When supply does not meet demand:
Mismatch size, main causes, improvement strategies and a forecasting solution for
Koga b.v.
Master thesis Business Administration - Operations & Supply Chains
By Gabe Sytema Amstelstraat 89 9725 KV Groningen g.j.sijtema@student.rug.nl
Preface
This paper is my master thesis for the Master Business Administration, specialization Operations and Supply Chains at the University of Groningen. It is a study of the extent, effects and main causes of the mismatch between supply and demand and possible strategies that can improve the match between supply and demand for Koga. This study has been challenging for me in many aspects, however I have learned a lot from it.
I would like to thank dr. Ir. J. Slomp for putting me into contact with Koga. L. Vink, my attendant at Koga, for giving me the opportunity to conduct this research and to patiently answer the many questions I had about the different aspects of the supply chain at Koga. I would also like to thank G. Everts, S. Henstra, S. De Jong, H. Lammertsma, P.J. Rijpstra, and S. Slagman for sharing their knowledge about the different processes at Koga and providing me with the necessary data when I needed it. C. Helfrich for the cups of coffee, providing many useful insights and sharing his knowledge of Koga with me. Finally I would like to thank all the people at Koga for the great atmosphere and constructive attitude which made me really enjoy my time at Koga.
Special thanks go out to my supervisor. dr H. Broekhuis for making sure I kept heading in the right direction with my thesis and for providing me with very useful and sometimes challenging
feedback that enabled me significantly improve this thesis. My second assessor T.D. Bodea for his valuable comments. M.Boekema for making my texts just that bit better. And last but not least, my girlfriend Janneke, for her continuous support and patiently enduring my lack of time for anything other than work and this thesis for the last seven months.
Gabe Sytema,
Abstract
The purpose of this study is to investigate the extent and main causes of the mismatch between supply and demand for Koga and which strategies can improve the match between supply and demand. A preliminary study revealed that total mismatch costs are 2.9 million. 22 percent of these cost arise from lost sales, 78 percent from markdowns. Main causes are the limited volume and mix flexibility due to a six month delivery lead time on bicycle frames and a high product variety. The preliminary study identified two types of strategies to improve the match between supply and demand: postponement- and forecasting strategies. The management team opted to further explore the possibilities of forecasting strategies.
The most promising forecasting strategy is a Delphi panel expert forecasting session followed by allocation of the forecasted sales levels to the available production capacity via the accurate response technique. This strategy is further developed and tailored for Koga and a forecast session for the race segment is executed. This strategy provides Koga with a structured manner of forecasting, a more detailed insight in the estimated demand uncertainties per model and
Table of contents Preface 2 Abstract 3 1 Introduction 5 1.1 Koga B.V. 5 1.2 Problem Statement 5
2. Preliminary study: mismatch between supply and demand 8 2.1 Volatility and uncertainty in demand 9 2.2 Variety and flexibility in supply 9 2.2.1 What creates product variety? 9 2.2.2 How product variety affects the supply chain 10
2.2.3 Types of flexibility 10
2.2.4 How flexibility affects the supply chain 11 2.3 Strategies to match demand and supply 11
2.3.1 Postponement strategies 11
2.3.2 Forecasting strategies 12
2.4 Supply and demand for Koga 13
2.4.1 Research method 13
2.4.2 Results 13
2.4.2.1 Mismatch size at Koga 13 2.4.2.2 Koga and product variety 15 2.4.2.3 Koga and flexibility 15 2.4.3 Diminishing the mismatch between supply and demand 16 2.4.3.1 Forecasting opportunities for Koga 16 2.4.3.2 Postponement opportunities for Koga 17 2.5 Conclusion preliminary research 19
3 Forecasting theory 20
3.1 Introduction 20
3.2 The Delphi panel expert forecasting technique 21 3.3 The accurate response technique 22
4. Forecasting at Koga 23
4.1 Procedure 23
4.2 process 26
4.2.1 Initial forecast 26
4.2.2 Forecast revision 27
4.3 Forecast Race segment 27
4.3.1 Forecast process 27
4.3.2 Forecast results 30
4.3.3 Forecast conclusion and limitations 33
4.3.3.1 Forecast conclusions 33
4.3.3.2 Forecast limitations 33
5 Conclusion and discussion 34
5.1 Conclusion 34
5.2 Managerial implications 36
5.3 Limitations and further research 37
1 Introduction 1.1 Koga B.V.
Koga B.V. is a producer of high-end, innovative bicycles. The company was founded in 1974 and is currently part of the Accell Group. Originally focusing on Race bicycles, production has expanded into six other bicycle segments: Light Touring, Comfort, Mountain Bikes, Fast Movers, Trekking and E-bikes. Furthermore, customers can choose to compose their own bike within the Signature program. With about 90 employees, the company produces 40.000 bikes per year. Koga bicycles distinguish themselves as durable, agile or sportive, innovative and well designed. This implies that Koga bicycles have to meet high standards in areas as durability and safety, but also have to be light and agile.
1.2 Problem statement
The bicycle market is characterized by annual production seasons. A bicycle sales season runs from September till August. The sales season shows two peaks in demand. One is in September, when dealers order new models for their showrooms, and there is a peak in consumer demand between April and July. Most bicycle models are updated every season. An annual update often involves at least a new paint scheme, making old frames or bicycles obsolete. Aggregate demand for Koga is stable and predictable. Demand for an individual model, however, is sensitive to several external factors. These include the popularity of that seasons' paint scheme, the popularity of other models in that specific segment and the weather. These factors make the demand for an individual model considerably less predictable.
Koga intends to match customer demand as closely as possible with their supply of bicycles. In their supply chain, however, Koga is restricted in its response to changes in demand. Since the bicycle frame determines the specific size and model of a bicycle, the exact product mix that Koga will produce is determined when these frames are ordered. The delivery lead time on bicycle frames for Koga is six months. This implies that frames need to be ordered at least six months before bicycle production. Therefore, in the current supply chain setup, Koga has to determine which specific bicycles it will make long before actual demand is known. Changing the volume and mix of bicycles is only possible by ordering extra frames. Therefore there is at least a six month lag between the moment Koga decides to change their product volume and mix and the moment this changed volume and mix enter the market. Hence, this six month delivery lead time on frames prohibits Koga from responding properly to any deviation from expected customer demand during the annual sales peak in spring and early summer. Koga currently tries to offset these problems by creating a forecast before ordering the first frames for a season, and a revised forecast as soon as order data is available. However, Koga considers the performance of this forecast to be inadequate. The main backdraw of the current system is that it requires a first forecast eight months before the start of the sales season1. At that moment no sales information is available and often the definitive component specification of the bicycle is not even known yet. Therefore, the match between supply and demand for Koga is currently suboptimal, resulting in lost sales on high selling models and obsolete stock on low selling models. Koga hence feels that their ability to match their supply with customer demand should be improved.
1 six months delivery lead time, one month production lead time and one month for the new model introductions. Thus
The problem of matching supply with customer demand whilst dealing with long lead times as described above is not unknown in the literature. Randall and Ulrich (2001) for instance state that “long lead times and forecast horizons increase exposure to demand uncertainty that increases
market mediation costs (p. 1590)”. Fisher et al. (1997) model a company’s production flexibility
by dividing production capacity into a speculative part and a reactive part. Speculative capacity is defined as the capacity filled on the basis of forecasts while reactive capacity refers to the capacity filled on the basis of actual sales data. Fisher et al. henceforth describe how one can increase reactive production capacity.
One of the reasons for this mismatch between supply and demand at Koga is the limited ability of their supply chain to respond to changes in customer demand. There is an extensive literature about the relationship between the degree of supply chain flexibility and the ability of a company to let supply meet demand. Most flexibility literature builds on the taxonomy of Slack and Lewis (2002) who divide flexibility into product, mix, volume and delivery date flexibility. Milner (2005), and Zhang (2003) show that volume, mix and delivery date flexibility have a strong relation with market mediation costs and customer satisfaction. Fisher (1997) describes the relationship between flexibility and cost: in order to diminish total costs, the degree of flexibility in the supply chain should match the degree of innovativeness of the product. The underlying reason for this relationship is that innovative products often face high volatility in demand. Higher volatility makes demand less predictable. For that reason, firms producing innovative goods should ensure that their supply chain is flexible enough to face these challenges.
The problem of mismatch between supply and demand that Koga faces bears much resemblance with the apparel industry as described by Fisher et al. (1994). The apparel industry increasingly faces market volatility and increasing purchasing lead times in a similar supply chain structure characterized with subsequent speculative and reactive ordering as modeled by Barnes-Schuster et al. (2002). Fisher and his colleagues (1994) (1996) (1997) (1997) describe the case of Sport Obermeier, that faces a similar problem, and present a technique called accurate response. This technique is a combination of a panel expert forecasting session and consequent intelligent use of the forecasting data to allocate production over available speculative and reactive production capacity. This maximizes the use of the available reactive production capacity.
The existing literature does not apply directly to Koga’s business processes for one important reason. A lot of existing literature on this subject assumes one or more of the following
situations: components are relatively inexpensive and/or are in ample supply, the manufacturer has complete control over the production process of all critical components, or the entire production process is outsourced and the company focuses on selling the product. In other words, this literature is often written from the viewpoint of a complete product and a
manufacturer that has complete control over all critical components. For Koga, bicycle frames are expensive, as a result the cost of an unused frame exceeds the loss incurred on a bicycle sold at a discount. Therefore, if a frame is ordered, it will be used to assemble a bicycle. A high quality frame is an important selling point for Koga and the number of suppliers that can deliver the required quality of frames is limited. For Koga, bicycle frames are hence a detailed control part. This implies that production of frames is outsourced but “The buyer remains responsible for both
functional specifications and detailed engineering” (Asunama (1985) as presented in Hsuang
Koga currently feels that the match between their supply of bicycles and customer demand can be improved. However its knowledge about the extent and the main causes of this mismatch is limited. For instance there are some sources of data on the size of the mismatch but these have never been combined. No data about the nature and main causes of the mismatch is readily available. Given this lack of information it is very difficult to formulate a correct research question. First, more insight needs to be gained on this mismatch: what is the size of the mismatch, what creates the mismatch and which strategies can improve the match between supply and demand. Henceforth, this thesis starts with a preliminary study into the nature and the extent of the mismatch between supply and demand at Koga in chapter two. This preliminary study examines a relatively broad preliminary research question:
How can Koga organize its supply chain and the corresponding bottleneck supply of framesets in such a manner that their supply of bicycles corresponds with customer
demand?
From chapter three onwards this thesis will focus on the strategy chosen by Koga's management to reduce the mismatch between supply and demand. After the results of this preliminary study were presented to them, they opted to further investigate the benefits and possibilities of an improved forecasting strategy. The above preliminary research question was henceforth reformulated into:
What is the optimal forecasting strategy for Koga to adjust their supply of bicycles to customer demand?
2 Preliminary study: mismatch between supply and demand
As was mentioned in chapter one, Koga possessed only limited formal knowledge about the size and the main causes of their mismatch between supply and demand. Henceforth, a preliminary study has been conducted in order to gain more insight into the nature and the extent of the mismatch between customer demand and what Koga supplies. This preliminary study tries to answer the following research question:
How can Koga organize its supply chain and the corresponding bottleneck supply of framesets in such a manner that their supply of bicycles corresponds with customer
demand?
The problem Koga faces is expressed by the mismatch between supply and demand. This
mismatch can be measured by market mediation costs. Koga’s market mediation costs primarily consist of two components: lost sales/stockouts if demand exceeds supply and obsolete
stock/markdowns if supply exceeds demand (Fisher et al. 1994; Randall and Ulrich 2001). One can identify two main possible causes why supply does not meet demand: the predictability of demand, and the ability of a company to match their supply with demand. Predictability of demand is determined by the volatility and uncertainty of demand (Makridakis et al. 1998). As demand becomes more unpredictable or unstable, it becomes more difficult to match supply with demand. The degree to which a company is able to let supply meet demand is
predominantly determined by product variety (Ulrich et al. 1998) and supply chain flexibility (Slack and Lewis 2002). (figure 1)
Figure 1, supply chain flexibility, demand volatility and the mismatch between supply and demand
Product variety is the number of different versions of a product offered by a firm at a single point in time (Randall and Ulrich 2001). More product variety implies more submarkets where demand needs to meet supply, making the match between demand and supply more complex and thus more difficult. Flexibility is the degree in which a company can respond to changes in market conditions and customer demand. From a supply chain perspective this implies that the more flexible a
2.1 Volatility and uncertainty in demand
Every production process requires lead time. This implies that a companies' decision how many products it wants to produce has to be taken before the products are available. If that product has not been made to order, but on stock, this production decision will have to be made on the basis of expected future demand. Predicting future demand is easier and more accurate if demand is relatively stable and if existing demand patterns (for instance season patterns)
continue into the future. If, however, demand is more volatile or future demand is less related to past and current demand then predicting future demand is more difficult. The degree of demand volatility and uncertainty can be measured by the variance of demand (Makridakis et al. 1998). Nahmias (1997) states that the variance of demand for a product group is lower then the variance for an individual product. This implies that aggregate demand can be predicted more accurately than individual demand.
2.2 Variety and flexibility in supply 2.2.1 What creates product variety?
Ulrich et al. (1998) studied the effect of product variety on production strategy for the US
Mountain Bike industry in the mid ‘90’s. They found that the primary driver of variety is related to the way a company exploits its core competences or competitive distinctions as shown in table 1.
Company Source of variation Core Competence Production strategy CODP Cannondale Frame
geometry Frames Everything in house Paint
Specialized Frame material Material (own composites) Everything outsourced (including assembly) Finished product VooDoo Assembly
(components) Assembly Frames outsourced Assembly
National Color
Production lead time and flexibility
Everything in house Paint
Table 1, US bicycle companies and their core competences, based on Ulrich et al. (1998)
Table 1 depicts the source of variation, core competences, production strategies and Customer Order Decoupling point of four US based Mountain Bike producers. All four producers have linked their source of variety to their core competence. The core competence of Cannondale, for
instance, is frame production. Hence, they offer many different types of frames. Since
2.2.2 How product variety affects the supply chain
Randall and Ulrich (2001) state that product variety imposes two types of costs on a supply chain: production costs and market mediation costs. Production costs include incremental fixed
investments needed to produce an extra variant of a product. Examples are the design costs of an extra variant or costs for adapting the production line in order to produce the extra variant. Market mediation costs arise when supply exceeds demand and a product has to be marked down or when supply falls short of demand, resulting in lost sales opportunities and dissatisfied customers (Fisher 1997). Increasing product variety implies more submarkets where supply needs to meet demand, making the match between demand and supply more complex and thus more difficult. As a result, the probability that supply meets demand will decrease for each submarket. Hence, total market mediation costs for a product will increase.
Every source of product variation has its own cost structure; for one source of variation market mediation costs will be dominant whilst for another source of variation production costs will be dominant. Randall and Ulrich (2001) define the dominant type of cost as the costs that have the greatest potential for cost reductions. Therefore, given the source of variation and corresponding cost structure, a company should design its supply chain with a focus on reduction of the
dominant cost. Randal and Ulrich (2001) analyzed the US Mountain Bike industry as well. They found that a bicycle has four different sources of variation: frame material, frame geometry and size, color and component specification (see table 2).
Source of variation Type of variation
Frame material Production dominant
Geometry and size Market mediation dominant
Color Market mediation dominant
Components specification Market mediation dominant
Table 2, source of variation and corresponding type of variation for a bicycle
For most of these sources of variation, market mediation costs are dominant. Therefore most supply chains in the bicycle industry should be designed in such a manner that market mediation costs can be reduced.
2.2.3 Types of flexibility
Slack and Lewis (2002) define four types of flexibility: product flexibility, mix flexibility, volume flexibility and delivery flexibility. Product flexibility is “the time necessary to develop or modify
products (..) and processes that produce them to the point where regular production can start.”
This is thus the time spend between the moment a company starts to develop a new product until the moment regular production can start. Mix flexibility is “the time necessary to adjust the
mix of products (..) being produced”. Volume flexibility is “the time taken to change the
2.2.4 How flexibility affects the supply chain
Fisher et al. (1997) state that the degree of flexibility of a supply chain needs to correspond with the type of product a company produces. Functional products with stable demand, a long product life cycle and therefore low mismatch costs need a supply chain focused on efficiency and cost. Innovative products, with unstable demand, a short product life cycle and high mismatch costs need a flexible2 supply chain focused on reaction speed to changes in demand. Such a supply chain is often combined with a modular product design in order to be able to respond quickly to changes in demand. Mismatch costs arising from problems with aligning supply with demand can be seen as a sign that the supply chain is not flexible enough for a specific product.
2.3 Strategies to match demand and supply 2.3.1 Postponement strategies
Postponement techniques involve strategies for delaying the definitive specification of a product and thus the moment one needs to decide on the final specification of a product (Brun and Zorzini 2009). A company is then able to determine the definitive product mix in a later stage of the supply chain. Therefore the lead time between the moment a product receives its final
specification and the moment it enters the market is reduced. This creates a more reactive supply chain, which enables a company to respond faster to changes in demand.
There are several ways to classify postponement, as is shown in table 3. Van Hoek (1998) classifies postponement by its nature, Olhager (2003) by its influencing factor and Yang et al. (2007) by its supply chain (for more details see appendix 1) These different classifications do not exclude each other. For example, a time postponement strategy can be aimed at reducing delivery lead time, which is a market related influencing factor. Such a strategy can be achieved by making a product market specific after transporting it to that market, which is logistics postponement.
Nature (Van Hoek 1998) Influencing factors (Olhager 2003)
Supply chain phase (Yang et al. 2007)
Time postponement Market related Logistics postponement Place postponement Product related Production postponement Form postponement Production related Purchasing postponement
Product development postponement
Table 3, several classifications of postponement
All types of postponement are closely related to the concept of the Customer Order Decoupling Point (CODP) since the chosen strategy of postponement is linked to a decision or activity that determines the final specification of a product. A product that has received its final specification is always produced for a specific order. This can be an existing customer order or an expected or forecasted customer order; ideally a company tries to postpone final specification until a
customer order is received. This enables a company to link the final specification of a product
2
directly to a customer order and thus, effectively, to produce on customer order. If a company cannot postpone the final specification until customer orders are received, the definitive product will not be produced for a specific customer order but is produced on stock. The amount of products kept on stock, however, is based on expected sales. Therefore, such a product is actually produced for a forecasted customer order.
A concept often used in combination with postponement is modularity (Hsuan 1999; Ernst and Kamrad 2000; Brun and Zorini 2009). One can consider a product as a combination of several components. Some components might be product specific, others are more general. If one tries to keep components more general, these components can be used in a more flexible manner, making the supply chain more flexible as well. In essence, a bicycle is a modular product (Ulrich et al. 1998) for which the frame is the interface. Suppose one considers a frame also as a modular sub-system with the different tubes and other parts as components. If one assumes that some tubes will be more general and others more specific, the real bottleneck is formed by the more specific tubes. Therefore, making the specific tubes more general might imply a significant increase in supply chain flexibility.
2.3.2 Forecasting strategies
Forecasting can be defined as improving the accuracy of predicted future demand with the help of statistical techniques (Makridakis et al. 1998). There are three possible types of techniques: causal techniques, time series and expert opinion. Causal techniques relate demand to specific indicators. The correlation between such indicators and demand is often only satisfactory for aggregated demand (Nahmias 1997). Hence, this technique is less suitable for forecasting
demand for individual bicycle models. Time series forecasting is used to estimate future demand on the basis of previous demand. This is only possible if demand is strongly correlated with past demand (Makridakis et al. 1998). Finally, expert opinion methods convert the knowledge from experts objectively and convert it into statistical data which can be used to estimate future demand and demand uncertainty. Two important expert opinion techniques are the Delphi method (Chambers et al. 1971, Georgoff and Murdick 1986) and the accurate response method (Fisher et al. 1994)
The Delphi technique translates expert estimations of a forecast into statistics, takes their mean and lets respondents consequently motivate their deviation from the mean. This process repeats itself until the respondents reach consensus. The accurate response improves the Delphi
technique by not repeating the process until respondents reach consensus. Instead, the variance between their estimates is used as a measure of demand uncertainty. This measure of demand uncertainty can then be used to determine for each model how much one can produce
speculatively and how much reactively (This type of forecasting will be explained in more detail in paragraphs 3.2 and 3.3)
2.4 Supply and demand for Koga 2.4.1 Research Method
Actual customer demand for Koga is difficult to obtain due to the fact that bicycles are primarily sold through independent dealers. Koga therefore only knows in detail what they sell to the dealers. They do not know which part of the bicycles they sell to their dealers is already sold to customers and which part these dealers still hold on stock. An estimation of dealer supplies can be made by using externally provided retail panel data, but exact data is not available3.
Henceforth for this study sales from Koga to its dealers are used in order to construct a consistent data set. For every model a list of monthly deliveries to dealers in the period January 2007 until January 2012 was obtained. This list was subsequently expanded by adding the segments to every model as they are represented on the Koga website. From the resulting list data on monthly sales per segment could be generated. Monthly storage costs per bicycle were obtained from the accounting department.
Obsolete bicycles are sold with an extra discount. The cost of obsolete stock can therefore be determined by the markdowns on these bicycles. Hence, a list of bicycles that were sold at a higher than 25 percent discount was retrieved for the seasons 2007 to 20114. The resulting list was supplemented with the models segments. Appendix 3 presents key statistics for the markdown cost for the seasons 2007 till 2011. For the season 2011 a list of lost sales due to stockouts was kept by the sales department. They kept the tally of the number of lost sales per model during this season. It was possible to supplement this list with the models segments, the size of the bicycle and the gender type of the bicycle. These statistics are given in appendix 4 Four different types of flexibility were described in paragraph 2.2.3: product, mix, volume and delivery flexibility. In order to gain insight into the degree of flexibility that Koga exhibits the purchasing, planning and sales process were mapped, as well as the connections between them and their connections with production. For this purpose several interviews with key employees in purchasing, sales and planning were conducted. The final result is depicted in an actor activity diagram, a flow chart for the production process and a description of the sales and purchasing process. These are presented in appendix 5.
2.4.2 Results
2.4.2.1 Mismatch size at Koga
Appendix 2 presents graphs for total sales and sales per segment from 2007 till 2012. In order to quickly scan for the share of different models in total sales per segment and the stability of those shares, appendix 2 presents the cumulative sales per model for every segment. One problem that remains with this data is that Koga usually tries to sell what they produce. Hence, there could be a bias in the dataset. However, the extent of the bias is unknown.
3
Except for E-bikes, which need a software update before they can be used. This update is automatically registered via the Accell group EDI. Hence for these bicycles exact sales data can be retrieved.
4
The extent of the problem that Koga faces is determined by the mismatch between supply and demand. This mismatch manifests itself in two ways; via stock outs/lost sales and obsolete stock/markdowns. Key data is given in table 4. To assess the amount of obsolete stock, from the monthly sales data, all sales that occurred after the end of the season were retrieved out of the above described sales data. The most important result found by analyzing this data is that bicycles that are not sold before the end of the season are on average sold within 1.58 month after the seasons’ end (see table 4).
From the sales data described in appendix 2 one can derive that total sales have been relatively stable and predictable, both in total annual sales and in the pattern of sales during the season. Monthly sales for the high volume segments follow the same seasonal pattern every sales season. However in the smaller segments the sales pattern is different every season. Monthly sales for individual models are wholly unpredictable. Moreover only 35 of the 86 2012 models are already in production since 2008. Hence there is only sufficient historical data for these individual models to forecast future sales statistically. These results thus indicate a relationship between previous and future sales in case of total sales and in case of monthly sales in the largest
segments5. Previous monthly sales on a lower aggregate however display such high variance that it cannot be predicted by previous sales. Therefore, statistical predictions of sales will only yield reliable results on an aggregate level.
Table 4, key figures for mismatch costs during the 2011 season
From table 4 one can derive that total mismatch costs amounted to 2.9 million euros in 2011 (measured in lost sales potential). Around 22 percent of these costs are due to lost sales and thus result from under stocking, whilst 78 percent of the total mismatch costs are due to discounts on obsolete stock and thus result from over stocking. This, however, does not necessary implies that Koga is producing too much bicycles. This is apparent from table 1, in appendix 3 which shows that markdowns differ significantly between models and segments. For one model or segment Koga might have planned too much bicycles while they might not have planned enough bicycles for another. From appendix 2 one can derive that monthly sales follow distinctly different seasonal patterns for every segment. This indicates that the demand characteristics per segment differ. These findings stress the need for Koga to take differences in market characteristics between segments into account when trying to diminish the mismatch between supply and demand.
5
Comfort, Light Touring and E-Bikes
Average time extra on stock obsolete bicycles in months
1,58
Storage cost per month € 4,95 Estimated lost sales 2011 € 635.000
2.4.2.2 Koga and product variety
Koga values four core competences: durability, sportivity, innovativeness and design. These competences themselves do not imply an increase in variety but their interaction does imply such an increase. Durability implies a solid frame, sportivity implies an agile frame and design implies an attractive frame. The optimal mix for a durable, agile, well designed frame for every model drives up the variation in frames and frame tubes for Koga. This interaction of competences, which acts as a variation driver, implies that frames are relatively different for every model. It is also what drives Koga to offer a broad range of frame sizes per model. These variation drivers also explain why new product development at Koga is often product driven. Koga often creates new products in order to be one of the first to offer new developments in the bicycle sector. Koga is therefore more product oriented and gives less priority to gathering market intelligence. The primary source of variation for Koga is frame geometry and frame size. From table 2 (page 9) one can derive that the costs for this type of variation are market mediation dominant.
Therefore, according to Randall and Ulrich (2001) Koga should focus on the reduction of market mediation costs incurred through increased product variety.
2.4.2.3 Koga and flexibility
As explained in the introduction, Koga is currently limited in their flexibility to respond to changes in customer demand by a six month lead time on bicycle frames. A frame is an expensive part. If a frame is ordered, it will be used to assemble a bicycle because the cost of an unused frame exceeds the loss incurred on a bicycle sold at a discount. The characteristics of a frame determine the exact specification of a bicycle for a specific model; they determine the specific type6 and size of the bicycle. Therefore, at the moment Koga orders the frames the exact volume and mix of bicycles to be produced is determined. The taxonomy of Slack and Lewis (2002) divides flexibility in four different types: Product, Volume, Mix and Delivery flexibility (see paragraph 2.2.3). The performance of Koga on these four types of flexibility is as follows:
Koga’s lead time for the development and testing of an new frame is up to five months. The lead time from the moment the frames are ordered until the moment they are delivered is six months. Therefore, this delivery lead time is the main source of limitation for product flexibility
There are few limitations on mix flexibility and volume flexibility for Koga until the moment the frames are ordered. From this moment onwards, the exact mix and volume is determined. After this order is placed, the volume and specification mix of a model can only be adapted by
incorporating it in the next frame order. Due to the six month lead time on frames, changes in the product mix and product volume therefore do not enter the market until six months after the product mix and product volume is determined.
Delivery flexibility for Koga is unlimited until frame orders are placed. Between the moment
frame orders are placed and the moment production is planned in the ERP system, delivery date flexibility is limited. When orders are planned and entered into the ERP system production can either be advanced two weeks or delayed six weeks. This limitation is mainly due to supply chain constraints of other components. This limitation is accepted as extra costs need to be incurred for the other components to arrive earlier.
6
Another aspect of flexibility is the relative sizes of speculative and reactive production capacity. Speculative production capacity is the part of production capacity that is employed before any demand indicators are available. Reactive production capacity is the production capacity employed after first demand indicators are available (Fisher et al. 1997). The number of frames ordered in a monthly batch is based on a long term forecast. This forecast is constructed in January7. To produce this forecast the manager operations and logistics constructs a first proposal that is discussed in a management team meeting and changed where deemed
necessary. This changed proposal is the definitive forecast. Based on this forecast, 60 percent of the total expected annual material is reserved at the supplier and monthly orders are placed. In September when the first sales orders are known and the stock of materials at the supplier is depleted, this forecast is updated. From this moment onwards monthly frame orders are determined based on actual sales. Since production capacity at Koga is directly linked to the supply of frames, 60 percent of production capacity is speculative and 40 percent is reactive. However, this reactive power is severely limited due to the six month delivery lead time of the frames, which determines the response time.
2.4.3 Diminishing the mismatch between supply and demand
A solution to improve the match between supply and demand at Koga can be found in two directions. The first solution can be found in the reduction of the impact of the delivery lead time by increasing knowledge about future demand. The second solution lies in the reduction of lead time, in order to increase the ability to react to changes in demand. This hence leads to two different viable strategies to address the problem in hand: forecasting and postponement.
2.4.3.1 Forecasting opportunities for Koga
A suitable forecasting strategy for Koga needs to give forecasts for every model for different parts of the sales season. As was explained in section 2.3.2, causal forecasting methods are often used estimate the effect of specific indicators on the future value of a variable (Makidrakis et al 1998). For instance what is the effect of an increase in advertising on sales. This type of forecasting is unsuitable for Koga for three reasons. Firstly, the correlation between such indicators and demand is often only satisfactory for demand on a segment or total sales level aggregate. (Nahmias 1997). A suitable forecast for Koga would need to be on a model level. This would require more detail then would be possible with causal data. Secondly there is no sufficient data available to estimate relationships for such indicators. Thirdly, many indicators that are
considered important for sales at Koga, like the color of a bicycle or the weather are often very hard to quantify. The data presented in sections 2.4.1 and 2.4.2 (see also appendix 2) show that for Koga forecasts based on monthly historical data will only yield reliable results on the level of total sales or for the high volume product segments. This method, hence, also will not be able to produce reliable forecasts on the level of detail needed. Moreover, Only 35 of the 86 different models offered in the 2012 season have been offered since the 2008 season8. Hence, there is only a sufficient dataset for these 35 models.
7 When the first frames are ordered in January, the first bicycles can be produced in late July and early August,
presented to the dealers in late august and sales can start in September. 8
Expert opinion methods such as the Delphi technique and the accurate response theory might produce a suitable forecast for Koga. These techniques could also provide more insight into the type of information that would be required to make accurate forecasts. Currently, the forecast is determined within a management team meeting. The management team feels this process could be improved. One of the main issues that became apparent from discussions with management team members, is that they often feel that they do not have the necessary data available when they determine the forecast. Moreover, they do not exactly know which data would be helpful to determine the forecast. The analysis of the mismatch between supply and demand and the supply chain process9 revealed that part of the required data is currently not recorded or is recorded on an ad hoc basis10. This deprives the management team of a complete overview. Structuring the forecasting process according to the Delphi technique (see paragraph 2.3.2) would take the forecasting process out of its current black box. It would also give more insight into which parts of the forecast are undisputed within the expert panel and which parts are a cause for debate. Furthermore, an effective expert opinion forecasting process needs to be formally structured in terms of process and required data. Such a more structured forecasting process would provide the management team more insight in the nature of the forecast and possible forecasting risks. In order to prevent any possible group biases in the estimate, the Delphi technique requires that the experts create their individual forecast without consulting each other and the individual forecasts need to remain anonymous for the panel experts. Individual biases should become apparent through the differences between the individual forecasts. By using market experts such as sales representatives as panel members, this expert session can also provide early market information about the new seasons' models. This would give Koga better information and the opportunity to respond to market demand and fully exploit their reactive production capacity.
There however is one main backdraw of this strategy that remains, as explained in paragraph 2.3.2, it does not attempt resolve the underlying problem of limited flexibility but only attempts to improve the match between supply and demand. As was pointed out in the above paragraphs, the mismatch between supply and demand is a symptom of this limited flexibility. This might reduce the effectiveness of this strategy.
2.4.3.2 Postponement opportunities for Koga
From paragraph 2.3.1 one can derive that potential postponement strategies are time, place or form postponement strategies and these can take place in the logistics, production, purchasing or product development phase of the supply chain. The postponement strategy can be influenced by market, product or production factors (for more details see table 3 (page 10) and appendix 1) When one applies these postponement typologies11 to the supply chain process at Koga12, one can distinguish several options for postponement.
9
See paragraph 2.4.1 and 2.4.4 10
For instance the list of lost sales was only kept for 2011 on an ad hoc basis but there is no structural knowledge on the extent of lost sales at Koga.
11
see also paragraph 2.3.1 and appendix 1 12
First, Koga can try to keep materials on supply at the frame supplier, thus postponing the decision on the definitive form13. The frame supplier quotes a three month lead time on purchasing raw materials and a total delivery lead time of six months. This suggests that keeping raw material on stock at the frame supplier has significantly reduces lead time (production related influencing factor14). However, the more raw materials are relatively unique for a single model, the smaller will be the benefit of this type of postponement (product related influencing factor11).
A second postponement option for Koga could be to reduce transport time for those frames that need a shorter lead time. The default method of shipment is by sea. Total transport time by sea is one month. A faster but more expensive option is shipment by air. Koga sometimes uses this option on an ad-hoc basis if shipment by sea implies that the frames will be too late for scheduled production. They could incorporate this type of transport in a more structural manner for bicycles that need a shorter lead time. Take, for instance, a situation when Koga is doubting whether to produce 450 or 500 units of a specific model. They could then order 450 frames, and postpone the decision to order 50 additional frames for one month. If they then decide to order the extra frames, they could ship them by air and still receive them on the same date as the 450 frames ordered before. Therefore, for the additional cost of transporting the frames by air, Koga is able to delay the decision on the definitive production volume and mix for a part of the production. Reducing transport time relates to logistics postponement. The required reaction time and thus the required lead time is determined by market related influencing factors. The delay of the definitive mix decision can be typified as time postponement11.
Third, if the unpainted specification of a frame does not change between seasons, Koga could opt to let a portion of those frames to be delivered unpainted and to paint them in Europe. They can be painted into the current season’s colors when needed in the current season or into the coming season’s colors when not needed for the current season. Delaying the decision on the season specificity of the frames is form postponement. Painting them after transport is production postponement. The extent to which a particular frame differs between two consecutive seasons can be classified as a product related influencing factor. The flexibility necessary for supply to match demand for a specific model is determined by market related influencing factors.
Koga could also opt to combine these three postponement options. If raw materials are kept on stock at the supplier and Koga is willing to use air shipment for frames with relatively uncertain demand, delivery lead time for a frame could, in theory, be reduced to three and a half month15. If one regards a Koga bicycle frame as a modular product of which the different tubes are the different components, one will find that many tubes are used in a range of different frames. Moreover, for a certain specific frame component, a lot of the inter-model difference between tubes stems from variations on a basis tube (see appendix 6, table 1). A problem is presented, however, by the fact that often a frame does incorporate one or two unique tubes, which creates a bottleneck for frame material variety. Consequently, improvements in those bottleneck parts will have a direct and significant effect on frame material flexibility. Improvements can be obtained in several manners. For example, if tubes are variations of each other Koga could continue only the most advanced or highest quality variation and use it in all places where
currently less advanced variants are used. Koga could also reduce tube variety in a similar manner by designing specific advanced, high quality tubes for groups of frames. These tubes have the potential to be developed into a Koga design feature.
13
Form postponement, see appendix 1 for more details 14
See appendix 1 for more details 15
Reducing tube variability would allow Koga to create product groups with similar frame components. Koga could then work with more reliable aggregated forecasts for those product groups. Demand patterns, demand uncertainty and demand volatility differ between the different segments of bicycles. For instance, demand for Mountain Bikes is concentrated in the autumn. Demand for Light Touring bikes is more volatile and sensitive to, amongst others, weather patterns whilst the demand pattern for e bikes is currently dominated by production numbers, since demand for e bikes exceeds supply. According to Fisher et al. (1997) the findings discussed above imply that Koga’s supply chain should not be equal for all models but should be tailored to the demand characteristics of a particular segment. Some segments show much more demand volatility and uncertainty, which also requires more flexibility in the supply chain. If the above mentioned product groups can be created based on such market characteristics, Koga would be able to tailor the supply chain for those product groups to their demand characteristics, allowing them to improve the match between supply and demand.
2.5 Conclusion preliminary research
In this chapter the following research question was examined:
How can Koga organize its supply chain and the corresponding bottleneck supply of framesets in such a manner that their supply of bicycles corresponds with customer demand?
The mismatch between supply and demand is driven by several factors, both on the supply side and on the demand side. On the demand side the volatility and uncertainty of demand are important since they determine how difficult it is to predict demand. On the supply side supply chain flexibility and product variety are essential since they determine how well a company can respond to changes in demand.
The main problem with letting supply meet demand is that there is a lead time of around six months between the moment that the definitive product volume and mix is determined and the moment those products enter the market. To reduce the mismatch between supply and demand, a postponement strategy can be applied. This includes delaying the definitive decision on product volume and mix until more market information becomes available. Another option is to improve the technique which one uses to forecast future demand.
The lead time between determining the product volume and mix for Koga is six months. Product volume and mix are definitive when Koga orders framesets from their suppliers. An analysis of Koga’s sales data yields three important issues. First, demand patterns and demand differ significantly between the various segments. Second, a first analysis reveals that demand most likely will only be statistically predictable at a relatively high aggregation level. Third, bicycles sold after the seasons’ end are kept on stock on average 1.58 months extra. An analysis of the
mismatch between supply and demand revealed that in 2011 about 22 percent of total mismatch costs arose from lost sales whilst 78 percent arose from markdowns on obsolete bicycles. Key data are given in table 4 (page 13).
An initial analysis of sales data reveals that statistical forecasting strategies will most likely only produce reliable results for Koga at a relatively high aggregated level. An expert opinion
forecasting strategy is able to produce forecasts on a more detailed level. This kind of strategy however is very sensitive to the quality of the experts, henceforth the best experts on sales for Koga need to be recruited for this expert panel. The accurate response technique builds upon Delphi expert opinion forecasting by using the detailed results from an expert opinion forecasting session to optimize the use of total production capacity.
These results were presented to the Koga management team. They concluded that there is one important drawback of postponement: it requires cooperation with the frame supplier. They considered this as a serious obstacle. Second, postponement strategies are only achievable on a long term basis. Forecasting strategies could be implemented on a more short term basis. Therefore, they opted to further investigate the possibilities for improvements to their current forecasting system. Henceforth, this will be the subject of study in the following chapters.
3 Forecasting theory 3.1 Introduction
Narrowing the research scope of this thesis to an optimal forecasting strategy for Koga implies that the preliminary research question needs to be reformulated:
What is the optimal forecasting strategy for Koga to adjust their supply of bicycles to customer demand?
A forecasting strategy for Koga requires sales forecasts for every model, for different parts of the sales season. There are three different types of forecasting techniques: causal techniques, time series techniques and expert opinion techniques (see paragraph 2.3.2). Causal techniques provide no suitable solution for Koga (see 2.4.5.1). Time series techniques are only possible under two conditions. Firstly, there needs to be sufficient sales data to estimate a relationship between consecutive levels of sales. Secondly, future sales predicted are based on past sales. Hence, there needs to be a relation between past and future sales. Only 35 of the 86 different models offered in the 2012 season have been offered since the 2008 season16. Hence, there is only a sufficient dataset for these 35 models. Therefore, if a significant relation would exist between past and future sales for a model, it is only possible to produce a time series forecast on both a monthly17 or seasonal level for these 35 models. Moreover, paragraph 2.4.5.1 already explained that the relationship between past and future monthly sales is not strong enough to provide satisfactory forecasting results for individual models.
16
See paragraph 2.4.1 17
Panel expert forecasting techniques are relatively easy to implement. This type of forecasting also allows the use of the accurate response technique since it provides multiple individual forecasts. The accurate response technique uses the difference between these individual forecasts to estimate the degree of demand uncertainty. Based on this estimated demand uncertainty, the accurate response technique is able to efficiently allocate bicycle production over the available production capacity. Combining both techniques implies that one can forecast on a seasonal level and use the accurate response technique to allocate the forecasted season level sales to a lower level. This approach would enable Koga to create a forecast for every model for different parts of the sales season. The Delphi panel expert forecasting technique and the accurate response technique are described in the following two paragraphs.
3.2 The Delphi panel expert forecasting technique
The Delphi technique is a panel expert forecasting methodology. It is an “intuitive methodology
for organizing and sharing expert forecasts (…). It operates on the principle that several heads are better than one in making subjective conjectures about the future, and that experts (…) will make conjectures based upon rational judgment and shared information” (Weaver 1971, p269). It was
named after the Greek oracle of Delphi that was consulted to forecast the future in Greek mythology (Loo 2002). In the late 1940’s, the US military army started to use it and since then it has evolved over the decades into many different variations (Linstone and Turoff 1975). These variations however share four characteristics (Landeta 2006):
1. The process is repetitive: The consulted experts are consulted at least twice on the same subject so they can reconsider their answer based on the answers from the rest of the experts in the previous rounds.
2. A certain degree of peer anonymity: The individual answers from the experts and often even the identity of the experts are only known to the group coordinator but not shared amongst group members.
3. Controlled feedback: The group coordinator controls feedback to the group members in order to eliminate all irrelevant information and to ensure peer anonymity
4. Statistical response: The questions are formulated in such a manner that they can be transformed into statistics. All the given answers are processed into statistics and results are presented in a statistical manner.
One of the main sources of variation between the different Delphi variants is the degree of consensus. In the original Delphi technique the process would be repeated until the experts reach consensus. Later variants do not necessarily search for consensus but use the differences
between the opinions or simply use the Delphi methodology as a method of structuring group communication to resolve complex problems18 (Landeta 2006, Kennis 1995).
18 The Delphi technique incorporates a strictly organized group communication process. This “Delphi”
Through structuring the group process, the Delphi technique has several advantages over face to face groups19. The most important advantage is that the anonymity of the panel experts and the fact these panel experts create their own forecast without consulting others prohibits any group bias. It for instance prohibits the opinion of a single panel member to dominate the result of the process20 or unwanted group think and bandwagon effects. Individual biases will become apparent from the difference between the individual forecasts. These differences might be causes by information only available to the particular expert or might stem from a biased expert. These differences need to be addressed when asking for feedback. Finally, the iteration process forces members to consider their answers carefully, thus invoking more rational results.
3.3 The accurate response technique
The Accurate response technique was first introduced by Fisher et al. in 1994. After its
introduction, this technique has been refined in several research papers (Fisher and Raman 1996; Fisher et al. 1997; Fisher 1997; Fisher et al. 1999; Ulrich et al. 1998). The main improvement of this technique is that it integrates forecasting with the planning and production processes in a supply chain (Li and Ha 2008). This integration is basically achieved by dividing production capacity into a speculative and a reactive part. Consequently, the results of a Delphi style expert opinion forecasting panel are used to allocate production of the forecasted sales levels between speculative and reactive capacity. Production is allocated in such a manner that the available amount of reactive production capacity can be optimally exploited.
Figure 2, speculative and reactive production, source: Fisher et al. (1997)
Total production capacity can be divided in two parts: speculative production capacity and reactive production capacity. Speculative production capacity is the production capacity employed before any demand indicators are available. Reactive production capacity is the production capacity employed after first demand indicators are available (Fisher et al. 1997). Figure 2 gives a schematic timeline of this concept for a single season. At the start of the timeline, one has to decide what to produce in the speculative production phase. Due to lead times for material acquisition, this speculative production phase starts at t0. At time t1 the first demand indicators are available. Based on these demand indicators, forecasted production amounts can be revised and reactive capacity can be filled with the remaining production needs. Due to material acquisition lead times, the reactive production capacity cannot start immediately but at t2. Factor throughput and transport lead times restrict reactive capacity till t3 ; subsequently the season ends at t4.
19
For a comprehensive list of deficiencies in face to face groups that are tackled by the Delphi technique see Kennis (1995) p. 6.
20
The forecast in the accurate response technique is a Delphi-based panel expert forecasting method. The variance between the different forecasts provided by the individual panel members is used as a statistical measure of demand uncertainty. Using standard probability theory, one can calculate the mean and the standard deviation of the panel estimates. When combined wit a specific probability distribution, these statistics can be used to estimate the probability that demand will reach a certain level and thus the probability of over- and underproduction. If one has calculated the mean and variance of every bicycle model and the amount of
speculative and reactive production capacity is known, one can consequently allocate production to speculative and reactive production capacity. The basic principle of this allocation is that the estimated probability of over- and underproduction is identical for every model. This can be established in the following manner. For each model, one needs to subtract an identical fraction of the (model-specific) standard deviation from its average forecasted sales level. The sum of the above results needs to match total speculative production capacity. Or in formal terms, for every model one needs to find:
j j j s =µ −λσ Such that: ) ( 1 j n j j S =
∑
µ −λσ = Where∑
= = n j j s S1 is the exogenously determined total speculative production, s
j is the speculative part of the annual production for model J, µ is the estimated mean for model J, σ is the estimated standard deviation for model J and λ is the weighing factor that is identical for every model (Fisher and Raman 1996).
4 Forecasting at Koga 4.1 Procedure
In order to improve forecasting at Koga, one does not only need to improve the forecasting technique but also the forecasting process. There are three important reasons to restructure the forecasting process. First of all, the current forecasting process is a black box process: during a meeting, the management team deliberates on what they, given all their expertise, expect future sales levels to be. Since this current process is based on the expertise of the management team members, it is often unclear which external sources of data might help to determine future sales levels and henceforth this data is often not readily available. Second, an accurate response technique with Delphi style panel expert forecasting sessions is a relatively complicated process since it involves a number of different process steps by several participants. Third, the Koga management team wishes to automate part of the forecasting process in a later stadium. Due to the problems they currently experience in their forecasting process, by not having the necessary sources of data readily available They feel that being able to digitally consult the input and output data from the forecasting process, would be helpful when they discuss the forecast in a
Given the above described arguments, the proposed forecasting process is constructed and presented in an Actor Activity Diagram (AAD; see figure 4). Following the methods of the Delphi technique and the accurate response technique, this AAD presents an overview of the necessary process steps from the necessary actors and the necessary data input and output. This AAD was constructed as follows. First, a concept AAD was constructed for an accurate response forecasting process. Subsequently, this AAD was reviewed and discussed with the manager operations and logistics, who currently has a leading role in the forecasting process. The results from this review were integrated into the definitive AAD.
Figure 4 presents the AAD for a general accurate response forecasting strategy for Koga. The main goal of this AAD is to provide insight in the proposed forecasting process. Henceforth the exact details needed for the application like the panel experts and the specific data sources are not filled in yet. These details will be filled in when the application of this proposed strategy is discussed in paragraph 4.3 (see figure 5).
Every function or role in the process has its own column, so called swim lanes. Three different roles are distinguished in the AAD. First, there is a panel leader who needs to guide the
forecasting process. This panel leader needs to be independent; in order to guarantee that individual forecasts are anonymous, he or she is cannot be a stakeholder in the forecasting process. Second, there are several panel experts, who provide individual forecasts. Third, there are the sales managers. They are the experts on sales for Koga and as such provide their individual forecast. They are also the primary source of sales
information for Koga. They hence need to provide part of the data input for the forecasting process. Besides the three general roles, the forecasting process needs data input and generates data output. The different forms of relevant data are depicted in the green boxes next to the swim lanes. Finally, black lines depict the process flow, blue lines depict data input and green lines depict data output. The blue boxes depict the final forecast output.
The forecasting process can be divided into two sections: an initial forecast and forecast revisions. Each of these two sections can also be divided into two parts. The first part is the Delphi style panel expert forecasting session. The second part is the allocation of forecasted production to speculative and reactive production following the accurate
response technique21. Figure 3, Legend for AAD
21
4.2 Process
4.2.1 Initial forecast
In December an initial forecast should be made that offers sufficient data for the first orders to be placed. This process has to be finished before the first frame orders need to be placed in January. It is started by the panel leader who asks the panel experts to give their first forecasts and
provides them with the data needed. The experts use this data as a benchmark for their
forecasts, however need to use their own knowledge of Koga sales to assess the sales effects of the various differences between the old seasons' version and the new seasons version of a particular model as well as the effect of possible changes in market conditions between the old season and the new season. It is important that every expert is able to make his or her own assessment of future sales without being affected by other panel members in order to prevent any group bias. Therefore, these experts should not consult each other during the forecasting process and the individual forecasts should remain anonymous among the experts
The panel leader combines the individual forecasts and calculates the means, medians and standard deviations. For the forecasts that deviate significantly from the mean and median, the planner subsequently asks the experts for feedback and if he or she wants to reconsider. There are three important reasons for asking for feedback. Firstly, it gives the experts the possibility to rethink their forecast, thus creating a well considered forecast. Secondly, by confronting experts with their deviation from the mean and median forecast, reasons for that deviation only known to that particular expert or possible individual biases can be filtered out of the forecast. Thirdly, since the reasons for deviating are recorded in the deviation book, it provides Koga with
important market information that previously was only available as panel experts' tacit
knowledge. The reasons for deviating given by the expert and any other feedback are gathered in the deviation book. This book hence functions as an archive for important considerations given by the experts. Forecast revisions are incorporated in the mean, median and standard deviation from the forecasts. Unlike early Delphi variants, the goal is not to reach consensus. Via the standard deviation the difference between the individual forecasts is used as a measure of demand certainty in the following accurate response process. Hence only one feedback loop is needed.
The next part of the forecasting procedure is to allocate production resources to speculative and reactive capacity by using the standard deviation as a measure of demand uncertainty (see paragraph 3.2). Speculative and reactive production capacity is external input. For Koga, 40% of total production capacity is reactive and 60% is speculative (see paragraph 2.4.4). Due to the long lead time on frames and the desire to smooth bicycle production over the season, the relative sizes of speculative and reactive production capacity cannot be changed. After allocating bicycle production to speculative and reactive production capacity with the accurate response
technique, the resulting distribution of production needs to be adjusted to allow for limitations. These limitations stem from production and sales characteristics such as minimum sales
4.2.2 Forecast revision
A forecast revision can be executed periodically or it can be triggered by the arrival of new sales information. However, a useful forecast revision needs some type of new sales information. One can distinguish several moments when important sales information becomes available. When the color scheme for the bicycles is finished in the spring, in August at the introduction of the next season’s models and at the end of September or early October, when the first sales information is available. The sales managers are usually best informed about new sales information. Hence, they are most able to provide this information for the forecast revision process.
The forecast revision process is started by the panel leader who asks the panel experts if they want to deviate from their original forecast. If the experts want to deviate from their original forecast, they are asked to give a reason for this deviation. The rest of the process is identical to the initial forecast: the individual panel expert forecasts are again combined by the panel leader who subsequently asks for feedback and, where possible, reconsideration of the individual forecast if it deviates significantly from the other forecasts. After the panel expert forecasting session bicycle production is again allocated over speculative and reactive production capacity, and assigned to a specific month based on the sales and production characteristics. There is one important difference in comparison to the allocation of production during the initial forecast. One now also needs to take into account that a part of the annual production is already completed and part of the annual production is already in the pipeline. This part of the production is based on the original forecast. Hence, in the revised forecast this original forecast needs to be taken into account in the allocation of production. The revision preferably is finished before the next monthly order of frames.
There are two general concluding remarks to be made. First, the management team has no role in the creation of the forecast. These forecasts provide decision-supporting information for the management team. They can always deliberate the provided forecast and decide upon the definitive production levels. Second, the accuracy of the new forecasting process can be
improved if Koga has implemented it for a couple of seasons. Then, they would, for instance, be able to replace badly performing experts or weigh the forecasts by the past accuracy of the panel experts. They might also be able to estimate a relationship between panel standard deviation of demand and actual standard deviation of demand (see Fisher et al. 1994).
4.3 Forecast Race segment 4.3.1 Forecast process
