview of the main variables in this research project and their expected relations can be seen in figure 3.

The variables in this framework are selected based on interviews and scientific literature. The decision to focus on reducing the impact of demand uncertainty instead of eliminating on reducing demand uncertainty itself was made because most factors causing demand uncertainty are structural factors innate to the semiconductor market. The goal of exploring these variables and their relations is to reduce the impact of demand uncertainty.

of the inventory costs as SiTel will try to maintain their delivery performance. The same increase of demand uncertainty will likely lead to a decrease in delivery performance as it becomes more

difficult to forecast demand accurately. However this can be compensated by holding more inventory to reduce the impact of demand uncertainty on delivery performance.

Based on scientific literature there are two factors that are expected to reduce the impact of demand uncertainty on inventory costs and delivery performance. The first factor is more information about demand. By increasing the availability of demand information there will be more insight into the causes of demand uncertainty. By improving insight into these causes it becomes possible to address and reduce the influence that these causes have on demand uncertainty and as such reduce demand uncertainty itself. The impact of demand uncertainty will be reduced as demand uncertainty is reduced. The second factor is the ability to deal with the impact of demand uncertainty. When it is not possible to reduce the influence of a cause of demand SiTel can increase their ability to deal with or absorb the impact. By creating a buffer, either in time or inventory, it is possible to absorb the impact and as such improve delivery performance. However when the impact is absorbed by creating an inventory buffer the inventory costs are likely to remain equal or increase.

2.1.2 Reliability and validity

In order to make sure that the factors that are being measured accurately reflect the actual situation, the measurements need to be both reliable and valid. Reliability will be discussed first and the extent to which a study’s operations can be repeated with the same results. Secondly, the validity of the measurements or the extent to which differences that are measured actually reflect true differences is discussed.

Reliability

According to Cooper and Schindler (2003) there are four potential sources of measurement errors that can affect the reliability. The respondent, the situation, the measurer and the data collection instrument can all contaminate the research outcome. Each potential source of error and its application to the research conducted for this thesis will be discussed shortly.

A respondent’s ability to respond accurately and fully may be limited due to fatigue, anxiousness, distractions, impatience, unwilling to express certain opinions or by reluctance to admit that he or she does not know the answer to a question. The ability to cross-check answers of the respondents to the data stored in the data warehouse as well as with the answers provided by the other

respondents makes the respondent an unlikely source of error in SiTel.

Situational factors such as the presence of a third party or expected blowback from certain answers can affect the interviewer-respondent rapport. During the conducted interviews all topics were discussed openly. There were other persons present during some of the shorter interviews, but since the short interviews mainly served to clarify previously discussed topics the chance of errors due to situational factors is low.

detail, with other persons and also when the interpreted answer was applied to the available data, it is possible that some may have slipped by unnoticed. As such the measurer is a possible source of error.

The instrument or the way in which information is shared, can cause errors by being confusing and ambiguous or by a poor selection of content. Since most information was gathered through open interviews any ambiguity could be directly addressed. Also the cross-checking of gathered

information through other interviews and the use of the data warehouse reduce the possibility of errors through ambiguity. Because the factors affecting demand uncertainty have been researched thoroughly with the help of scientific literature it is unlikely that any important issues have been omitted.

Validity

Besides reliability, the validity of the measurements that are being used is important to ensure that what we want to be measured is actually measured. This research pays attention to two types of validity between which Cooper and Schindler (2003) distinguish; content validity and construct validity.

Content validity measures the degree to which the content of the items adequately represents the universe of all relevant items under study. By basing the measurements that have been used on numerous articles of the current scientific literature concerning demand uncertainty, inventory costs and delivery performance a high content validity can be confirmed.

Construct validity is the extent to which correct operational measures are established for the concept being studied (Voss et al, 2002). The use of several sources of evidence from literature, interviews and data have been checked by respondents and provide a solid basis from which construct validity can be confirmed.

Interpretation of the results

When considering the reliability and validity of the measures that are proposed at the end of this thesis to reduce the influence of the factors which are causing demand uncertainty, it should be taken into account that these measures are based on literature and not on practice. The current situation of SiTel has been taken into account while proposing the measures, yet as none of these are implemented into practice before this thesis is completed it is difficult to tell whether these results of these proposals will be as expected. As such the reliability and validity of these measures and their possible outcomes should be doubted as they are just that; possible outcomes, and should

3. Business processes

In this chapter the current processes of SiTel will be described. Based on the descriptions of these processes, several uncertainties in meeting demand, caused by the organization of the processes can be identified. This information in turn serves as input for the development of methods which either decrease these uncertainties in meeting demand or increase SiTel’s ability to handle them.

According to Slack & Lewis (2008) an important factor in SiTel’s ability to meet customer demand is their production capacity and the way this capacity is managed. Therefore the methods by which the expected capacity requirements needed to meet customer demand are forecasted and planned are discussed. After this an overview of the production process and production capacity will be given. Here SiTel’s control over the production capacity and the effect of outsourcing on the production capacity is also discussed. Thirdly the behaviour and importance of SiTel’s end customers is discussed to provide a better understanding of what can be expected in terms of customer demand. The input of these customers into the planning and forecasting process is described as well. The end customers are divided into three different segments based on their sales volume and negotiation behaviour.

3.1 Forecasting& planning

SiTel uses a six month rolling forecast to plan their orders. The basic forecast for each month is based on the incoming orders and the expected orders minus the available inventory. The basic forecast is adjusted by a forecaster based on knowledge of the market, customer demand fluctuation and trends. Once a month, the forecast is send to the wafer manufacturers and to the assembly, module and test companies. This is done at predetermined times in order to match the planning cycle of these suppliers. SiTel uses two methods to improve delivery performance:

Consignment stock

Consignment stock is merchandise which is stored at the customer’s site, but which is owned by the SiTel. The customer is not obliged to pay for the merchandise until they remove it from consignment stock or until 45 days have passed. The customer has a pull obligation which means that return of consignment stock is not possible. The use of consignment stock reduces the total lead time from SiTel to the customer which allows orders to be placed later. Because there is no need to take demand during the production time into account, the use of consignment stock also provides more accurate information about the actual customer demand.

Customer buffer stock

Buffer stock is customer inventory which is reserved by a customer. For this reservation the customer pays a service charge made up of; the warehouse storage, handling and management cost, which is usually about 3% of the average sales value of the materials on stock. Holding buffer stock reduces the lead time, allowing orders to be placed later.

can be ignored to allow a rush order priority over an already planned order when the changes this causes are judged to be acceptable by the planner.

After the orders have been received they are scheduled using forward planning. Forward planning allows for a period of time in which some changes can be made to the planning schedule while orders can still be delivered on time. It regularly occurs that customers want to change either the order request date or the number of chips they have ordered. This is usually possible, but any changes that affect the manufacturers must be communicated two weeks in advance by SiTel. This means that no more changes can be made two weeks before production begins. The complete scheduling process, from forecast to planned orders, takes around 4 weeks. After that the manufacturing process begins.

3.2 The manufacturing process

The manufacturing process that is used to create chips is consists of four basic processes. Sometimes a fifth process is added in which specialized modules are placed on a chip to add additional

functionalities.

The production process

1. Wafer fabrication.

A wafer is a thin, round slice of semiconductor material, usually made from silicon. In the wafer fabrication process, silicon, in the form of wafers, is transformed into integrated circuits, on which different layers and patterns of metals are constructed. SiTel produces chips with three different sizes of semiconductor strips. Two of the three sizes are also available with flash memory. Chips with a flash memory take longer to produce than “regular” chips. The wafer production process lasts 40 days for a regular chip and 51 days for a flash chip. The customer order decoupling point is located in this process and is based on the point of differentiation of the chips. For regular chips it is located at 27 days and for flash chips it is located at 38 days.

2. Sorting.

In the sorting process the dies are inspected and defect dies are removed. This process takes around 7 days.

3. Assembly.

Here the individual dies are sawn from the wafer and final assembly and packaging takes place, which takes around 14 days to complete.

4. Testing.

The chips are tested to determine if there are faults in their connections or in other aspects. This process takes 7 days.

An additional fifth process is sometimes used to add modules to chips. These modules add additional functionality to the chips and can take up to 42 days of lead time. Modules are added to only a small part of all chips.

SiTel currently follows a make-to-order strategy, were production is initiated based on customer orders (full start) or based on a forecast which is used to resupply a forward inventory location in the wafer fabrication process. At this inventory location, called the romstore, wafers on which a number of general production steps have been performed are stored to reduce the delivery lead time. From the romstore inventory products are processed further upon receiving a customer order. The production process with the customer order decoupling point at the Romstore location in the wafer fabrication process can be seen in figure 4.

Wafer fabrication 40 days/ 51 days (flash) Sorting 7 days Assembly 14 days

1 day 1 day 1 day

Romstore 27 days/

38 days (flash Wafer

fabrication 14 days Sorting

Warehouse Customer Customer

order 2 days

1 day/

2-4 days Shanghai 2-4 days

CODP location Testing 7 days Modules 42 days 1 day 4 days

Figure 4, Manufacturing process and processing times

Production capacity

SiTel’s outsources its production and is depended on their suppliers to supply them with the production they need. To increase the production capacities flexibility and reduce the risk of

disruption SiTel follows a secondary source procurement strategy, whereby a secondary supplier acts as backup in case the main supplier is unable to deliver when needed.

Each step in the production process has a main supplier and a secondary supplier. The most important of the main suppliers are TSMC which produces wafers and UTAC which does most assembly and testing. These are both large international companies for who SiTel is only a small customer. Because SiTel represents only a small sales volume to these companies they have a relatively weak negotiation position. It can therefore be difficult for SiTel to see that their capacity demands are met.

changes can still be made fairly easy; as long as the suppliers are notified twee weeks in advance and they do not conflict with product specific capacity.

It is possible that the demand forecast fluctuates significantly per month. Because SiTel is depended upon their suppliers for their production capacity, who in turn have to balance the capacity demands of all their customers, it is not always possible to deal with increases in demand as they occur. Sitel can in theory easily switch over to their secondary source suppliers when their main supplier cannot supply them with sufficient production capacity. However a new factory has to be qualified before a customer will accept their products. This qualification process is performed to ensure that the products made by the supplier are up to the customer’s standards. This qualification process is an expensive process which can last anything from 30 till 100 days to complete. Because of the length and costs of this process, suppliers are not qualified in advance by SiTel’s customers. From the perspective of SiTel’s customers it is easier to either delay the order for the required products one month or to order them from another company who can supply them in time. The qualification process significantly reduces the capacity flexibility that the secondary source strategy could provide. No qualification is required to switch test locations. However the testing hardware between different locations must be correlated before the test results can be guaranteed. This quasi qualification process creates additional costs and takes about 3 months to complete, creating another barrier in the use of the second source strategy.

These qualification and correlation processes reduce the capacity flexibility and make it difficult to deal effectively with large fluctuations in demand. To be able to react better to these demand fluctuations SiTel shares their forecast, background information regarding the forecast and product expectations with their suppliers.

The orders that SiTel receives are different per customer group. Each group and their ordering behaviour will be discussed shortly.

3.3Customers

The end customers of SiTel can be divided into three groups, based on the size of the customer, which is measured by looking at the potential and actual sales they represent for SiTel, and their negotiation power and behaviour.

The first group are the large customers with a sales volume of more than $40 million each. Together they represent 72% of the total sales volume. SiTel has relatively little negotiation power when dealing with these large customers. These customers have a dominant negotiation strategy, whereby SiTel is given the specifications to which they must adhere. An example of this is the maximum delivery lead times which are set by some of the large customers.

The third group is made up of a large number of small sized customers with a sales volume of less then $2,5 million each, who represent about 13% of the total sales volume. These customers have a ‘follower’ negotiation strategy. This means that they do not make specials demands or special specifications but buy products under standard terms.

Information sharing

Most large and medium customers share information with SiTel in the form of forecasts. However it is next to impossible to determine the accuracy of these forecasts. Since SiTel’s customers are located further down in the supply chain then SiTel, it could be expected that their forecasts are more accurate. But due to limited demand visibility, which is only about 7 days for the retailers that SiTel’s customers supply, the accuracy of these forecasts is not very dependable. Besides the limited demand visibility SiTel also has to take artificial increases of the forecast into account. Expected demand might not materialize and the difference between the expected and actual demand can lead to fluctuating forecasts which do not correspond to actual demand from the customers and as such send a false signal to SiTel.

The information that SiTel receives from their customers, besides a forecast, has decreased over the last couple of years. The knowledge about customers, transferred to suppliers by information

4. Current performance level

To be able to determine the inventory costs, delivery performance and demand uncertainty SiTel faces and whether or not they are related, the current performance level will have to be determined. The performance of these factors also provides an additional indication of problems areas, which can support or refute qualitative information. After the causes of these performances and their

correlation are known it becomes possible to suggest improvements.

The current performance level will be determined by measuring the inventory costs and delivery performance. To create a reliable overview of the performance level, data from 2008 to 2010 are examined. 2008 is chosen as the lower limit because the 2007 data is incomplete and not 100% reliable according to SiTel. Therefore the 2007 will not be included. The inventory costs are measured by calculating the inventory carrying costs per month, the inventory depreciation costs and the inventory turnover. The delivery performance will be measured by calculating the customer service ratio or RLIP, the delivery reliability rate or CLIP and the average weighted lateness of the delivered orders(Schönsleben, 2007) (Acar et al., 2010)

4.1 Inventory costs

According to Schönsleben (2007) inventory costs can be measured by the following performance indicators:

The inventory carrying costs. This includes:

o The cost for the work-in-process (WIP) and finished goods inventory.

o The costs of financing or capital costs, which is the cost of immobilizing money in inventory.

The inventory depreciation costs. This includes:

o Technical obsolescence. This results from changes in standards or the emergence of improved products on the market.

o Perishability, certain items can be stored only for a particular limited period of time. This is the case with ‘living’ products such as food or biological pharmaceuticals, but also with ‘non-living’ products such as certain electronics items.

o Damage, spoilage, or destruction due to unsuitable handling or storage such as the rusting of sheet metals.

Each of these factors will be discussed in turn. Inventory carrying costs

Wafer fabrication 40 days/ 51 days (flash) Sorting 7 days Assembly 14 days

1 day 1 day 1 day

Romstore 27 days/

38 days (flash Wafer

fabrication 14 days Sorting

Customer order decoupling point Wafer inventory Wafer inventory Die inventory Warehouse Customer Customer order 2 days 1 day/

2-4 days Shanghai 2-4 days Consigned/ finished goods

inventory CODP location Testing 7 days Modules 42 days 1 day 4 days

Figure 5, Inventory points in the manufacturing process

SiTel only pays inventory carrying costs for inventory that is stored at the warehouse. For inventory that is held at any other point no carrying costs are charged. However there are limits set by these manufacturers determining how much space the WIP is allowed to take up, the maximum value of the WIP and the minimal turnover of the WIP. When products are stored at the warehouse, SiTel only has to pay storage and handling costs for products that are stored there for longer then 2 months. For these products the storage and handling costs are based on the pallet spaced that is being used.

In august 2008 a reduction in inventory carrying costs was achieved by focussing on ‘customer specific production’. SiTel negotiated agreements about the minimum order quantity with customers of who the expected demand for the next two years was less then the twice the minimum lotsize. The relatively low order quantities of these customers meant that the risk for excess inventory was relatively high. By making the minimum order lotsize equal to the minimum productionlotsize, the excess inventory andthe costs of carrying it have declined. This can be seen in figure 6.

Figure 6, Inventory warehousing costs.

$0 $20.000 $40.000 $60.000 $80.000 $100.000 $120.000

The costs of financing inventory

Inventory ties up financial resources. The money that is used to produce the (work-in-process) inventory cannot be used for other purposes until the inventory has been sold. To calculate the cost of the money that is tied up in inventory a percentage of the mean return on investment will be applied when internal capital resources have been used to finance the inventory. These costs represent the possible amount of money that could have been earned if the money had not been tied up in inventory. If the inventory is financed by a third party, the bank interest rate which represents the additional cost of the lending of the money tied up in the inventory, is used to calculate the costs of financing inventory. SiTel follows an inventory policy in which they try to minimize the inventory that they hold. They attempt to hold inventory only at the romstore location. The amount of WIP held at the romstore location is based on the forecast plus excess stock from inaccuracies from previous forecasts. However, due to differences between the minimum order quantity and minimum production quantity there are ‘left over’ finished goods that are kept in stock. Beside the regular WIP and inventory, SiTel is contractually obligated to hold safety stock for

Microsoft. This safety stock has a value close to $2 million. In figure 7 the location of the inventory and inventory value for the years 2008 to 2010 can be seen.

Figure 7, Inventory value

SiTel uses a bank loan with 12% interest a year to finance their inventory. This means that the carrying costs of inventory are 1% of the total inventory value per month.

In November ’09 there is no longer any value assigned to the inventory at the romstore location. This is because, from this point onward, the value of the romstore inventory is assigned at the moment the wafer is completed and not longer at the romstore point.

Inventory depreciation costs

A total of the inventory costs, consisting of the obsolescence costs, warehousing costs and financing costs can be seen in figure 8.

Figure 8, Inventory cost per month

4.2 Delivery performance

According to Schönsleben (2007) delivery performance can be measured by looking at: The fill rate or customer service ratio

The delivery reliability rate.

The fill rate or customer service ratio is the percentage of products that is delivered on time according to the delivery date requested by the customer. It is measured by dividing the number of products delivered on the desired delivery date by the number of products ordered. It is called the Requested Line Item Performance (RLIP) within SiTel.

The delivery reliability rate is the percentage of products that is delivered on time according to the delivery date promised by SiTel. This delivery date is set based on the lead time required to produce and deliver the order. It is measured by dividing the number of products delivered on the confirmed delivery date by the number of confirmed products. It is called the Confirmed Line Item Performance (CLIP) within SiTel (Schönsleben, 2007).

These performance indicators will show whether an order has been delivered on time. However, besides knowing whether an order is delivered on time it is also important to know that when an order has been delivered to late, how much too late that order was delivered and, in the case an order is broken up into different suborders, what percentage of that order has been delivered to late. For example, an order of which 90% is delivered one day to late will get a RLIP or CLIP score of 10%, but an order of which 90% that is more then a week to late get’s the same score. To

differentiate between this the average weighted lateness is measured. An overview of the current performance and the target performance is shown in figure 9.

Figure 9, RLIP & CLIP performance

The RLIP and CLIP generally follow the same course although not at the same percentage level. The importance of the comparison between the LRIP and CLIP scores lies in their differences. A decrease in the RLIP line compared to he CLIP line signifies a likely external cause for the difference, while a decrease in the CLIP compared to the RLIP signifies a likely internal and thus possibly preventable cause.

Average weighted lateness

The average weighted lateness is measured by taking the percentage of each order that is delivered late (orders can consists of more than one delivery) according to the delivery date requested by the customer (RLIP). The percentage that is delivered late is multiplied by the number of days it is delivered late. The results are shown in figure 10.

Figure 10, the average weighted lateness of late deliveries

The average weighted lateness tends to fluctuate around 6 days, with a deviation of around 2 days. When this figure is compared to the RLIP score, a relation between the deviation in the average weighted lateness and the RLIP becomes apparent. An increase of the lateness to 8 days appears to be related to a reduction of the RLIP of around 10%, while a decrease of the lateness to 4 days appears relate to an increase of the RLIP of 10%. Based on this data a reduction of the production lead time of 2 days on average could result in an RLIP score close to the 85% target.

50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100%

RLIP & CLIP performance

RLIP target rlip CLIP target clip 0 2 4 6 8 10 12 D a y s l a t e

Summary

Inventory costs are made up of carrying costs and depreciation costs. The majority of the structural inventory costs are made up of the costs of financing inventory. Because SiTel can store a limited amount of WIP for free at their suppliers, these costs are almost non-existent. Another source of costs are the obsolete inventory costs. Because customers stopped the production of certain products, any leftover inventory of these products were made obsolete and as a result had to be written off as expenses.

SiTel measures delivery performance through the use of a customer service ratio (RLIP) and the delivery reliability rate (CLIP). The customer service ratio measures the percentage of products that are delivered on time according to the delivery date requested by the customer. The delivery reliability rate is the percentage of products that is delivered on time according to the delivery date promised by SiTel.

5. Demand uncertainty

In this chapter the construct demand uncertainty and its impact on SiTel’s current performance will be will be discussed. This will be done by measuring demand uncertainty and determining whether a statistical causal correlation exists between demand uncertainty and the performance factors used to determine the inventory costs and delivery performance. Secondly the factors that are causing demand uncertainty will be examined.

5.1 Measuring demand uncertainty

According to Jeunet (2006) the expected demand (d^) is made up of the demand pattern (d), the fluctuation of demand ( ) around the demand pattern and the ‘pure’ forecast error ( .

The pure forecast error includes the choice of an inaccurate forecasting method and errors in identifying and estimating the parameters of the demand pattern. From this formula we can conclude that any difference between the demand pattern and expected demand, usually taken as the difference between the actual value and the expected value is caused by demand fluctuation and the pure forecast error.

Thus, there are two sources of uncertainty in the demand prediction process. The first source is the forecasting error, which can be reduced by getting closer to the optimal forecasting method for the underlying demand pattern. The second source, the process error, is the random variation in the demand pattern itself. This random variation contrasts with the systematic variability due for instance to phenomena like seasonality and trend. This random demand can be reduced only by attempting to manage the demand process, for example by attempting to change customers’ behaviour or through collaborative forecasting (Fildes & Kingsman, 2011).From this it can be concluded that demand uncertainty has two aspects: the fluctuations in demand ( and the ability to detect these fluctuations timely and accurately ( .

The fluctuations in demand can be measured by the coefficient of variation or COV (Milliken, 2006). The coefficient of variation is a measure of relative dispersion in probability distribution and is calculated with the following formula:

COV=

Figure 11, Coefficient of variation of customer demand.

The figure shows that the COV of demand in SiTel lays around 150%. This means that the value of the orders that are received each month vary with as much as 150% from the average order value. However, there variation in demand could be caused by just a few difficult products which are difficult to forecast. Therefore the COV per product group will be examined. According to Milliken (2006) a COV equal or less than 80% is manageable. The COV per product family for the period of 2008 till 2010 is shown in figure 12. In the case of SiTel, only three product families; 11, 46 and 51 out of 30, have a ‘manageable’ COV lower than 80%.This shows that SiTel has to deal with a high level of demand fluctuation over its entire product range, represented by the average COV of demand per month of 115%, rather than a distortion caused by one or two product families.

Figure 12, Coefficient of variation of demand per product family

According to Jeunet (2006) the best or optimal forecasting technique is usually the one that gives the closest results between the forecasted demand and the actual value, based on historical sales figures. The second source of demand uncertainty, the forecast error is the deviation of the actual from the forecasted order quantity. This is measured by subtracting the forecasted order quantity from the actual order quantity divided by the actual order value.

0% 50% 100% 150% 200% 250% 300% 350%

Coefficient of variation

The result of these calculations is shown in figure 13.

Figure 13, Forecast error

In SiTel the forecast two months prior to the ‘romstore production start’ is used for every planning cycle. This forecast is used due of the longer lead time (4-5 weeks) of products that can not be supplied directly from the romstore and need a ‘full start’ or to resupply the romstore itself. Demand uncertainty is pointed to as cause of increased inventory costs and increases in both the weighted lateness of orders and number of orders that are delivered late (Acar et al., 2010) (Davis, 1993) (van Kampen et al, 2010). To discover if demand uncertainty also has this impact at SiTel, the correlation between demand uncertainty and the performance factors of inventory costs and delivery performance will be calculated.

5.2 The impact of demand uncertainty on current performance

In this paragraph the impact of demand uncertainty, represented by the COV and the forecast error, on the current performance will be measured. The current performance will be represented by the RLIP, the CLIP and the inventory cost. The inventory costs are divided into two groups, the inventory costs with and the inventory cost without obsolete inventory. This is done because the date of obsoletion cannot be linked to a forecast or start of production date. A comparison between COV and inventory costs could therefore become distorted due to a write-off of obsolete inventory. To remove this distortion the COV will also be compared to the inventory cost with obsolete inventory costs.

When two factors are compared with each other, the time that has passed between the

measurements of each factor needs be taken into consideration. For example, when the COV is being compared to delivery performance, the two months it takes a received order to be produced and delivered means that the COV of the orders received in January should be compared to the delivery performance of these orders in March. In this case of the forecast error two more months are

required, since the duration of the forecast and subsequent romstore resupply or full production start is one month for and the planning of the customer orders which are produced from the romstore supply last another month. Because measurements are taken from 2008 to 2010, this also changes the sample population to respectively 34 and 32.

The COV is compared to the RLIP, the CLIP and the inventory cost using statistical correlation. A positive result or significant correlation between two variables indicates that there is a relation between these variables. The correlation coefficients are shown table 1.

Demand uncertainty Performance indicators

Coefficient of variation Forecast error

Inventory costs 0.176 -0.283

Inventory costs without depreciation 0.615* -0.454*

RLIP -0.191 -0.148

CLIP 0.096 -0.296

Average weighted lateness 0.057 0.134

Table 1, the correlation between demand uncertainty and performance indicators.

* Significant at 1% level

Based on a 1% level of error, the correlation between the COV and the inventory costs without depreciation is significant. This means that an increase in demand variation will result in an increase in the inventory costs without depreciation 99% of the time. This increase in inventory costs without depreciation is most likely caused by an increase in WIP held at the romstore location to compensate for the process error to ensure that the percentage of on-time delivery remains acceptable to customers and by additional excess finished goods inventory that is held. The finished goods inventory increases due to the difference between the minimum order quantity (MOQ) and minimum production quantity (MPQ), whereby the MOQ is smaller than the MPQ. This difference leads to residual finished goods inventory after the order quantity has been supplied. An increase in the COV will increase the average time this residual finished goods inventory is held, thus increasing the inventory costs without depreciation.

The correlation between the inventory costs without depreciation and the forecast error is significant at the 1% level. This correlation indicates that a reduction in the forecast error will lead to increase inventory costs without depreciation or that an increase in the forecast error will lead to a decrease in inventory costs without depreciation. This appears to be an unexpected result at first. However further research shows that each increase in the forecast error is the result of a decrease in demand. Because the forecast does not follow, or only slowly follows a reduction in demand and remains on the same level for some time, each decrease in demand leads to a relative increase in the forecast error. This slow response could be the result of either a delay in information that SiTel receives from their customers or a bias on the part of the forecaster.

actual demand over the last three years. Because of this forecast bias there is usually enough stock in the romstore location to meet demand. The forecast and actual demand are shown in figure 14.

Figure 14, forecast and actual demand

From this figure we can conclude the actual demand increases significantly in June and July each year. A small increase can also be detected in March. However the actual demand during these periods differs per year. The difficulty in predicating the scale of these increases seems to be responsible for a large part of the forecast bias.

In the following paragraph, the factors which cause demand uncertainty within SiTel will be discussed. 0 2.000 4.000 6.000 8.000 10.000 12.000 14.000 16.000 18.000 20.000

Forecasted and actual demand

5.3 Causes of demand uncertainty

According to Ambrose et al. (2006) a lack of information concerning demand levels or specifications of goods or services can lead to uncertainty in demand. The uncertainty can arise from lack of data, or from data not being available soon enough to allow for effective planning of operations. However to be able to reduce the uncertainty in demand the causes of the lack of data must be researched. According to Ho et al. (2005) demand uncertainty can be caused or affected by a number of factors:

1. The number of sales channels – demand will be more diffused when there are more sales channels.

2. The heterogeneity of channels – lower channel heterogeneity leads to a greater impact of customer characteristics on demand and increases demand uncertainty. A greater variety of channels also leads to higher demand uncertainty.

3. The frequency of channel or customer replacement – more frequent customer replacement corresponds to unstable channel cooperation, and would cause higher demand uncertainty. 4. The rate of new product introduction – new products without a sales history are more

difficult to predict.

5. The product lifecycle – uncertainty is greater in markets with shorter product lifecycles in which competition tends to be fiercer.

6. Product variety – a greater range of products increases the difficulty of managing product demand.

7. The predictability of product demand – a pattern of demand for products is frequently associated with its characteristics. Products that are difficult to forecast are likely to face greater demand uncertainty.

8. Sharing demand forecast with customers – customers have more knowledge and information about the market, therefore sharing information results in reduced demand uncertainty. 9. The frequency of order expediting – Unexpected events associated with an order that will

change associated business tasks such as order processing and production scheduling. 10. The frequency of change in order content –uncertainty is greater when there are more

changes in order content disrupting the program of related manufacturing and administrative processes.

The main conclusions that can be drawn from the questionnaire are that a decent number of customer purchase products directly from SiTel. But because the majority of sales come from less than 10 customers it is unlikely that the 30-40 customers that purchase directly will increase demand uncertainty. SiTel delivers to a limited amount of customers of which most order through a contract manufacturer or distributor. Once a customer orders a product from SiTel it is unlikely that they will change suppliers due to the high change-over cost.

A possible source of demand uncertainty is the high rate of new product introduction. According to data from SiTel, on average new products are introduced every six months or less. Gaps in new product introduction can last more than six months, which can explain the answers given in the questionnaire. The average lifecycle is reasonably long for an innovative industry but demand data shows that the demand pattern remains difficult to predict even after one or two years. Because the romstore has a limited variety, especially compared to the other production stages, the total product variety has only a limited impact upon demand uncertainty.

The demand pattern is difficult to predict which is reflected by both the high COV and forecast error, indicating that this could be a major cause of demand uncertainty. The forecast is shared with the customer to a medium to high degree and SiTel receives forecast information from several customers itself. However the reliability of these forecasts can differ greatly per period and per customer. Possible causes of the inaccuracies would therefore be worth exploring. Changes affecting planning and scheduling or order content can be easily accommodated if they are received before the final production schedule has been send out. However orders changes received after the production schedule has been send out are more difficult to deal with and are most likely moved to the next month. These factors therefore do not so much affect demand uncertainty itself, but rather SiTel’s ability to deal with the impact of demand uncertainty.

Question Reply

1 30-40 customers purchase products directly (3) 2 There are less then 100 customers in total (1,25) 3 The frequency of customer change is low (2)

4 A new product is introduced every six months to one year (4,25) 5 The average product lifecycle is 2-4 years (1,75)

6 There are 50-100 product variants (2,75) 7 The demand pattern is difficult to predict (3,75)

8 The demand forecast is shared with the customer to a medium to high degree (3) 9 There is a medium rate of unexpected changes affecting planning& scheduling (2,75) 10 There is a medium rate of changes in order content (3,25)

From the responses and available data the high rate of new product introductions, the difficulty of predicting the demand pattern and the lack of accuracy of the shared forecasts appear to be the factors causing demand uncertainty. Besides these factors the frequency of order changes appears to be significant in dealing with the impact of demand uncertainty and as such will also be discussed.

5.4 Influence of the causes of demand uncertainty

Each of the causes of demand uncertainty affects the demand uncertainty in a different way. Here each factor and its influence on demand uncertainty will be explored. This in turn will form the basis for solutions specific to each cause of demand uncertainty. These solutions will be discussed in the next chapter.

Predictability of the demand pattern

According to Holweg et al (2005) and Lee et al (2004) unpredictable or non-transparent demand patterns have been found to cause artificial demand amplification in a range of settings. With artificial demand amplification orders to the supplier tend to have larger variance then sales to the buyer (i.e. demand distortion), and the distortion propagates upstream in an amplified form (i.e. variance amplification.

According to Lee et al. (2004), demand amplification can be caused by three factors; demand signal processing, the rationing game, order batching and price variations.

Demand signal processing

Demand observed at the retailer is transmitted to the supplier in an exaggerated form. As the retailer processes demand signals, the original sales information is distorted and its variance amplifies when passed upstream to the supplier. Additionally, a long lead time in replenishing orders from upstream tends to aggravate the distortion even further.

The rationing game

When a products demand potentially exceeds supply due to limitation in production capacity or uncertainty of production yield the manufacturer would ration the supply of the product to satisfy the retailers’ orders. Assuming the manufacturer allocates output in proportion to the size of the orders, each retailer will try to secure more units by issuing orders which exceed in quantity what the retailer would order if the supply of the product is unlimited, amplifying demand.

Order batching

Depending on the ordering pattern of the buyer, the same expected amount of orders generates different variances. If all buyers place their order at the same time a higher variance exist than when buyers place their orders at different times. A higher variance leads to a greater amplification. Price variations

Variations in the product sales price, from the buyers’ point of view, lead to changes in the purchasing behaviour of buyers. A higher sales price will lead to a minimum of orders, usually just enough to resupply basic inventory needs, while buyers will increase their inventory when products can be bought for lower sales prices.