1
Thesis MSc. Technology Operations Management
A dual-mode supply chain structure where stock out is not allowed:
a case study
Faculty of Economics and Business
University of Groningen
June 2013
Name: Anna Paques
Details: Oppenheimstraat 80
9714 ET Groningen +31 6 4233 6543
paques.anna@gmail.com
Student number: 1689886
Supervisor University of Groningen: Dr. X. Zhu
Co-assessor University of Groningen: Dr. N.D. van Foreest Supervisors COMPANY X: Bas Wijnbergen
Mark Leenheer
2
Preface
In February 2013, I started this research at COMPANY X in order to complete my master Technology Operations Management at the University of Groningen.
There are several people that I would like to thank. From COMPANY X I first want to thank Bas Wijnbergen for giving me the opportunity to do research at COMPANY X and for the discussions we had during the process. Moreover, I would like to thank Mark Leenheer and Erik Pricker. The regular feedback sessions I had with you gave me so much new input and increased the value of this research. Next, I want to thank the colleagues I had during these 5 months for welcoming me and their moments of jokes and fun! I want to thank all employees at COMPANY X that helped me understand the process under investigation. Their openness and responsiveness were extremely valuable to my research.
From the university I want to thank my first supervisor dr. Zhu for guiding me during this period and that I always could contact him. Besides this, I want to thank dr. van Foreest my second supervisor for his insights into the problem.
I hope my work will be a contribution to COMPANY X. June, 2013
3
Management summary
This research was started since COMPANY X faces high additional costs in their bulk supply chain to paper customers. Failing to meet customer demand in this industry is unthinkable, since the cost of stoppage at a paper mill is enormous. To help COMPANY X decrease the additional costs, we intend to:
‘develop a robust system that fits the situation at COMPANY X, in order to satisfy all customer demand with a minimization of total cost and a reduction of the workload’.
A series of steps are planned in the methodology in order to accomplish the objective of this research. First of all, interviews and observations gave insight into the bulk supply chain process. Second, a literature review was executed that discussed various inventory policies that could be applicable to COMPANY X. Various assumptions concerning: demand, lead time, stock out and a supply chain structure were found for the utilization of these policies. Next, we performed the case study where we discovered that COMPANY X has a dual-mode supply chain structure were stock out is not allowed. A normal and a rush order (preventing stock out) can be placed. Such a model is not addressed in literature. We therefore developed a spreadsheet simulation and modeled this dual-mode supply chain structure.
The literature review and case study revealed that a base stock, (s,S) and MRP policy could be useful to COMPANY X and were therefore utilized in the spreadsheet simulation model. Three different analyses were executed. The objective of the first analysis was to investigate whether the use of one of the policies would have yielded a better performance than the actual procedures used by COMPANY X. The results show that a base stock and (s,S) policy would have performed better. The objective of the second analysis was to investigate the robustness of the results of the simulation used in the first analysis. We made some minor changes and the results show that all three policies perform well when they are implemented in the current situation at COMPANY X. The objective of the third analysis was to increase the understanding of the relationship between input variables and output values. The results demonstrate that a more stable lead time will yield better performance regarding: cost per container and number of rush orders. Furthermore, a more continuous review will result in lower cost while keeping a low percentage of rush orders. The positive impact of a more continuous review is highest for the base stock and (s,S) policy. Finally, as expected when a more fluctuating demand pattern occurs, more inventory is necessary in order to retain a low level of rush orders. This will result in higher cost per container.
4
Table of content
Preface ... 2 Management summary ... 3 List of figures ... 7 List of tables ... 8 1.Introduction ... 10 2. Research design ... 12 2.1 Problem situation ... 12 2.2 Scope ... 13 2.3 Research objective ... 13 3. Literature review ... 143.1 Push and pull policies ... 14
3.1.1 Push policy: MRP ... 14
3.1.2 Hybrid policy: CONWIP ... 14
3.1.3 Pull policies ... 15
3.2 Assumptions for the utilization of inventory policies ... 17
3.3 Information sharing & quality ... 19
3.4 conclusion ... 20
4.Research framework ... 21
4.1 Research questions... 22
5. Methodology ... 23
5.1 Case study selection ... 23
5.2 Case study data collection and analyses ... 23
5.3 Spreadsheet simulation model ... 23
6. Case Study ... 25 6.1 Product flow ... 25 6.2 Information flow ... 26 6.2.1 External transport... 27 6.2.2. Internal transport ... 27 6.3 Costs ... 28 6.4 Customer demand ... 29
6.5 Replenishment lead time ... 30
6.6 Necessary information ... 33
5
6.6.2 Performance measures to evaluate the forecast ... 34
6.7 Linking theory to practice ... 35
7. Spreadsheet simulation model and results ... 37
7.1 Explanation of the simulation ... 37
7.2 Analysis 1 and results: Comparison with history ... 40
7.3 Analysis 2 and results: robustness of the model ... 41
7.4 Analysis 3 and results: Sensitivity analyses ... 42
7.4.1 Results sub analysis a: Lead time is constant 7 days ... 43
7.4.2 Results sub analysis b: periodic review: twice a week ... 43
7.4.3 Results sub analysis c: continuous review ... 44
7.4.4 Results sub analysis d: higher mean demand ... 45
7.4.5 Results sub analysis e: lower mean demand ... 45
7.4.6 Results sub analysis f: higher standard deviation ... 46
7.4.7 Results sub analysis g: lower standard deviation ... 47
7.5 Conclusion ... 47
8. Implementation ... 48
8.1 Selection of the optimal policy for the organization ... 48
8.2 How does the policy work and what information is necessary ... 48
8.2.1 Base stock policy... 48
8.2.2 Necessary information ... 49
8.3 Responsibilities of the various employees ... 49
8.3.1. Line planner ... 49
8.3.2 Shipping agent ... 49
8.3.3 Account manager ... 50
8.4 Frequent evaluation of the base stock policy ... 50
9. Recommendations to COMPANY X ... 51
10. Discussion and Conclusion ... 52
References ... 54
Appendix ... 60
1. Bulk supply chain COMPANY X ... 60
2. Inventory diagrams for the different policies ... 61
3. Literature review ... 62 4. Day after which the graph for the inventory level for various begin stocks becomes the same
6
5. Values of S, (s,S) and safety stock used in analysis 1 ... 65
6. Values of S, (s,S) and safety stock used in analysis 2 ... 66
7. Valuess of S, (s,S) and safety stock used in analysis 3, sub analysis a ... 67
8. Values of S, (s,S) and safety stock used in analysis 3, sub analysis b ... 68
9. Values of S, (s,S) and safety stock used in analysis 3, sub analysis c ... 69
10. Values of S, (s,S) and safety stock used in analysis 3, sub analysis d ... 70
11. Values of S, (s,S) and safety stock used in analysis 3, sub analysis e ... 72
12. Values of S, (s,S) and safety stock in analysis 3, sub analysis f ... 73
7
List of figures
Figure 1: Supply chain COMPANY X... 10
Figure 2 Bulk supply chain of COMPANY X from loading location in the Netherlands to paper customers in Europe ... 12
Figure 3 Percentage of additional costs for all countries that make use of local terminals ... 13
Figure 4 material and information flow for Kanban and CONWIP ... 15
Figure 5 material and information flow for push and pull policies introduced in the literature review ... 21
Figure 6 Three customers ... 25
Figure 7 Specific product route from the Netherlands, via terminal A to a customer in Sweden (route terminal A) and the rush delivery route ... 26
Figure 8 Information flow route terminal A ... 28
Figure 9: Total costs route terminal A for financial year 2012 ... 28
Figure 10 Customer demand for terminal A (financial year 2012)... 29
Figure 11 the lead time of the containers and their departure days ... 31
Figure 12 the cumulative lead time with the agreed replenishment time range of 15 till 19 days ... 31
Figure 14 lead time containers (data after 31-3-2012) ... 32
Figure 13 lead time containers (data before 31-3-2012) ... 32
Figure 15 departure and arrival days of the containers (data from 31-3-2012) ... 33
Figure 16 Forecast SAP and actual demand ... 35
Figure 17 Inventory level for the base stock policy (with S = 27) for various begin stocks. ... 39
8
List of tables
Table 1 Assumptions/requirements for the utilization of different kind of systems and policies & some of their characteristics (see appendix 3) ... 19 Table 2 the three policies (with continuous or periodic review) and their requirements form the research framework ... 22 Table 3 Cost information additional activities... 26 Table 4 information concerning the four materials. ... 30 Table 5 performance measures for the forecast in SAP (demand in containers, 25.500 ton each) (Hopp & Spearman, 2008) ... 35 Table 6 The optimal values of α & β for each material to develop the best forecast based on
9
Table 23 Results Analysis 1 for the base stock policy. Time horizon is 341 days. ... 65
Table 24 Results Analysis 2 for the (s,S) policy. Time horizon is 341 days. ... 65
Table 25 Results Analyis 1 for the MRP policy. Time horizon is 341 days. ... 65
Table 26 Results Analysis 2 for the base stock policy. ... 66
Table 27 Results Analysis 2 for the (s,S) policy ... 66
Table 28 Results Analyis 2 for the MRP policy ... 66
Table 29 Results sub analysis 3a for the base stock policy ... 67
Table 30 Results sub analysis 3a for the (s,S) policy. ... 67
Table 31 Results sub analyis 3a for the MRP policy ... 67
Table 32 Results sub analysis 3 b for the base stock policy ... 68
Table 33 Results sub analysis 3 b for the (s,S) policy ... 68
Table 34 Results sub analyis 3 b for the MRP policy ... 68
Table 35 Results sub analysis 3 c for the base stock policy ... 69
Table 36 Results sub analysis 3 c for the (s,S) policy ... 69
Table 37 Results sub analysis 3 c for the MRP policy ... 69
Table 38 Results sub analysis 3 d for the base stock policy ... 70
Table 39 Results sub analysis3 d for the (s,S) policy ... 71
Table 40 Results sub analyis 3 d for the MRP policy ... 71
Table 41 Results sub analysis 3 e for the base stock policy ... 72
Table 42 Results for sub analysis 3 e for the (s,S) policy ... 72
Table 43 Results sub analyis 3 e for the MRP policy ... 72
Table 44 Results sub analysis 3 f for the base stock policy ... 73
Table 45 Results sub analysis 3 f for the (s,S) policy ... 73
Table 46 Results sub analyis 3 f for the MRP policy ... 73
Table 47 Results sub analysis 3 g for the base stock policy ... 74
Table 48 Results sub analysis 3 g for the (s,S) policy ... 74
10
1.Introduction
The motivation for this research is the high inventory cost in the supply chain (SC) of COMPANY X, see figure 1. COMPANY X produces ingredients for different industries, where we focus on the paper industry. Failing to meet customer demand in this industry is unthinkable, since the cost of stoppage at a paper mill is enormous. Therefore, COMPANY X needs to meet customer demand at all times. One of the most crucial processes to coordinate a supply chain is inventory management (Salcedo et al., 2013; Ryu et al., 2012). The use of optimal inventory policies can minimize the total system wide cost (Routroy, 2010). However, inventory policies described in literature, frequently make use of assumptions like the Poisson customer demand, which is most of the time not in line with reality (Cattani et al., 2011; DeHoratius & Rabinovich, 2011). Therefore, this research evaluates the relevance of different assumptions in traditional inventory policies. Moreover, a replenishment policy is recommended to COMPANY X so that the total supply chain cost is minimized while customer orders are satisfied at all times.
Traditional inventory control made use of a single echelon (stage/level) with deterministic independent demand for non-perishable products without any constraints (Routroy, 2010). Each echelon in such a SC strives to develop local strategies for optimizing the local inventory control, without considering the performance of the complete SC (Ryu et al., 2012). However, these traditional models are often restrictive to reality (Cattani et al., 2011; Routroy, 2010).
Literature research shifted from this traditional single echelon approach to a multi –echelon supply chain inventory planning approach. In such multi-echelon networks, new product supplies are first stored in a regional distribution center. From these local distribution centers the products are distributed to customers. As stated earlier, various assumptions are utilized that are not in line with reality and therefore, this particular area lacks research (Cattani et al., 2011; Routroy, 2010). As early as in 1985, Graves (1985) addressed that the assumption of a deterministic shipment time is restrictive. Furthermore, literature addresses demand as if it were Poisson distributed, which is often a simplifying assumption for real demand (Forsberg, 1997; Nahmias & Smith, 1994) Cattani et al. (2011) mention that traditional inventory research considers a simple supply chain structure as described above, while they introduce a dual-role central warehouse. They believe this alternate supply chain structure is common in practice. Their research shows that the current theoretical models are not applicable in real world settings. They argue that more robust models need to be generated which fit much better with reality. A final assumption that is often made in academic literature, is the allowance of backordering (Axsäter, 1997; Mak et al., 2005). DeHoratius & Rabinovich (2011) report that the literature in operations and supply chain management lacks field research. They argue that field research should be used to investigate the relevance of various assumptions in traditional inventory models.
We contribute to the existing literature by filling the gap of robust inventory policies that fit real life settings. With the help of an extensive literature review, a case study at COMPANY X and a robust spreadsheet simulation model a suitable inventory policy is selected. The objective of the policy is to minimize total cost, while satisfying customer demand at all times.
12
2. Research design
2.1 Problem situation
This research was started, since COMPANY X faces high additional costs in their bulk supply chain. To the best of our knowledge the current literature does not address a detailed solution to this specific problem.
The problem owner has the perception that the delivery of bulk items to their paper customers (named: bulk supply chain, see figure 2) is related with a high amount of additional costs. Moreover, the workload related to this process is assumed to be high and a peak workload occurs when rush orders need to be executed. The organization of a rush order results in extra communication (phone, email, etc.) between various departments of COMPANY X. The problem area is assumed to be all paper customers in Europe that are supplied through local terminals.
Figure 2 Bulk supply chain of COMPANY X from loading location in the Netherlands to paper customers in Europe
In a general sense, the bulk supply chain starts with the transport of bulk items to local terminals (production is omitted). The customers are supplied from these local terminals. Every route from the loading place to the customer has a predetermined cost amount. This amount is set by the shipping-agent and COMPANY X (purchase transport management). However, the total costs are higher than this predetermined amount. These additional costs, which are yearly recurring, can have several causes: (1) additional loading time, (2) additional unloading time, (3) demurrage (container and terminal hire, also mentioned inventory or holding cost), (4) rush deliveries (see figure 2) and (5) weekend deliveries. The percentage of total additional costs for each country is illustrated in figure 3. The total additional costs are approximately 430.000 euro per year. Demurrage costs represent almost 74% of these costs. More information about COMPANY X, the bulk supply chain process and the problem considered can be found in appendix 1.
bulk production at
factory local terminal
delivery from external silo’s
rebulking big bags at external supplier
local terminal
customer
customer
13 Figure 3 Percentage of additional costs for all countries that make use of local terminals
2.2 Scope
Based on the discussed problem, the focus of this research is on the flow of bulk products to paper customers in Europe. The used transport is called: intermodal transport, which consists of the combination of transportation through road, rail and water.
2.3 Research objective
The objective of this research is the following: ‘develop a robust system that fits the situation at COMPANY X, in order to satisfy all customer demand with a minimization of total cost and a reduction of the workload’.
14
3. Literature review
This literature review discusses inventory control policies which could be applicable in the situation at COMPANY X. We first address the push policy MRP, second the hybrid policy CONWIP and third we will address the pull policies: Kanban, (R,Q), (s,S) and base stock. Next the assumptions that are made for using these policies will be summed up. Finally, the usefulness of sharing information and information quality in the inventory policies will be addressed.
Literature addresses two principles for the management of material flows: push and pull policies (Olhager & Östlund, 1990). More recently, a combination of these policies is specified, known as hybrid (Dalalah & Al-Araidah, 2010). Those pull, push and hybrid policies (for example: Kanban, CONWIP and MRP) consist of both a production process and a movement part (Karmarkar & Kekre, 1989; Masuchun et al., 2004). Since in the specific case at COMPANY X we only take into account the movement of material, the focus will be on this specific part. The more statistical reorder point policies (like: (R,Q), (s,S), and base stock) focus on the movement of work (amount and timing).
3.1 Push and pull policies
The essential difference between push and pull policies is the mechanism that triggers the movement of work (Hopp & Spearman, 2008). In a push system the release of work is based on (actual or forecasted) demand, while a pull system or policy authorizes the release of work based on system status (e.g. inventory void).
3.1.1 Push policy: MRP
This system became a prominent approach to manage the flow of raw material on the factory floor in the late 20th century (Mabert, 2007). It describes a broad approach for managing operations; production planning and control (Zipkin, 2000). MRP uses independent (final product) demand to derive production schedules for dependent demand components (Hopp & Spearman, 2008; Nicholas, 1989). The procedure starts with the scheduled due-date for an end item, and then works backward in time to determine the quantities and timing for components that go into the end-item. MRP schedules the release of work based on (actual or forecasted) demand and is therefore said to be push (Masuchun et al., 2004). Thus, MRP derives schedules for the production and movement of raw material, semi-finished products and finished products. In the specific case at COMPANY X only the movement of finished products is taken into account and an actual or forecasted demand will trigger that movement.
Traditionally, lead times are assumed to be constant in MRP, when in fact, actual lead times vary considerable (Karmarkar, 1991). Therefore, planners typically choose longer lead times, which result in larger inventories. Moreover, conventionally MRP used deterministic demand, this is however often too restrictive, since most systems are stochastic (Kwak et al., 2009; Louly & Dolgui, 2013). Several researchers devoted attention to uncertainties in demand and lead time (Louly & Dolgui, 2013; Masuchun et al., 2004). However, taking into account uncertain demand ánd uncertain lead times simultaneously are the least studied (Dolgui & Prodhon, 2007).
3.1.2 Hybrid policy: CONWIP
15 (figure 4), and therefore some researchers refer to such a system where some of the stages are pull and some are push as hybrid (Dalalah & Al-Araidah, 2010; Grosfeld-Nir et al., 2000). The level of work in process (WIP) is maintained constant in the total material flow, since a new material is released into the system immediately once a product exits the system.
3.1.3 Pull policies
We first address, the well-known Kanban, that just like CONWIP uses cards for the authorization of work. Next we address the more statistical (R,Q) policy, the (s,S) policy and the base stock policy. Those statistical pull policies address a specific problem that can be separated into two distinct parts that focus upon: (1) determining the amount of inventory to replenish, and (2) determining the reorder point or level of inventory that will trigger a replenishment order (Mabert, 2007). These statistical policies can be reviewed periodically and continuously. In a continuous review, inventory levels are continuously reviewed, while a periodic reviewed policy is one in which the inventory level is observed only at equally spaced points in time (Haji & Haji, 2007).
Figure 4 material and information flow for Kanban and CONWIP
Kanban
Kanban is seen as a tool for realizing Just-in-Time. The word kanban is Japanese for card, and in the Kanban system, cards are used to manage the delivery and/or production of parts, items, or raw material (Lage Junior & Filho, 2010). The cards connected to production are called P-kanban and a conveyance kanban, or C-kanban is an authorization to move a container from an upstream buffer to a downstream buffer (Nicholas, 1989; Schonberger, 1983). The dual card configuration, where both the production and transportation are controlled by P- and C-kanbans is the most familiar.
This pull system is triggered by the demand at the most downstream location. Production or delivery at each inter-mediate stage is initiated on a signal or authorization coming from a downstream stage. So, the difference between CONWIP and Kanban is that authorization cards in CONWIP move only
16 from the last stage in the process back to the first stage, whereas in Kanban these cards move between each and every subsequent stage (figure 4). In Kanban, material flows only when authorized from a downstream stage and material is pulled from operation to operation. Moreover, compared to CONWIP, the WIP in Kanban is constrained at every operation.
To achieve Kanban, several questions need to be answered. For example, when and how should authorization signals be sent upstream and what size should the buffers be? This last question is similar to the problem of determining the reorder point in the more statistical policies. A simple reorder point system is used to answer these questions (Nicholas, 1989).
Traditionally, the lead time in pure Kanban is zero while the lead time in CONWIP is expected to be small. Lead time is the total time required to replenish a buffer, so the elapsed time between an order and the replenishment. In Kanban, it is assumed that a part is held in the outbound stock point when requested, while in CONWIP the levels of WIP are kept low and therefore lead times are somewhat longer. So, the utilization of the traditional Kanban system can result in improvements in reduction of inventory, waste labor and customer service (Wang & Sarker, 2006). However, the traditional Kanban system is not adequate in any situation. In a Kanban system, some material must be held in the buffers between the operations, and to justify these buffers, a somewhat continuous, stable demand is required (Takahashi & Hirotani, 2005; Lage Junior & Filho, 2010; Tardif & Maaseidvaag, 2001)). However, researchers acknowledge the more realistic stochastic systems and try to find Kanban systems that fit better to these real life settings (Gupta & Al-Turki, 1997; Tardif & Maaseidvaag, 2001).
Finally, if the entire system consists of only two stages, Kanban and CONWIP are the same (Nicholas, 1989), as is the case at COMPANY X. Thus, we address these two policies as identical in the remainder of this research. Moreover, since we disregard production, only the movement of work (C-kanban) with the help of cards in such systems has our interest. The objective of Kanban/CONWIP is now equal to the more statistical reorder point policies.
(R,Q)
In an (R,Q) policy, when the inventory position (on-hand inventory – backorder + orders) declines to or gets below the reorder point R, a size Q is ordered (Axsäter, 2003). In the basic (R,Q) policy the inventory level is continuously reviewed and demand is assumed to be random and follows a normal distribution (Hopp & Spearman, 2008). The (R,Q) policy if frequently used in a simple two level inventory system with one warehouse and various retailers (Axsäter, 2003; Forsberg, 1996) or a somewhat more difficult three-echelon inventory system (Hajiaghaei-Keshteli et al., 2011). Ordering and inventory holding costs are incorporated and stock out is allowed. Holding costs are defined as costs for capital that is tied up in inventory, while shortage costs occur if an item is demanded and cannot be delivered due to a stock out (e.g. extra administration). In the (R,Q) policy a fixed cost (the ordering cost) is associated with a replenishment order. Normally, the replenishment lead times are assumed to be fixed and known and literature addresses the lead time as if it is constant (Axsäter, 1997; Seo et al., 2002).
(s,S)
17 the inventory level is reviewed continuously (Mak et al., 2005). When the (s,S) policy is applied with a continuously reviewed inventory level and demand arrives in time buckets of one unit, the (s,S) policy is similar to an (R,Q) policy under continuous review (appendix 2), with R = s and Q = S – s (Haji & Haji, 2007; Presman & Sethi, 2006).
In the traditional policy the replenishment lead times are assumed to be constant and demand patterns are deterministic (Duan & Warren Liao, 2013; Mak et al., 2005). However, since actual demand and lead time patterns are often random, the current literature addresses stochastic distributions (Mak et al., 2005).
Base stock level
The base stock, (S-1,S) or one-for-one policy is used when the inventory level is periodically or continuously reviewed. This policy only answers the question ‘how much stock to carry?’ (Hopp & Spearman, 2008). When a base stock policy is used, the stage in the supply chain (for example the warehouse) determines a target inventory level that is called base-stock level (S). During each review, the inventory position is determined and the warehouse places a replenishment order (of one unit at the time) every time the inventory position reaches r (= S – 1) (Hopp & Spearman, 2008). The continuous reviewed base stock level is similar to an (R,Q) policy with R = S – 1 and Q = 1, see appendix 2 (Axsäter & Rosling, 1994). Therefore, when the ordering cost is negligible in the (R,Q) policy, a base stock policy is used (Haji & Haji, 2007). Moreover, a continuous base stock policy is similar to a Kanban policy where the number of kanbans equals S. Thus, the number of cards in a Kanban system is the same as the base stock level (S) and the same as the (R,Q) policy with R = S -1 and Q = 1. However, the Kanban policy and the base stock policy differ in that a base stock policy has no limitation on the amount of work that can be in process, while in Kanban the amount of cards limits the occupancy of the stages in the supply chain (Zipkin, 2000; Hopp and Spearman, 2008).
3.2 Assumptions for the utilization of inventory policies
The assumptions made for the use of the traditional policies described previously are restrictive. First of all, traditionally lead times are assumed to be deterministic; known and often constant lead times are used (Duan & Warren Liao, 2013; Karmarkar, 1991). But research shifted towards more realistic lead times; stochastic distributions.
Secondly, inventory policies work best with a deterministic demand; system parameters are known or can be estimated with certainty. However, in practice this is often not the case (Cattani et al., 2011; Kwak et al., 2009). Demand is then said to be be stochastic. When demand is assumed to be stochastic in the articles investigated, a Poisson distribution is applied in most instances. But the use of this demand distribution is also debatable (Saidane et al., 2013; Forsberg, 1997), since this distribution is most appropriate for relatively low volumes of demand (Cattani et al., 2011; Nahmias & Smith, 1994).
Third, most of the investigated articles allow for backordering when stock out occurs. When demand is backorderd a shortage, penalty or backorder cost incurs (Axsäter, 1990; Seo et al., 2002). Delayed orders are for example delivered on a first come-first served basis or when customers are unwilling to wait backorders will result in lost sales.
18 three-echelon inventory system is for example addressed by Hajiaghaei-Keshteli et al. (2011). In these systems the material is transported by one route. However, these simple supply chain structures are restrictive in real life settings (Cattani et al., 2011). Organizations do have more complicated supply chains, where they can choose between two alternatives for refilling stock at some stage (Arts & Kiesmüller, 2013). Kiesmüller et al. (2005) consider a dual sourcing supply chain structure. The supplier can choose between two different tranportation modes, a fast mode, for example transportation by road, and a slow mode, for example transportation by rail or water. They apply a base stock policy and inventory is periodically reviewed. The main objective of their research is to reduce total costs. However, they do allow for stock out to occur, since unsatisfied demand is backordered. Likewise, Arts & Kiesmüller (2013) and Chiang & Gutierrez (1996) both address an inventory system with two modes of resupply; a regular order and an emergency order. Chiang (2003) studies a periodic inventory system with long review periods. A regular and emergency order can be placed each period. If the inventory position is negative at a review moment, an emergency order should be placed, that raises the inventory position to a non-negative level. However, orderings are placed periodically, therefore stock out can occur in the mean time and unsatisfied demand will be backlogged. Literature also focussed on placing regular order periodically, while emergency orders are placed in case of imminent stock outs (Teunter & Vlachos, 2001). These orders follow an order-up-to policy on the inventory positon. Moreover, unfilled demand can occur, which is backordered and filled whenever a regular order or an emergency order arrives. Summarizing, it may be stated that literature addresses more realistic supply chain structures, though stock out is likely to occur in these dual-mode supply chains.
19
Policy Usage requirements:
Demand Replenishment lead time
Characteristics:
Review Stock out Triggering rule
MRP deterministic &
constant stochastic
stochastic yes order due date –
estimated lead time Kanban/CONWIP deterministic stochastic deterministic stochastic stock withdrawal = order arrival (R,Q) constant & continuous stochastic (mostly, Poisson)
constant continuous yes/no stock withdrawal
and inventory level rule = order arrival
(s,S) deterministic stochastic (mostly, Poisson) stochastic constant zero continuous /periodic
yes stock withdrawal
and inventory level rule = order arrival
Base stock level stochastic (mostly,
Poisson) deterministic stochastic constant continuous /periodic
yes stock withdrawal =
order arrival Table 1 Assumptions/requirements for the utilization of different kind of systems and policies & some of their characteristics (see appendix 3)
3.3 Information sharing & quality
Now that we discussed pull and push policies and the restrictions for utilizing them, it is time to address the concept of embedding the policies into the company. Kanban and CONWIP policies for example, use cards to manage the flow of material. More recently, radio frequency identification (RFID) technologies are used to improve the product traceability and the visibility among supply chains (Sarac et al., 2010). In this sub section we focus on the usefulness of sharing information and more importantly: the quality of the information which is necessary when implementing policies into an organization.
The accuracy of data and the information flow within a system are of crucial importance to the performance of a system (Karmarkar, 1991; Masuchun et al., 2004). Information sharing among supply chain members reduces inventory costs in the supply chain (Cachon & Fisher, 2000; Gavirneni, 2002; Sarker et al., 2013; Seo et al., 2002).
Supply chain systems are filled with uncertainties (Datta & Christopher, 2011). Demand and supply uncertainties are mentioned and addressed in literature (Baghalian et al., 2012; Chen & Paulraj, 2004). Uncertain lead times or transportation times (Schmitt, 2011; Simchi-Levi & Zhao, 2005; Van Landeghem & Vanmaele, 2002), price fluctuations and information delays (Van Landeghem & Vanmaele, 2002) are other sources of uncertainty. Information sharing within the supply chain can reduce this (Datta & Christopher, 2011; Ouyang, 2007).
20 should also be of high quality. Forslund and Jonsson (2007) used four variables to measure information quality;
- in time - accurate
- convenient to access - and reliable
Information being in time means that the information arrives in the agreed time. Accuracy concerns the degree of obvious mistakes in the data. Convenient to access deals with the ease of accessing the data without further processing. And finally, reliability refers to the probability that the information remains unchanged.
They reveal that supply chain performance is higher for suppliers with access to customer forecasts compared to suppliers without access to forecasts. However, they cannot verify if the supply chain performance is positively correlated with forecast information quality. This might be due to the fact that their sample consisted of suppliers that where rated as the most important to the company investigated, and that in this relationship a deeper cooperation can be expected. Hartono et al. (2010) measured information quality in terms of its usefulness, the accuracy and the accessibility. Compared to Forslund and Jonsson they found a positive relationship between the quality of shared information and the operational supply chain performance as well as the overall firm performance.
3.4 conclusion
21
4.Research framework
In the literature review we addressed several policies. This chapter will address the framework in which the research will continue. The framework consists of various policies to select from for the specific case of COMPANY X, see table 2. In addition, this chapter will concentrate on the research questions.
As mentioned before, for the situation at COMPANY X Kanban and CONWIP are similar. Furthermore, ordering costs are negligible at COMPANY X. Thus, a continuous (R,Q) policy is the same as a continuous base stock policy. The literature review also indicated that a continuous base stock policy is similar to Kanban. We therefore only consider the MRP, base stock and (s,S) policy (with periodic or continuous review) that could be applicable to COMPANY X (table 2). Figure 5 shows the material and information flow of these policies.
Demand distribution, replenishment lead time and information sharing and quality seem to be important aspects for using these policies and for managing inventories succesfully (Simchi-Levi, 2009). Table 2 shows the three different policies and their usage requirements concerning these three aspects.
Figure 5 material and information flow for push and pull policies introduced in the literature review loading location terminal: stock loading location terminal: stock customer customer material flow Information flow
Kanban, CONWIP, (R,Q), (s,S), base stock MRP
authorization of work: forecast or actual
demand
22
Policy Demand Replenishment
lead time Necessary information MRP deterministic and constant stochastic deterministic stochastic reliable demand forecast Kanban/CONWIP/continuous base stock/ continuous (R,Q)
(In the remainder of this research: continuous base stock policy)
deterministic stochastic (mostly, Poisson) deterministic constant stochastic reliable inventory level
continuous (s,S) stochastic (Poisson) constant
reliable inventory level periodic (s,S) deterministic stochastic constant zero stochastic reliable inventory level
periodic base stock stochastic (Poisson) stochastic
constant
reliable inventory level
Table 2 the three policies (with continuous or periodic review) and their requirements form the research framework
4.1 Research questions
Although this research is initiated based on a problem at a real company, a case study (explained in detail later) is also beneficial in the development of exploratory research (Meredith & Samson, 2002). Cattani et al. (2011) and DeHoratius & Rabinovich (2011) mention that firm specific data could be used to evaluate the relevance of different assumptions in traditional inventory models. Thus, this research evaluates various assumptions that are made in literature for the utilization of the earlier mentioned MRP, (s,S) and base stock inventory policies in a real life setting. Therefore, the research questions are as follows:
1. What inventory policy (from the three policies in table 2) fits best to the case study (concerning the three research aspects: demand, lead time, information sharing & quality)? 2. What adjustments need to be made to the chosen policy in order to fit the case study at
COMPANY X?
3. What adjustments, concerning the three research aspects, should COMPANY X make in order to use the chosen policy?
23
5. Methodology
This research presents an explorative case study conducted with the aim of evaluating the relevance of assumptions made in traditional inventory models. In addition, an inventory policy must be selected and when necessary adjusted to fit the case study in order to reduce the costs in that specific supply chain. A spreadsheet simulation model is derived in order to find a policy that minimizes total cost.
The case study was carried out following Eisenhardt's (1989) process for case study research. We used a case study to explore the system in its natural setting, learning from the case and through observing actual practice generating theory from practice (Benbasat et al., 1987). Together with the information from the literature review, the results of the case study will be used to answer research questions one and two. Moreover, research questions three and four will be answered with the help of the findings of the case study and the results from the spreadsheet simulation model. In line with Eisenhardt’s (1989) recommendations, the research was set up without any particular theory or hypotheses in mind in order to retain theoretical flexibility.
5.1 Case study selection
The case study used in this research is the flow of bulk items to customers in Sweden. These customers are supplied through the terminal A. This case study is selected because the additional costs are highest in Sweden. Furthermore, the terminal supplies three customers, of which one customer is the largest consumer in this total bulk supply chain through Europe.
5.2 Case study data collection and analyses
Multiple data collection methods were used in order to conduct the case study. Tellis (1997) indicates the importance of multiple sources of data to enhance the reliability of the research. Semi-structured interviews with employees from the departments: planning, order management, front office, transport management and an interview with the account manager were executed. These open-ended interviews gave understanding of the bulk supply chain to paper customers in Sweden. Moreover, structured interviews or short phone calls where concrete questions were asked later on in the data collection stage, gave more specific information concerning this process. Observation at the office gave insight into the method of working. Archival sources or documents such as annual reports, power point presentations, information in the SAP system (the enterprise resource planning system), data concerning historical demand and stock level obtained through the shipping-agent were gathered and analyzed. The findings were frequently discussed with two members of a project team within COMPANY X to share ideas and give comments for further investigation.
5.3 Spreadsheet simulation model
After the case study data was collected and analyzed, we learned that no policy would fit the case at COMPANY X directly, since the organization makes use of another supply chain structure compared to what is frequently mentioned in the literature. We build a spreadsheet simulation model in excel to simulate this supply chain structure and implemented the MRP, base stock and (s,S) policy to answer research question four. Three different kinds of analyses were executed:
1. Comparison with history
24 demand and the actual lead times that occurred in financial year (FY) 2012. We compared the base stock policy, the (s,S) policy and the MRP policy in order to find out whether one of these policies would have performed better. The input variables and output values are more explicitly explained in the chapter ‘spreadsheet simulation model’.
2. Sensitivity analyses – to test the robustness of the results of the model in the presence of uncertainty
During this analysis we compared the three policies again. Moreover, demand and lead time were random variables with a known probabilistic function. We used Palisade @Risk, which was used by Abuizam (2011), to iterate 1000 times in order to test the robustness of the results.
3. Sensitivity analyses – to gain increased understanding of the relationship between input variables and output values
In this third analysis we changed the lead time, the review moment or the demand distribution. We tried to demonstrate how changing the input variables may affect the output values. This also validates the correctness of the proposed model, because the output values react in a reasonable way to the changes in input variables.
25
6. Case Study
The case study in this research focuses on the supply of customers in Sweden via terminal A. In the current situation bulk products are loaded in the Netherlands and shipped to this terminal, from where three customers are supplied (see figure 6) with four different products. This process is called the ‘route terminal A’, where production is omitted. First, we propose the product and information flow of this route. Second, we illustrate the total cost in financial year (FY) 2012 and explain the additional costs. Then, we address the three research aspects; demand, replenishment lead time and information sharing and quality. Finally, we identify which theoretical inventory policies could fit the case study at COMPANY X in the subsection ‘linking theory to practice’. This chapter deals with the data collection and data analyses.
Customer A - XXX Customer B - XXX Customer C - XXX
Figure 6 Three customers
6.1 Product flow
A container is loaded in the Netherlands and then transported by road to XXX. From there, the bulk containers are placed on a boat and shipped to Rotterdam. The boat from XXX to Rotterdam leaves once a week on Thursday. In Rotterdam the bulk containers are placed on a different boat (leaves once a week on Saturday or Sunday) and are shipped to terminal A. Most of the time, the containers arrive on Thursday at the terminal, where they are stored. The shipping-agent (shipping-agent B) transports the desired amount of bulk by road to the paper customer. As mentioned before, failing to meet demand from these paper customers is unthinkable, since the cost of stoppage at a paper mill is enormous. Therefore, COMPANY X needs to meet customer demand at all times.
26
Loading (NL) XXX Rotterdam (NL) Terminal A Customer (SE)
internal transport external transport
rush delivery
Figure 7 Specific product route from the Netherlands, via terminal A to a customer in Sweden (route terminal A) and the rush delivery route
Compared to external transportation, the internal transportation is executed by shipping-agent A. Furthermore, shipping-agent A organizes the complete transportation from loading in the Netherlands to unloading at the customer side. Together with the department procurement transport management of COMPANY X, shipping-agent A determines a fixed freight rate of € XXX for the internal part of the route. The terminal is not owned by COMPANY X or the shipping-agent. Shipping-agent A has contact with the owner of this terminal, and the handling costs at this terminal are incorporated into the freight rate.
The invoice from the shipping-agent A can differ from the precalculation of the freight rate. These additional costs can have several causes as explained in appendix 1. The additional costs for the route terminal A are displayed in table 3 below. Loading in the Netherlands and unloading at the customer side can both take two hours. When loading and unloading takes longer, an additional cost occurs. A rush delivery occurs when no stock is available at the terminal, see figure 7. The cost as a result of the rush delivery, depends on the urgency of the delivery and it therefore depends on the transport-vehicle and route taken at that moment. A truck with big bags is send to Sweden, which needs to be re-bulked. We assume that the cost for this route is € XXX from loading in the Netherlands to unloading at the customer side. However, we can now save the predetermined cost (of € XXX) for the external and internal part of the route terminal A. So, the additional cost for a rush delivery is € XXX per container. The lead time for this route is assumed five days, however, we will assume in our analyses that the lead time is one day, the reason for this will be explained in the chapter “Spreadsheet simulation model”.
Additional cost Costs
Loading XXX € / hour Unloading XXX € / hour Demurrage (terminal + container rent) 2 – 14 days : XXX € / day 15 – 30 days : XXX € / day 31 – 60 days : XXX € / day 60+ days : XXX € / day
Rush delivery XXX € per container
Weekend delivery Saturday : XXX €
Sunday : XXX € Table 3 Cost information additional activities
6.2 Information flow
27 6.2.1 External transport
Specific for this casus is, that a second shipping-agent (shipping-agent B) is incorporated in the process. This is due to the fact that the customer only operates with this specific shipping-agent. Shipping-agent B makes a phone call to the customer side almost every day (also in weekends) to obtain the current silo level. They decide if and when the customer needs to be supplied. The delivery is executed by shipping-agent B, after which shipping-agent A reports the purchased and transported amount of bulk and the delivery date to the sales office (DE). The sales office enters a sales order in the SAP system the same day. Subsequently, during the night the delivery and shipment are automatically derived in SAP. Shipping-agent A needs to write off this shipment in SAP, so that it is concrete that the delivery has taken place. These documents are prepared for the external transport (from the terminal to the customer), and are therefore prepared after the actual transportation has occurred. We refer to figure 8 for graphical representation of this process.
6.2.2. Internal transport
The internal transport (from the loading location to the terminal) is organized by the line planners in the Netherlands, see figure 8. The line planner replenishes the terminal on a weekly basis. Shipping-agent A enters information about the transportation of bulk items from the terminal to the customer into SAP. However, the specific information in SAP is delayed, as mentioned by the line planners. Further, the line planners receive an excel file from the shipping-agent about the amount of stock at the terminal and the expected amount of bulk arriving at a certain date. The account manager is responsible for the customer, and has regular contact with him. When the account manager receives meaningful information to the planning department (e.g. the customer decides to stop production for a while) he forwards this information.
The line planners make use of the excel file, experience and historical demand to find a trend so they can determine the optimal re-order amount. Next they prepare a stock-transport-order, which contains a delivery order and a shipment. The shipment contains information about the amount of bulk that needs to be transported to the terminal, where and when this bulk needs to be loaded and the delivery date. This shipping order goes to shipping-agent A. The delivery order goes to the production location. This document contains information about the amount of bulk that needs to be available for loading at a specific date. The stock-transport-order is specifically meant for the internal transport between loading in the Netherlands and unloading at the terminal.
28 Customer Shipping agent B Accountmanager Shipping-agent A Line planning
Sales office Order management
Shipping-agent B
Production location silo level
various info.
delivery info delivery info delivery info
shipping order
delivery order specific info
excel file delivery info
information internal transport
information external transport
XXX XXX 71 2 10 5 1 10 1 XXX External transport Internal transport Terminal + container rent Loading
Sunday loading Saturday loading Unloading
Rush delivery: direct from NL Rush delivery: Rerouting
Additional cost (%)
Figure 8 Information flow route terminal A
6.3 Costs
The total cost of financial year 2012 for the route terminal A is illustrated in figure 9. Demurrage cost (terminal and container rent) is 71% of the total additional cost for the terminal, which is almost € XXX for financial year 2012.
29
6.4 Customer demand
In this sub section we provide insight into the customer demand flow. Data was gathered from the shipping-agent and from the SAP system and is analyzed in this section. We assume that the delivery date at the customer side is the true moment of demand. We assume this, because paper customers request direct delivery, with minimum lead time. The customer demand (per product) for the route terminal A is illustrated in figure 10. The figure shows that the minimum demand in a time window of one year is one container and the maximum amount supplied is 18 containers, with a total demand of 511 containers.
The material E is disregarded, since this material is only purchased two times. In this time period, the material XXX was transported from a different route. Because of this deviant route and the fact that XXX a is not taken into account in the additional costs, we do not take this material into consideration in this research.
Material A is the material that is demanded the most. The demand goes up and down during the year, but is always in the range of zero till eleven containers at most. However, at the turn of the year the demand curve is zero for four weeks. A reason might be a production stop during the holiday season. Demand for material C is relatively stable, but at a much lower demand rate than material A, namely between zero and three containers. Materials B and D are only purchased during several weeks in the time range of one year and are demanded in a low amount. Since the demand per product is diverse, we use the demand per product during the remainder of this research and not the total demand. Information regarding the demand distribution, which is used for later analyses, concerning the four materials can be found in table 4.
30
Material A B C D
start / end date of demanded period * 21-7-2011 / 15-8-2011** 22-9-2011 / 3-2-2012 *** 4-8-2011 / 22-8-2012 5-4-2012 / 23-8-2012 demand distribution Normal distribution (0.021 > 0.01). With µ= 6, & σ = 2.687 Poisson distribution (X2 < 11.431, with a 0,01 significance level). With λ = 1.579 Poisson distribution (X2 < 5.991, with a 0.05 significance level) With λ = 1.034 Poisson distribution (X2 < 11.431, with a 0,01 significance level). With λ = 2.286
Table 4 information concerning the four materials.
* the additional costs mentioned earlier are also based on this time period. ** In implementing the policies in the next chapter, we do not take into account the zero demand at the turn of the year. We assume that the demand is mediated during the year. *** Demand for this material occurred also far before this start date and far after the end date. We disregard this small amount of demand, in order to have a constant flow of demand.
6.5 Replenishment lead time
This section addresses how data concerning the lead time was gathered and illustrates the analyses of the data. Two approximately normal distributions appear in the lead time data (figure 11). Therefore, we derive two hypotheses that are examined in order to find out why the data appears like these normal distributions.
The replenishment lead time in figures 11 and 12 are calculated by subtracting the arrival time at the terminal from the departure time at the loading location in the Netherlands. Two arrival times could have been used: (1) the arrival time mentioned in SAP, or (2) the arrival time mentioned by the shipping-agent A (an excel file). These arrival times should be similar, however, the data showed minor differences. We assume that there is a delay in the SAP system and therefore we used the arrival time in the terminal mentioned by shipping-agent A to calculate the replenishment lead time. Furthermore, the data consisted of the same data as was used for the customer demand, so the time window is also one year (July 2011- August 2012). However, of the 511 deliveries, only 438 deliveries contained complete arrival time data. Thus, these figures are based on 438 deliveries.
31 0 20 40 60 80 100 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 # o f co n tain e rs
Lead time (days)
Lead time
Departure day: Monday Departure day: Tuesday Departure day: Wednesday Departure day: Thursday Departure day: Friday Departure day: Saturday Departure day: Sunday
0 20 40 60 80 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 cu m u lativ e % o f to tal # co n taine rs
Lead time (days) Cumulative lead time
49 97
From figure 11, two hypotheses are derived to explain the appearance of two approximately normal distributions in the lead time.
1. The material departures on various days. The boat is taken directly or products need to wait for the next boat to arrive, causing diverse lead times.
2. There is a difference in lead time between the route before 31-3-2012 and afterwards. Hypothesis 1 cannot be confirmed. Figure 11 shows no explicit explanation that the diverse lead times are caused by different departure days. To examine hypothesis two, we separate the lead time into two graphs, one for lead times before 31-3-2012 and one for after 31-3-2012 (figures 13 and 14). Figure 13 looks like a normal distribution with outliers in lead times for 10 and 11 days. These lead times (10 and 11) would fit perfectly in figure 14. It might be that the two different routes were used simultaneously for a short moment of time.
Figure 11 the lead time of the containers and their departure days
32 We focus our attention on the lead time from 31-3-2012, since this route is still taken. This route never took longer than 18 days. Almost 20% of all containers had a lead time of 15 till 19 days, while the other 80% were in the range of 7 till 15 days. Figure 15 shows the various departure and arrival days of the containers based on figure 14. As was mentioned earlier, the boat from XXX is assumed to leave every Thursday. The data concerning this lead time strengthen this assumption. Normally, containers loaded during the week will be shipped on this boat. Containers loaded on Thursday sometimes are shipped at the same day, but more often they are shipped on the boat that leaves one week later. After the containers are loaded on the first boat, the route will take 7 or 8 days. This depends on the departure day of the second boat; leaving on Saturday or Sunday. The containers then arrive on Thursday or Friday in the terminal A. However, it seems that some containers (14%) are not shipped on the first Thursday after they are loaded (the bottom three lines in the figure). We assume that they are shipped one week later and therefore, their lead times are longer. However, a clear explanation why these containers are not on the first boat remains unclear. A possible reason is that the boat is full or that the containers arrive too late at the harbor. One outlier in the data is the one container that is loaded on Friday and arrives 7 days later in the terminal (first line in the figure). Possible reasons for this extraordinary route are: the boat left one time on Friday, the data was not registered correctly or the container was send by road due to a rush delivery. However, no explicit explanation for the route of this one container has been found.
0 20 40 60 80 100 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 # o f co n tain e rs
Lead times (days)
Lead time (before 31-3-2012)
Departure day: Monday Departure day: Tuesday Departure day: Wednesday Departure day: Thursday Departure day: Friday Departure day: Saturday Departure day: Sunday
0 10 20 30 40 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 # o f c o n taine rs
Lead time (days)
Lead time (after 31-3-2012)
Departure day: Monday Departure day: Tuesday Departure day: Wednesday Departure day: Thursday Departure day: Friday Departure day: Saturday Departure day: Sunday
Figure 13 lead time containers (data before 31-3-2012)
33 The replenishment lead time appears to be different for the containers that are transported before 31-3-2012, compared to those that are transported after 31-3-2012. In the next chapter we use the information regarding the lead time in the analyses. During analysis 1 (compare with history, as mentioned in the methodology) we make use of the actual lead time and therefore use the data concerning the entire year. For the analyses 2 and 3 we will work with the replenishment lead times from 31-3-2012 even though this figure is based on four months of data only. We therefore measure the lead time as follows: days between loading day and Thursday + 7 days. In 14% of the cases the containers take 7 days longer due to the fact that the container is not shipped on the first boat.
6.6 Necessary information
This section deals with the information that needs to be shared in order to utilize the policies and the quality of this information. Forslund and Jonsson (2007) mention that the receiver of the information must judge the quality of this information. The two line planners are asked to give their judgment about the information concerning the inventory level and the forecast they can use to plan the replenishment to terminal A. Three variables: in time, accuracy and convenient to access mentioned by Forslund and Jonsson (2007) are used for judging the information quality. Moreover, we use performance measures to evaluate the forecast for the different materials.
6.6.1 Line planners judgment
The information concerning the inventory level at the terminal and the forecast in SAP that can be used to plan the replenishment to terminal A are:
1. Information regarding the past weekly delivery of a certain product. This information is brought forward by the shipping-agent and received on a monthly/weekly basis.
34 2. Information regarding the number of containers in the terminal, the recently delivered containers and the number and date of planned containers to arrive at the terminal. This information is received from the shipping-agent on a daily basis.
3. A forecast in SAP. This forecast is made by the account-manager for a one month period. 4. The inventory level in SAP.
Both line planner use information 1 and 2. But one line planner mentions that information number 1 is not free from mistakes. The weekly purchased amount sometimes differs from the information in the daily excel files. Furthermore, the daily received excel files are not completely accurate and convenient to access. For example, the file indicates that two containers have left the terminal and are delivered to the customer. The next day, the file presents these two containers as if they are still in stock. Therefore, this line planner generates her own monitoring excel file to indicate these minor mistakes. The information in SAP is not utilized by either of the line planners. The information is delayed and some delivered containers from a year ago are still present in SAP.
We conclude from the line planner’s judgment that the received information is in time. However, the accuracy and the convenience to access the information stays behind, especially for the information in the SAP system.
6.6.2 Performance measures to evaluate the forecast
The SAP system contains a forecast for the demand, see figure 16, that is derived by the account manager. We use the BIAS, MAD (mean absolute deviation) and MSD (mean square deviation) measures for evaluating the forecast (Hopp & Spearman, 2008).
The three measures are calculated as follows:
MAD = ∑ | ( ) ( )|
MSD = ∑ [ ( ) ( )]
BIAS = ∑ ( ) ( )
35 Figure 16 Forecast SAP and actual demand
Forecast
product MAD MSD BIAS Material A 6,5 74,6 2,0 Material B 4,9 32,1 4,8 Material C 2,3 7,5 0,9 Material D 3,6 37,2 -3,6 All products 15,1 337,4 14,0
Table 5 performance measures for the forecast in SAP (demand in containers, 25.500 ton each) (Hopp & Spearman, 2008) We used times series forecasting to see if it would be possible to forecast the demand better. Exponential smoothing (with a linear trend) was used, since this technique computes a smoothed estimate as a weighted average of the most recent observation and the previous smoothed average (and it also computes a smoothed trend or slope in the data) (Hopp & Spearman, 2008). For each material we used data from financial year 2011 to find the optimal values for α and β (the smoothing constants that are between zero and one). The Solver in Excel, with the objective of minimizing the MSD, and the constraint that the BIAS must be positive, displayed the optimal smoothing constants for each material based on the exponential smoothing (with linear trend) technique, see table 6. We can conclude that the forecast derived through the use of time series forecasting is slightly better for material A and C than the forecast in SAP.
Optimal values for α β MAD MSD BIAS Material A 0.745 5.1 40.5 0.50 Material B 0,564 0,498 5,6 40,6 0,01 Material C 0.651 1.1 1.5 0.20 Material D*
Table 6 The optimal values of α & β for each material to develop the best forecast based on exponential smoothing (with or without linear trend). * No actual demand in FY 2011 was available for this material
6.7 Linking theory to practice
We finalized the analyses concerning the case study. It is therefore time to link the theoretical policies mentioned in the research framework to the case study at COMPANY X and answer research questions 1 and 2. 0 500 1000 1500 2000 2 3 4 5 6 7 8 9 10 11 12 kt o n s o f p ro d u ct time (period)
Forecast and actual demand for all products
36 Table 7 shows which characteristics of each policy conform to the case study. The assumptions used in academic literature regarding demand and lead time distribution are not that restricting to COMPANY X. The MRP, continuous base stock and periodic (s,S) and base stock policy are directly applicable based on the demand and lead time distribution. Only, the continuous (s,S) policy does not seem to be investigated for a situation where both demand and lead time are stochastic.
Policy Demand Lead time
MRP X X
continuous base stock X X
continuous (s,S) X
periodic (s,S) X X
periodic base stock X X
Table 7. The X represents the similarity between the values in the case study and the use of the policy regarding demand and lead time
The case study revealed that COMPANY X cannot permit for stock out to occur, since customer demand should be satisfied at all times. Moreover, the supply chain structure of COMPANY X is dual-mode, since a normal route and a rush delivery route can be taken to replenish terminal A. The investigated literature only addressed dual-mode supply chain structures where stock out occurs. So, the policies are not directly applicable to the case study of COMPANY X. The assumptions made in literature regarding stock out and a simple supply chain structure are restrictive to COMPANY X. COMPANY X makes use of a rush delivery, which is more expensive than the normal route, to react to an imminent stock out in the terminal. Therefore, we must consider the supply chain structure of COMPANY X as a dual-mode supply chain structure. The goal of this structure is to prevent the terminal from running out of stock. Chiang (2003) and Teunter & Vlachos (2001) trigger in their research an emergency order based on the inventory position. However, the inventory position is the same or has a higher value than the actual inventory level and therefore stock out can occur. To prevent stock out to occur, we would suggest placing a rush (or emergency) order whenever the inventory level will drop below a certain level (rush order level). In addition, when the lead time of this rush order is small, say one day, no stock out will occur. The goal is to find an optimal rush order level, so that no stock out occurs. To the best of our knowledge such a dual-mode supply chain structure where stock out is not allowed, has not yet been investigated in literature.
