Quantifying the Impact of Deployment Practices on Interplant Freight Volatility by Kurn Ma Master of Science, Mechanical Engineering, University of Nevada, 2007 Bachelor of Science, Mechanical Engineering, University of Nevada, 2004 and Manish Kumar Post Graduate Diploma in Management, Marketing & Operations, LBSIM, Delhi, 2004 Bachelor of Engineering, Chemical Engineering, Punjab Technical University, 2002 SUBMITTED TO THE ENGINEERING SYSTEMS DIVISION IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF ARCHVES MASTER OF ENGINEERING IN LOGISTICS MASSACHUSE TT 7!NPTrUTE OF TECH N0LA.LGY AT THE JUL 16 2015 MASSACHUSETTS INSTITUTE OF TECHNOLOGY JUNE 2015 LIBRARIES ( 2015 Kurn Ma and Manish Kumar. All rights reserved. The authors hereby grant to MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part in any medium now known or hereafter created. Signature of Author.... Signature redacted Master of Engineering in Logistics Program, Engineering Systems Division Signature redacted Signature of Author..... May 8,2015 Master of Engineering in Logistics Program, Engineering Systems Division May 8, 2015 Signature redacted ........................... Certified by....................... Dr. Roberto Perez-Franco Research Associate, Center for Transportation and Logistics 1/ .. 01 Sianature redacted / Accepted by...................... A Thesis Supervisor .............................. Dr. Yossi Sheffi Director, Center for Transportation and Logistics Elisha Gray 11 Professor of Engineering Systems Professor, Civil and Environmental Engineering Quantifying the Impact of Deployment Practices on Interplant Freight Volatility by Kurn Ma and Manish Kumar Submitted to the Engineering Systems Division on May 8, 2015 in Partial Fulfilment of the Requirements for the Degree of Master of Engineering in Logistics Abstract As the US economy recovers from the recession of 2008, demand for freight transportation is getting stronger. However, the trucking industry is not ready to take on this growth in volume due to a shortage of truck drivers. From a shipper's perspective, availability of transportation is an important concern that needs to be addressed to ensure customer satisfaction, realize growth and keep costs down. Shippers can enhance their carriers' ability to provide consistent trucking capacity by reducing the variability of freight demand. This thesis creates a simulation model of deployment processes at a consumer packaged goods company to evaluate relationship between transportation variability and various management levers. Through the analysis of the simulation runs, the effects of management levers on the freight volatility are quantified. The findings of the research show that actively limiting the truckloads sent downstream, and thus eliminating the freight volatility through internal policies is a potential solution. The thesis concludes by presenting the benefits and trade-offs of this approach on the logistics costs of the sponsor company. If the company sticks to the current policy of immediate shipment as the need arises, the thesis shows that the best deployment schedule is a bi-weekly one. Thesis Supervisor: Dr. Roberto Perez-Franco Title: Research Associate, Center for Transportation and Logistics 2 Acknowledgements On behalf of Manish Kumar: I thank the entire SCM family for their support throughout my time at MIT. I would like to especially thank our thesis advisor Roberto Perez-Franco for his continued guidance and ideas. I am thankful to my wife and daughter, Ashmita and Aadhya, for their loving support that made MIT possible for me. I thank my parents for always believing in me and for their unconditional support. I am thankful to our guides from our sponsor company for getting all the relevant data and their ideas during the course of this thesis. On behalf of Kurn Ma: I would like to thank our advisor Roberto for his guidance during this project. I am particularly grateful for the support I have received from my family, both the one back home and the one here at MIT. 3 Table of Contents Abstract ................................................................................................................................................... 2 Acknow ledgem ents..................................................................................................................................3 1. Introduction..........................................................................................................................................8 1.2. Scope of Research ........................................................................................ 9 1.3. Sponsor Com pany Overview .................................................................................................... 9 1.3.1. Supply Chain Organization ........................................................................................... 10 1.3.2. Deploym ent Process Overview ................................................................................... 11 1.4. Thesis M otivation........................................................................................................................12 2. Literature Review...............................................................................................................................13 2.1. Trucking Industry in the United States................................................................................... 13 2.1.1. Driver Shortage ............................................................................................................... 14 2.2. Past Research in Volatility M itigation ..................................................................................... 15 3. M ethodology......................................................................................................................................16 3.1. Inventory Replenishm ent Policy.............................................................................................. 18 3.1.1 Dispatch Quantity Calculation....................................................................................... 18 3.2. Forecast Accuracy and Bias.................................................................................................... 19 3.2.1. Dem and Sim ulation.................................................................................................... 19 3.3. Production Output ..................................................................................................................... 3.3.1. Production Output Variability..................................................................................... 3.4. Cost Calculations ........................................................................................................................ 4. Sim ulation M odel 21 21 23 ............................................................................................................................... 24 4.1. Forecasts .................................................................................................................................... 24 4.2. Truckloads .................................................................................................................................. 24 4.3. Production and Review Schedules ......................................................................................... 26 4.3. M odel Validation........................ ................................................................... 28 5. Results and Discussion ...................................................................................................................... (30 5.1. Unm anaged Scenario Analysis with Biased Forecasts........................................................... 31 5.1.1. Im pact on Shipm ent Variability................................................................................... 31 5.1.2. Im pact on Inventory Holding Costs............................................................................ 33 5.2. Unm anaged Scenario Analysis w ithout Biased Forecasts ...................................................... 35 5.2.1. Im pact on Shipm ent Variability................................................................................... 35 5.2.2. Im pact on Inventory Holding........................................................................................ 36 4 5.3. M anaged Truckload Simulations ................................................................................................ 39 6 . C on c lu sio n ......................................................................................................................................... 4 3 7. Future W ork ...................................................................................................................................... 44 Refe re nces ............................................................................................................................................. 4 5 A p p e n d ix A ............................................................................................................................................ 47 5 List of Figures Figure 1: Locations of plants, warehouses, and distribution centers across the U.S........................10 Figure 2: Process m ap of the com pany supply chain ........................................................................ 11 Figure 3: Daily truckloads from plant to DC ....................................................................................... 12 Figure 4: Expected truck driver shortfall in the U S. ........................................................................ 14 Figure 5: Increase in line haul rates in the U .................................................................................... 15 Figure 6: Sim ple depiction of sim ulation m odel ................................................................................ 17 Figure 7: Distribution fitting of production variability ..................................................................... 22 Figure 8: Input variables for m odel baseline verification................................................................. 28 Figure 9: Baseline dem and com parison ........................................................................................... 29 Figure 10: Baseline truckload com parison ......................................................................................... 29 Figure 11: W eekly truckloads under scenarios 1, 2 & 3................................................................... 32 Figure 12: W eekly truckloads by product group .............................................................................. 33 Figure 13: Average inventory at DC for 3 deployment schedules and 2 MAPE values.....................34 Figure 14: Weekly truckloads under scenario 13, 14, & 15 .............................................................. 36 Figure 15: Average inventory at DC for 3 deployment schedules without Bias and production v a riatio n ................................................................................................................................................ 37 Figure 16: Average inventory in no. of days at DC for different schedules ...................................... 38 1......................................................................................... 40 Figure 18: M anaged truckload scenario 2......................................................................................... 40 Figure 17: M anaged truckload scenario 6 List of Tables Table Table Table Table Table Table Table Table Table Table 1: Daily figures of operations and processes of the supply chain ........................................... 2: Sample holding cost calculation ........................................................................................... 3: Forecast-related input variables and descriptions ............................................................... 4: W eekly truckload projections.............................................................................................. 5: Replenishment related variables and descriptions ............................................................ 6: Configurations for simulation run 1 .................................................................................... 7: Configurations for 2nd scenario analysis .............................................................................. 8: Managed scenario forecast sensitivity parameters ............................................................ 9: M anaged scenarios 1 through 4 financials.......................................................................... 10: M anaged scenarios 5 through 8 financials........................................................................ 7 10 23 24 25 27 31 35 39 41 41 1. Introduction The transportation of finished goods is a critical part of any supply chain. In the consumer packaged goods (CPG) industry, transportation management is commonly built on a foundation of contracts signed with carriers. The majority of these contracts are with carriers in the trucking industry because trucking provides a combination of flexibility, speed, and pricing that is appropriate for the value and nature of CPG products. As supply chains have matured and become leaner, companies have started to look at transportation as an area to build a competitive edge. In the U.S. especially, recent regulations in the trucking industry and a shortage of qualified truck drivers have prompted shippers to focus on improving transportation efficiency. Many companies are now adjusting their transportation strategies to be more aligned with the changing nature of the trucking industry. Through better coordination and long-term relationships with the carriers, CPG companies are seeking ways to reduce spending on freight transportation while still maintaining a consistent and high level of service to their customers. Despite the increased focus on improving transportation efficiency, shippers are still impacted by the bullwhip effect, which is the increasing volatility of demand as one goes further up the supply chain (Lee et al., 2004). This increase in volatility often leads to demand spikes which causes companies to ship more goods downstream to meet customer requests. Consequently, the companies will need to acquire more trucks to ship the extra volume of goods. If the contracted carriers are unable to supply the extra trucks, then the contingency plan for most shippers is to obtain trucks from the spot market. In recent years however, the cost of acquiring trucking from the spot market has risen sharply (primarily due to trucking capacity shortage), which is the main factor putting pressure on companies to improve transportation efficiency and reduce costs. 8 1.2. Scope of Research This thesis studied masked data of a freight lane between a production plant and a distribution center (DC) for a CPG company. The focus was to quantify the effects of internal deployment policies on freight volatility and to make recommendations that will reduce this volatility and the overall transportation costs. The production, shipment, and demand data for the 15 largest product groups by volume in the freight lane were studied. The study was done on volume considering the fact that products in CPG industry are voluminous and hence cube out before they weight out. These 15 product groups comprise approximately 50% of the overall freight transported from the plant to the DC. The initial phase of the project was the characterization of the data to develop formulations that reflect the relationships between the inputs and outputs. These formulations were the foundation of the simulation model used to determine the impact that internal policies and factors have on the freight volatility. The internal policies and factors studied are following: production variability, production schedules, demand forecast error, inventory review periods, safety stock levels, and stock replenishment points. This wide list of exogenous and endogenous factors was chosen in accordance with sponsor company's requirements and on the basis of literature review. Through the simulation model, the sensitivity of the freight shipment volume was measured against the model inputs to determine which variables have a meaningful impact on the freight volatility. Additionally, the simulation model also analysed how changes to the inputs affect overall costs and service levels. 1.3. Sponsor Company Overview The project sponsor is one of the largest CPG companies in the world with brands serving customers in more than 100 countries. Its products are sold through large retailers such as Walmart, Kroger, Target and others. As the retail market went through consolidation in the US, their sales volume consolidated around a few retailers. As much as 60% of the sales in US market come from 10 retailers. 9 1.3.1. Supply Chain Organization The supply chain organization works behind the scenes to ensure that company's iconic brands are available to customers at the right time and right cost. Table 1 shows the magnitude and complexity of the supply chain organization. Table 1: Daily figures of operations and processes of the supply chain Description ('000) per day Orders 1 Shipments 2 Tenders 3 Cases picked 325 Cases moved in warehouse 6,000 Potential Lane Combinations 23,000 Pallet-Miles 30,000 The company has more than 30 plants in the US producing over 800 stock keeping units (SKUs), and many more distribution centers to service the retail locations. The company also has a large base of external manufacturing partners that support the production needs when required. The overall supply chain n etwork in the US can be seen in Figure 1. Source: Sponsor Company Figure 1: Locations of plants, warehouses, and distribution centers across the U.S. 10 The focus of this project is on the dry retail portion of the company's supply chain. Dry retail is the term used for products that are transported in ambient temperatures and do not require special warehousing specifications. These products have sufficient shelf life to allow unconstrained supply chain planning. The dry retail products are manufactured in 8 plants across the US and customers are mainly served through shipments from the distribution centers. Some companies elect to have orders shipped to their own stores or warehouses directly from the plant. The dry retail supply chain has a structure typical of that in most manufacturing companies. There are several tiers of suppliers, manufacturing plants, finished goods warehouses, and distribution centers which serve the customers. A simple representation of the supply chain process map is shown in Figure 2. Tier N Suppoers r 1 Manjcurirg Warehouses Transpotatin Ret 191 RDCs S*piiers Retiler stcre shees R.LLL Source: Sponsor Company Figure 2: Process map of the company supply chain 1.3.2. Deployment Process Overview The production scheduling is based on monthly forecasts fed into the company's enterprise resource planning (ERP) system. Throughout the production process, finished goods are stored at a nearby warehouse to distribute to the DCs as needed. Dispatches of finished goods from the plant to the DCs are determined by the deployment teams, which are responsible for monitoring the inventory levels of their respective product groups. Inventory levels are measured in terms of days of service (DOS) that can be supported with the stock available at the DC. During the periodic reviews, if the deployment team is alerted by the ERP system that the DOS level for a product falls below a predetermined number, then a replenishment order is placed to obtain more stock from the plant warehouse. 11 The combined orders from the deployment teams aggregate into the daily number of truckloads of products from the plant to the DC. Because the replenishment orders are determined by inventory target levels and highly variable customer demand, the orders themselves are also highly variable. The variability of the daily truckloads sent from a plant to a given DC can be seen in Figure 3. 18 16 14 12 0 10 8 6 4 2 0 Date Source: Sponsor Company Figure 3: Daily truckloads from plant to DC 1.4. Thesis Motivation Mitigating the freight volatility is the central objective of this thesis because of the impacts it has on transportation costs and service levels. The sponsor company will determine the necessary amount of trucks needed per week throughout the fiscal year based on demand forecasts. When customer demand is higher than forecasted, it drains the available inventory at the DC faster than anticipated, which in turn increases the replenishment dispatch from the factory to the DC. These higher than anticipated dispatch quantities require the procurement of trucks beyond what was contracted from the carriers. 12 With the current strain on capacity in the trucking industry, it is often the case that carriers will not be able to supply the extra trucks needed. The contingency plan for the sponsor company in these instances is to hire trucks from the spot market, which have significantly higher rates than those of contracted carriers. It may also be the case that hiring trucks from the spot market is too difficult or too prohibitive, which can compromise the company's ability to service customer demand. Because the strain on trucking capacity is projected to worsen, there is an impetus to reduce the freight variability and to reduce the dependence on the spot market. 2. Literature Review Historically, the supply of trucking was such that shippers could meet the high shipment requirements caused by customer demand spikes. In recent years, however, the trucking industry has experienced a shortage of drivers. Due to a variety of factors, the rate of drivers being hired has not been sufficient to meet the growing demand for trucking from shippers. The resulting limited supply of trucking has compromised the ability of shippers to acquire extra trucking from their contracted carriers in high-demand situations. The situation caused by the driver shortage has also increased the rates of trucking acquired from the spot market. This literature review examines the causes of the driver shortage and the response of shippers to the situation. 2.1. Trucking Industry in the United States The trucking industry is the largest component of the freight transportation market in terms of loads carried and miles travelled. In 2011, the trucking industry generated $604 billion in revenue, accounting for more than 80% of the US freight market spending. In terms of tonnage, trucking moved 9 billion tons of goods which was 67% of the total US freight transported (USDOT, 2011). The industry is very fragmented, with approximately 97% of carriers owning 20 or fewer trucks (Rivera, 2014). This fragmentation adds to the difficulty of suppliers' attempts to build long-term strategic relationships with carriers in order to manage costs. 13 In general, the industry can be segmented in to three areas: 1. Full Truckload (FTL) 2. Less than Truckload (LTL) 3. Parcel Each of these segments has its own cost structure and industry dynamics. FTL involves the point to point shipment of fully loaded trailers. In the LTL segment, small loads are collected in a local area and consolidated for a long-haul shipment. Parcel shipments are the short-distance delivery of individual packages as seen in the domestic mail delivery market. 2.1.1. Driver Shortage During the recession of 2008, overall consumer demand decreased, which reduced the need for freight transportation services. During this time, companies and independent operators in the trucking industry either downsized or went out of business. As the economy recovered in the ensuing years, demand for goods began to rise as did the need for commercial freight trucking. However, the number of available drivers has not been growing at a rate fast enough to meet the increasing demand for trucking. According to the American Trucking Associations (Cassidy, 2014) the shortfall between the number of drivers required and the number of drivers available is expected to reach 239,000 by 2022. Figure 4 depicts the projected trend lines. Millions Trend-Une for Number of Tractor-Trailer Drivers Demanded 2.00 - 239,000 potential shortfall 1.75 - Trend-line for Number of Tractor-Trailer Drivers Supplied 1.50 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Source: American Trucking Associations Figure 4: Expected truck driver shortfall in the U.S. 14 __ -_ - --- ___ --- - I,-- -- __ --a- .............. I One of the reasons for this shortage of drivers is that the average age of the current driver workforce is 55, and that presently more drivers are retiring than are being hired (Roberts, 2014). Furthermore, recent government safety regulations have restricted the number of hours drivers can work per day. These regulations have also required additional monitoring of trucks on the road. The changes caused by the new regulations have led to increased discontent among truck drivers and workplace turnover has reached as high as 106% (Cassidy, 2014). Expectedly, this shortage of available trucking has led to increasing freight shipping rates. By August 2014, rates for contracted trucking rose by 8% on the year and spot rates rose by 14% (Carey, 2014). The rise in truck per mile line-haul rates is shown in Figure 5. Cass Truckload Linehaul Index"' A measure of changes In per-mile truckload linehaul rates. 127 125 123 121 119 117 1I5 113 109 107 105 103 101 99 97 JAM IJ NJWJ SNJA 205 0 2M5 ca" 2006 .IJni I$NJAWJ %N)NW1%NJJ 2007 n100~lwn sve"a 2009 2009 2010 n MINJMN IJd 2011 2012 AI %PiJ5 MJ 2013 INJAI 2014 %N 2015 WK and AwN0000 Parnno Figure 5: Increase in line haul rates in the U.S. 2.2. Past Research in Volatility Mitigation The main source of volatility in a supply chain is the inherent uncertainty in customer demand. In most cases, companies will plan production, warehouse inventory levels, logistics, finances, and other components of their operation based on demand forecasts. Therefore it is imperative to have the information inputs that go into developing demand forecasts be as accurate as possible. One common method of ensuring the quality of these inputs is to have a collaborative relationship among stakeholders in the supply chain. This involves having high visibility of data between 15 suppliers, manufacturers, shippers, and retailers (Wen, 2011). By sharing data, the various echelons of the supply chain are able to better align expectations and plans, reducing the potential to add further volatility through individual actions of the stakeholders. Another popular coordination technique used by suppliers and retailers is vendor-managed inventory (VM I), characterized by supplier-managed inventory at a downstream customer. VMI has been shown to be effective in reducing stock outs and increasing the efficiency of inventory management. Another benefit of VMI is the autonomy that a supplier has in deciding when to ship a replenishment order to the retailer. By having control over how to consolidate and ship goods to the retailer, suppliers are able to have a direct impact on the volatility of their respective supply chain link (Cetimkaya, et al., 2000). Companies have also improved their warehouse loading operations to increase the overall dependability of their logistics department. For example, through efficient facility designs carriers can load and unload trailers faster, reducing driver wait time and increasing efficiency. Besides ensuring on time deliveries, making facilities more convenient to load from contributes to making the shipper more attractive to the shippers in terms of accepting loads. The shipper is, therefore, more likely to acquire extra trucking when needed and better able to meet its transportation needs (Fugate, et al., 2009). 3. Methodology A model was created using Microsoft Excel and Visual Basic for Applications (VBA) to approximate the real world operations of the sponsor company's supply chain as it pertains to the transportation between a factory and a DC. Using data collected from the previous fiscal year, the relationships between production outputs, customer demand, and stock reorder policies were formulated to create the foundation of the model. The model is set up so that daily transactions are simulated over a year. Each day there are values for production output, inventory position, forecasted demand, actual demand, and replenishment dispatches. While it is not a depiction of the 16 actual model, Figure 6 gives a general idea of what kind of data is analysed in the simulation. All product quantities are given in terms of pallets. Description Date Product 1 Product 2 Product 3 Product 4 Product 5 Product 6 Product 7 Product 8 Product 9 Product 10 Product 11 Product 12 Product 13 Product 14 Product 15 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 11/11/2013 Production Plant Inv Forecast Actual APE DC Inv Target Inventory Reorder Point Dispatch Truckload 651.05 612.41 81.53 31.92 140.74 15.32 17.52 576.33 1077.94 28.75 16.31 43.49 339.76 410.44 1465.80 228.93 32.25 4.31 1.68 7.46 0.81 0.93 30.74 56.24 1.51 0.86 2.31 18.03 21.37 43.25 12.13 34.42 4.09 1.81 6.99 0.76 0.91 31.72 57.99 1.44 0.91 2.45 17.19 19.76 44.12 13.08 6% 5% 7% 7% 6% 2% 3% 3% 5% 5% 6% 5% 8% 2% 7% 353.6 46.8 18.4 80.8 8.8 10.2 339.9 612.3 16.4 9.3 24.5 197.3 240.2 475.5 133.7 645.0 86.3 33.7 149.2 16.2 18.6 614.8 1124.9 30.2 17.2 46.2 360.7 427.4 864.9 242.6 483.7 64.7 25.3 111.9 12.1 13.9 461.1 843.6 22.7 12.9 34.7 270.5 320.5 648.7 181.9 193.8 26.3 10.2 45.8 4.9 5.7 192.1 328.1 9.0 5.1 14.3 111.1 123.8 270.5 74.3 3.8 0.5 0.2 0.8 0.1 0.1 3.0 5.2 0.1 0.1 0.2 1.7 1.9 4.3 1.2 Figure 6: Simple depiction of simulation model During the simulation, all 15 products are evaluated one day at a time. In each day, the production output, actual demand, and arriving dispatches are simulated. These simulated values are then incorporated into a resulting inventory position at the end of the day. These end of day inventory positions are then used to simulate the production output and dispatches for the following days in the model. Lastly, the dispatches are converted into equivalent truckloads. In summary, for each day of the simulation the actual demand will be subtracted from the DC inventory. If the date during the simulation matches the day that a deployer will review the inventory position, then the DC inventory is compared to the reorder point. Whenever deployer review finds the DC inventory below the reorder point, then a dispatch is requested in an amount that will bring the DC inventory up to the target level. This dispatch will arrive the next day of the simulation and be added to the DC inventory. The dispatch volume will also be subtracted from the plant warehouse inventory. The plant warehouse inventory will be replenished from time to time based on the production schedule. 17 3.1. Inventory Replenishment Policy The sponsor company utilizes a periodic review system to determine inventory replenishment at the DC. The 15 products analysed in the simulation model are separated into four different production groups. Each production group has its own production schedule as well as a dedicated deployer that analyses inventory levels and determines the reorder quantities. Currently, the deployers review the inventory position for their respective product groups once a week. During the reviews, the deployers will assess the current inventory level at the DC and compare it to the reorder point. If the current DC inventory is below the reorder point, the deployer will order a dispatch from the plant to bring the DC inventory up to a target level. The sponsor company analyzes inventory levels in terms of days of service (DOS). For example, if the inventory falls below 15 DOS, then a dispatch will be ordered to bring the inventory up to 20 DOS. Under the current policy, all dispatch requests will be fulfilled. For the plant-DC link studied in this thesis, replenishment dispatches typically arrive the next day. Because the deployers conduct their inventory reviews independent of one another, spikes in demand for multiple products will compound and contribute to the overall need for trucking on a given day. 3.1.1 Dispatch Quantity Calculation The dispatch quantity is determined by a simple comparison of the current DC inventory and pre-set targets. Q - ic0, IC< IC > s where: Q= Reorder quantity s = Reorder point S = Order up to level Ic= Current DC inventory 18 (1) 1 Due to the seasonality of customer demand, the values for R and T are different for each fiscal month. These values are predetermined for each product and month based on historical data. 3.2. Forecast Accuracy and Bias The metric used by the sponsor company to assess forecast accuracy is the mean absolute percentage error (MAPE). The MAPE measures the absolute deviation of the forecasted demand values as a percentage of actual demand values, and then averages them. MAPE = E *100 (2) where: F = Forecasted demand on day i Ai= Actual demand on day i. In the simulation model, the MAPE for the demand forecasts are bounded between 0% and 100% because it makes more intuitive sense to measure accuracy between those two extremes. Bounding the MAPE is also the standard practice of the sponsor company. 3.2.1. Demand Simulation In the simulation model, the forecasted demand values are fixed and are based on historical values. To simulate the actual demand, the forecasted value will be modified by a simulated forecast error. For example, if the goal is to determine how an overall MAPE will affect the actual demand in the simulation, the absolute percentage error (APE) of the daily forecasts will be simulated by the following. APE ~ N(MAPE,2/3*MAPE) (3) This means that the APE for a given product on a given day will be approximated by sampling a value from a normal distribution with a mean value of MAPE and a standard deviation of 2/3*MAPE. A 19 standard deviation that is two-thirds of the MAPE is used in the simulation to mirror the historical data (Appendix A). Once the APE is simulated, the actual demand is determined by back calculating from the MAPE formula. The first step of the back calculation gives us the following. APE* |AI = IFi-Ai| (4) where: F = Forecasted demand A = Actual demand Because the formula uses the absolute difference between the actual and forecasted values, there are multiple possible values for the simulated actual demand to take into account the fact that it is possible to over-forecast, under-forecast and accurately forecast the demand. F/(1 + APE), Ai < Fi Ai = F/(1 - APE), Fi > Ai Fi, Fi = Ai (5) To determine whether demand is over or under forecasted in the model, values between 1 and 100 are randomly generated for each line of the spreadsheet. If it is assumed that there is no forecast bias, then for values between 1 and 50 the demand will be over-forecasted and for values between 51 and 100 it will be under-forecasted. Once it is determined whether the demand is over or under forecasted, the appropriate formula is used to get the final actual demand value. It can be seen that if the APE is zero, meaning that the demand is accurately forecasted, it does not matter which formula is used because both will result in A = F. For scenarios in which there is a bias, determining which formula to use will be based on an adjustment to the comparison with the randomly generated numbers. For example, if there is a 67% bias towards over-forecasting, then the 20 over-forecasted demand formula will be used when the randomly generated number is less than or equal to 67. 3.3. Production Output The production schedule of the manufacturing plant is determined by forecasting the inventory position of the whole system (plant and DC) two weeks ahead. Therefore, the production plan is calculated by the following. Ps = S - (IP +ID +=1 -~ FI) (6) where: S = Target inventory level for the whole system P15 = Amount to be produced on day 15 (2 weeks ahead) Pi = Amount planned to be produced on day i Fi = Forecasted demand on day i Ip = Current inventory at the plant warehouse ID = Current inventory at the DC warehouse. In this formulation, the model will take into account the current inventory level of the system and calculate the expected position 2 weeks ahead based on forecasted demand and other planned production in the same time period. If that position is lower than the predetermined system inventory target, then the plant will plan to produce an amount to reach the target. 3.3.1. Production Output Variability In analysing the sponsor company's weekly production data, we found a noticeable difference between what was planned to be produced and what was actually produced. The ratios of actual production to planned production were tabulated for each week and plotted using a software 21 package called EasyFit. A distribution was then fitted over the data and the result can be seen in figure 7. Probabdity Density Function 08 Cauchty I 072 I a 00428 1 0 064 056 048 04 0 32 024 0 16 008 0 02 04 06 08 1 12 14 16 0 Histogram - CauChy Figure 7: Distribution fitting of production variability The resulting best-fit distribution is a Cauchy distribution with a median parameter of 1.0 and a scale parameter of 0.0428. Therefore, in the simulation, each time there is planned production for a product, that amount will be multiplied by a factor that is randomly drawn from a Cauchy distribution based on the previously mentioned parameters. In summary, whenever there is planned production, the amount to be produced will be multiplied by factor determined by the following formula. PF = 0.428 * tan(7r * (RAND - 0.5)) + 1.0) where: PF = Production factor RAND = A random percentage between 0 and 100. 22 (7) In order to better reflect historical values, the values obtained from formula (7) are bounded between 0 and 2 in the simulation, since there is a possibility of obtaining values beyond these boundaries from the distribution that would be implausible in reality. 3.4. Cost Calculations After each simulation run, the total inventory holding cost of the DC inventory and the total cost of the truckload shipments are calculated. This is done by determining the average inventory of each product per week. The average inventory is then multiplied by the cost of goods sold to determine the overall value. The weekly inventory value is then multiplied by the holding rate, which for the sponsor company is 7%, to get the holding cost. Because the 7% holding rate is a yearly rate, the final step in calculating the holding costs is to multiply by a factor of 1/52 to get the cost of holding the inventory for one week. Table 2 shows an example calculation of the holding costs for a given week. Table 2: Sample holding cost calculation Description Week Average Inventory (Pallets) Cost per pallet 55,687.40 Product 1 610.97 14,601.26 Product 2 82.16 8,743.82 Product 3 32.09 7,750.27 Product 4 141.36 7,364.23 Product 5 15.35 5,391.68 17.69 Product 6 5,189.78 Product 7 584.57 3,910.24 Product 8 1070.42 3,068.58 Product 9 28.60 2,770.05 Product 10 16.29 43.87 2,322.50 Product 11 1,704.67 Product 12 343.06 1,446.11 406.72 Product 13 1,227.11 819.29 Product 14 997.48 228.69 Product 15 Total Value $ 34,023,432.70 $ 1,199,608.65 $ 280,564.72 $ 1,095,582.49 $ 113,037.62 $ 95,401.06 $ 3,033,765.97 $ 4,185,604.31 $ 87,753.39 $ 45,135.63 $ 101,882.98 $ 584,807.69 $ 588,164.43 $ 1,005,356.87 $ 228,118.57 Holding Cost $ 45,800.77 $ 1,614.86 $ 377.68 $ 1,474.82 $ 152.17 $ 128.42 $ 4,083.92 $ 5,634.47 $ 118.13 $ 60.76 $ 137.15 787.24 $ $ 791.76 $ 1,353.37 $ 307.08 For the calculation of the truckload costs, the dispatch loads are differentiated between contracted loads and spot market loads. These loads are multiplied by the rates the sponsor company paid over the time period studied. For regular contracted loads, the rate is $615 per truck. 23 For spot market loads, it is $873 per truck. For the purposes of this thesis, the costs associated with production change-over and set up at the factory are not considered. Only the effect of production output variability on the DC operation costs are taken into account. 4. Simulation Model This section discusses the assumptions, framework, and input variables of the simulation model. 4.1. Forecasts In the model, the forecasted demand is fixed and the actual demand is simulated. The forecasted demand values are set based on the historical data provided by the sponsor company. However, the data provided only includes the monthly forecasted demand for the products. To determine the daily forecasted demand, the monthly forecasts are simply divided by the number of days in the appropriate month. It is important to note that the sponsor company does not use a traditional calendar in its demand planning process. Rather, they divide the year into 12 fiscal months, with each month having either 4 or 5 weeks. Therefore, each fiscal month will have either 28 or 35 days for all relevant calculations in the model. Table 3 shows the relevant input variables associated with the forecasts as they pertain to the model. Table 3: Forecast-related input variables and descriptions Description Variable Name MAPE BIAS Percentage value representing the mean absolute percentage error of the weekly demand for all 15 products overthe entire year. This value will determine the shape of the distribution that the daily forecast error will be drawn from in the simulation. Binary flag which determines whether the forecasts are biased. A "0" means that the forecast is unbiased. A "1" means that the forecast is biased toward overforecasting. 4.2. Truckloads The forecasted demand is used to determine the baseline number of trucks required per week to transport the reorder inventory to the DC. For simplicity, all products belonging to a particular 24 manufacturing platform will have the same load factor. This load factor determines the number of pallets that can fit in a truckload. By converting the monthly forecasts for all the products into their equivalent truckloads and dividing by the respective number of fiscal weeks, the projected number of truckloads required per week is determined. Table 4 shows the final calculation of weekly trucks required in the different fiscal months for the simulation runs. One last consideration regarding the weekly truckload requirements is that, since demand forecasts are different for each month, a "day of supply" has a different meaning for each month as well. Therefore, 15 days of supply in December is different from 15 days of supply in January, for example. To account for the change in inventory position, the weekly truckload requirement in the first week of each fiscal month is adjusted up or down in the simulation. Through this process, the inventory at the DC is adjusted to meet the target level as determined by the input variable in the simulation. Table 4: Weekly truckload projections Month Weeks Total Trucks Weekly Trucks 25.34 101.36 4 JAN 24.48 97.90 4 FEB 22.20 111.02 5 MAR 23.79 95.17 4 APR 23.73 94.93 4 MAY 21.75 108.74 5 JUN 25.31 101.26 4 JUL 22.01 88.03 4 AUG 17.05 85.25 5 SEP 96.72 24.18 OCT 4 26.88 107.54 4 NOV 21.63 108.17 5 DEC It is important to note that for the historical data, there is a discrepancy between the number of trucks that were actually used than what would be calculated using the previously mentioned method. The reason for this difference is that in real life, the deployers are balancing the entire supply network for their assigned product group. This means that they are responsible for moving 25 goods across a number of plants and DC relevant to the products they manage. Also, in the real life operations of the sponsor company, products are pushed from the plants to the DCs even when there are no replenishment orders because of capacity constraints on the plant warehouses. The complexity of the actual deployment practices and the methods of determining the number of truckloads to contract in the real world are too difficult to model for this project. Therefore, for all simulation runs the truckloads sent from the plant to the DC are determined on a pull-based system as inventory at the DC falls below the trigger points. Lastly, the calculated truckloads from Table 4 are the baseline against which simulated truckloads are compared to for all runs of the model. 4.3. Production and Review Schedules Within the model, both the deployer reviews and plant production runs occur only Monday through Friday. For the production schedule, a constraint is in place so that the plant will produce products for only one product group that day. It is assumed that production runs last for only one day, and whatever is produced during the run is immediately added to the relevant product inventory at the plant warehouse. Regarding the deployer reviews, whenever it is determined that a replenishment dispatch is required, a truckload with the ordered goods will arrive the next day. The historical data shows that truckloads arrived at the DC seven days a week and the model is designed accordingly. In the current operations of the sponsor company, all replenishment requests are fulfilled and truckloads of inventory are sent the next day regardless of what is contracted with the carrier. These operational conditions are modelled in the simulation as the baseline scenario. As an alternative, the model provides an option to actively manage the number of truckloads by aggregating the required truckloads required during the current week and shipping them the following week. The required truckloads are also constrained to an upper limit that is determined by the demand forecasts. Lastly, the replenishment dispatches are distributed evenly across all seven days of the following week. A description of the remaining input variables can be seen in Table 5. 26 Table 5: Replenishment related variables and descriptions Variable Name Description DepoyerSchedule Binary flags determining which day of the week the deployers review inventory position for their product group. Production Schedule DC Target DOS DC Reorder DOS System DOS AVS Manage Over Projection Binary flags determining which day of the week the plant produces products for a particular manufacturing platform. Target inventory level of the DC in terms of Days of Service. Inventory level of the DC that triggers a replenishment order. Target inventory level of the entire system (DC and plant warehouses). Binary flag which determines whether the production output is varied. A "0" means that there is no variability and that whatver is planned to be produced will be produced. A "1" means that the planned production amount will be multiplied by a randomlv determined production factor. Binary flag which determines whether the truckload dispatches are managed. A "0" indicates the baseline scenario in which all replenishment requests are fulfilled the next day. A "1" means that the required shipments forthe week are delayed and evenly distributed the next week. Percentage value that determines how much above the forecasted truckloads to set the truckload limit in the managed scenario. 27 4.3. Model Validation Using input variables that match the current real-world conditions of the sponsor company, multiple simulation runs were run to verify that the simulated output of the model were in line with the historical data (Figure 8). Deployer Schedule Plat 2 Plat 3 Plat 1 MON TUE WED THU FRI 1 DC Target DOS DC Reorder DOS System DOS Plat 4 1 1 1 Bias (Ofor "No", 1for "Yes") AVS (0 for 100%, 1 for varying) Manage(O for baseline, 1 for managed) 0 Mape MON TUE WED THU FRI 35% Over Projection Production Schedule Plat 2 Plat 3 Plat I 1 1 Plat 4 1 1 Figure 8: Input variables for model baseline verification 28 After five simulation runs, the aggregate monthly demand in terms of pallets were compared to the historical data (Figure 9). Demand Comparison 8000.00 /000.00 6000.00 5000 00 Hist orIcal --- Sim I -Sim 4000.00 0~ 2 Sim 3 3000.00 - Sim 4 -- Sim S 2000.00 1000 00 0.00 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Figure 9: Baseline demand comparison The truckload dispatches from the simulations were also compared to the historical data (Figure 10). Truckload Comparison 140 120 100 M- 80 0 - Histor ical - Sim 1 Sim 2 Simn 3 60 40 20 0 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV Figure 10: Baseline truckload comparison 29 DEC - Sim 4 - Sim 5 From the comparison figures it can be seen that the simulation outputs are on the same order of magnitude as the historical data. The baseline scenario was therefore deemed acceptable as a reference point to test the sensitivity of the freight volatility to changes in the input variables. 5. Results and Discussion The analysis is primarily focused on two different operating paradigms: unmanaged and managed replenishment shipments. The unmanaged scenarios are based on the current operating conditions in which all replenishment requests are shipped the next day to the DC without regard to transportation costs. The simulations in this case seek to identify how changes in the internal policies of the sponsor company, modelled as the input variables, affect the volatility of the weekly truckload shipments. Changes in the input variables also have an effect on the inventory position and therefore also on the holding costs at DC. Consequently, the metrics analysed for the unmanaged scenario outputs are weekly truckload volatility and inventory holding costs. The managed scenarios, however, essentially eliminate the week to week volatility by delaying and evenly distributing the replenishment shipments. In the model, the replenishment orders in a given week accumulate during the simulation and culminate into a final replenishment order quantity. That final quantity will then be shipped to the DC in the following week, with the shipments divided evenly across the weekdays. The replenishment quantities are limited to the projected number of truckloads based on demand forecasts. If the replenishment quantity exceeds the limit, then the order amounts for each product are proportionally reduced in order to get the final quantity down to the limit. By not always reaching the target DC level through replenishment shipments, there is a risk of not being able to meet customer demand. Therefore, the metrics analysed for the managed scenario outputs are inventory holding costs and lost sales. For simplicity, lost sales are tracked in terms of pallets not delivered to the customer. 30 5.1. Unmanaged Scenario Analysis with Biased Forecasts The parameters used in this simulation run are shown in Table 6. Table 6: Configurations for simulation run 1 Bi-Weekly 45% 20 Daly 45% 20 Weekly 35% 20 15 1 1 15 1 1 1 20 Daily 35% 20 15 1 1 5.1.1. Impact on Shipment Variability Scenarios 1, 2, and 3 explored the effect the deployment review schedule had on shipment volatility. The weekly truckloads for the scenarios are shown in Figure 11. 31 40 35 30 -Sum of 25 Truckloadi C. -Sum 20 of Truckload2 15 tI Sum of Truckload3 10 5 0 1 3 5 7 9 111315171921232527293133353739414345474951 Weeks Figure 11: Weekly truckloads under scenarios 1, 2 & 3 The bi-weekly schedule has lower variability as a whole. Apart from one or two peaks, the band is narrower compared to weekly and daily schedule. This is also confirmed by the standard deviation for the simulations; the standard deviation for the bi-weekly schedule is the lowest, while the daily scenario has the highest variability. It is counter-intuitive to some extent if we consider the variability associated with daily deployment. In a daily deployment schedule, deployers are expected to make decisions immediately as the inventory goes down the reorder level and push dispatch. This almost continuous review should keep the size of dispatch (number of truckloads) relatively similar week to week. The only variation in dispatch should come from the small demand variation. However, a closer look at the data explains why that is not the case, the variation is magnified due to the accumulation of truckload numbers over the week. In daily deployment schedule, we find instances when dispatch is ordered more than once a week which create spikes in number of truckloads dispatched per week. These spikes are not possible in weekly deployment schedule and are much curtailed in bi-weekly schedule. 32 Another example, shown in Figure 12, of similar simulation runs for three deployment schedules but this time at MAPE of 35% and for one product group shows the observation more clearly. The red line represents the bi-weekly schedule output. 35 I q 30 25 - Sum of Truckload4 -Sum of Truckload5 20 -- Sum of Truckload6 15 10 5 0* 1 3 5 7 9 111315171921232527293133353739414345474951 Weeks Figure 12: Weekly truckloads by product group 5.1.2. Impact on Inventory Holding Costs We now go through the simulation output in terms of inventory held at the DC under three different deployment scheduling scenarios. Figure 13 shows the average inventory held under three scenarios and two different MAPE values. The average of inventory 1, 2 and 3 are for the weekly, bi-weekly and daily deployment schedules for a MAPE value of 45% whereas the 7, 8 and 9 line graphs are for a MAPE value of 10%. 33 280 260 240 220 - 20 -- 200Inventory 0 180 - 160 - 140 - 120 - Average of Inventory 1 Average of 2 Average of Inventory 3 Average of Inventory 7 Average of Inventory 8 Average of Inventory 9 100 1 3 5 7 9 1113151719 2123 25 27 29 3133 3537 39414345474951 Weeks Figure 13: Average inventory at DC for 3 deployment schedules and 2 MAPE values It is noticeable that the change in MAPE values does not impact the inventory holding. This is not in accordance with the inventory management theory which says that average inventory increases with MAPE. We attribute this observation to the fact that we have used constant reorder point and order up to points in all the MAPE scenarios, based on the real life situation. Hence there is a very small band for the average inventory move in. The core observation however is to do with the deployment schedule. The fact comes out clearly that a daily deployment maintains much higher inventory at the DC comparative to other deployment schedules. This inventory position is caused by more frequent review and hence much quicker identification of falling inventory. The identification allows the deployer to make dispatch which increases the inventory again to the order-up-to level. On the other hand, in weekly deployment, there is possibility for inventory to fall much below the reorder level before being identified. This creates possibility for minimum inventory to go lower than order point whereas the maximum inventory level is still same as in daily deployment, which causes average inventory to be lower in weekly deployment. Hence the decision of deployment schedule has to be seen along with 34 management's choice of service level and how flexible the company is with adherence to the service level. 5.2. Unmanaged Scenario Analysis without Biased Forecasts The parameters used in this simulation run are shown in Table 7. Table 7: Configurations for 2 "dscenario analysis W0eMy, 35% Bi-Weekly 35% Gaily 4-5%/ 15 20 15 0 IS Us0 0 5.2.1. Impact on Shipment Variability We shall now check robustness of our findings stated in section 5.1. We remove variability from two exogenous input parameters and then observe the output variable i.e. transportation between the plant and the DC. We remove bias from the demand forecast and also assume that the production schedule versus actual also does not have any variation. The only exogenous variable in the model now is downstream demand from the DC and it is kept at realistic 35% MAPE. The output is shown in Figure 14. 35 45 40 35 30 -Sum of Truckload13 25 --\ Sum of Truckload14 of Truckload15 '520 -Sum 15 10 5 0 1 3 5 7 9 11 13 15 17 1921232527 2931 3335 37 3941434547 49 51 Weeks Figure 14: Weekly truckloads under scenario 13, 14, & 15 The output reinforces the earlier observation that bi-weekly deployment scheduling creates least variability, compared to weekly and daily scheduling, for transportation requirement at the plant. The standard deviation is consistently lower for all the scenarios for the bi-weekly schedule while it is highest for the daily deployment schedule. 5.2.2. Impact on Inventory Holding The average inventory held at the DC also shows the same trend as in section 5.1. The inventory at DC is higher in case of daily deployment review for the reasons mentioned earlier, specifically the opportunity to catch drop in inventory without delay. This is shown in Figure 15, where the green line indicates the daily deployment schedule. 36 300 250 200 150 Average of inventory 13 Average of inventory 14 Average of Inventory 15 100 50 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Weeks Figure 15: Average inventory at DC for 3 deployment schedules without Bias and production variation It is important to note that the simulation runs associated with any of the three deployment schedules does not show stock-outs. Considering this in addition to the inventory levels in bi-weekly and weekly deployment schedule shown by blue and red lines, it can be noticed how a bi-weekly schedule avoids stock outs while carrying lower inventory levels. This gets even clearer by Figure 16 which compares the inventory in number of days for the three deployment schedules. It should be noted that the parameters fixed for inventory management were order-up-to level of 20 days and re-order level of 15 days. From a perspective of stock service level, bi-weekly schedule assures that the inventory never goes below the desired 15 days. The bi-weekly schedule also performs better from an inventory holding cost perspective since it holds much lower inventory on an average compared to daily schedule. 37 25 20 -Average of Inventorydays 13 Average of Inventorydays 14 15 Average of Inventorydays 15 6 z 10 5 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Weeks Figure 16: Average inventory in no. of days at DC for different schedules if we consider the outputs both in terms of transportation variability and inventory at DC, the biweekly schedule of deployment provides the best approach with lower inventory holding. 38 5.3. Managed Truckload Simulations The parameters used in the managed truckload simulations are shown in Table 8. Table 8: Managed scenario forecast sensitivity parameters 1 0% 10 1 20% 15 1 35% 17 12 35% 17 12 35% 15 35% 15% 20 15 15% 15 10 15% 15 10 0% 01 20% Scenarios 1 through 4 analyse the inventory holding costs and lost sales as a result of changes in the inventory target and reorder points at the DC. In scenarios 3 and 4, the Over Projection variable is set to 20% because historically, the sponsor company is generally able to acquire 20% additional truckloads above what is contracted in a given week from the carrier. The effect of this flexibility is measured as the DC inventory variables are modified. Figures 17 and 18 depict the weekly truckloads of the scenarios 1 and 2 to illustrate the reduction in volatility in the managed scenarios. 39 Scenario 1 Truckload Comparison 50 45 40 35 0 30 25 - Forecasted - Scenario 20 - Forecasted 15 - Scenario 2 20 15 1 10 5 0 1 4 7 1013161922 25 28 3134 37 4043 464952 Week Figure 17: Managed truckload scenario 1 Scenario 2 Truckload Comparison 45 40 35 V 30 0 25 10 5 0 1 4 7 10 13 16 19 22 25 28 31 34 3/ 40 43 46 49 52 Week Figure 18: Managed truckload scenario 2 The above figures show how in the managed scenarios, the truckloads do not exceed the forecasted amounts. There are also only a few cases where the truckloads are lower than what was forecasted. Overall, the truckloads do not deviate much from the weekly limits and the freight volatility is drastically reduced. Table 9 compares the inventory holding costs and lost sales of scenarios 1 through 4, and the baseline unmanaged scenario. 40 Table 9: Managed scenarios 1 through 4 financials Lost Sales (Pallets) Scenario Total Holding Costs Transportation Costs Total Cost 0.0 731,831.12 $2,481,318.17 1,749,487.05 $ Baseline $ 0.0 707,331.74 $1,872,431.64 1,165,099.91 $ $ 1 28.4 710,227.67 $1,556,361.51 846,133.84 $ $ 2 0.0 709,728.84 $1,697,347.37 $ 987,618.53 $ 3 186.5 714,929.74 $1,518,171.62 803,241.88 $ $ 4 From Table 9 it can be seen that managing the truckload shipments can have a significant effect on the overall costs of the DC operations. The savings primarily come from the reduced holding costs, because the DC is able to serve customer demand effectively with a lower amount of inventory on hand. Though there were lost sales for scenarios 2 and 4, the amounts are negligible considering that the total demand from the DC is over 60,000 pallets of products. While the transportation costs are similar for the scenarios, in the real-world the baseline transportation costs are likely higher. The model is limited in that it analyzes the truckloads on a weekly basis. It is therefore unable to capture the daily variability and the impact that has on how frequently the more expensive spot market trucks need to be acquired. Scenarios 5 through 8 analyze whether further costs can be reduced by adjusting the parameters related to forecast accuracy. The purpose of this analysis is to determine how much efficiency can be gained if the sponsor company is able to improve its forecasting methods. The results are shown in Table 10. Table 10: Managed scenarios 5 through 8 financials Lost Sales (Pallets) Scenario Total Holding Costs Transportation Costs Total Cost 955.5 $1,109,286.67 730,849.49 $ 378,437.19 $ 5 0.0 720,459.76 $1,746,671.58 1,026,211.82 $ $ 6 201.0 $1,485,896.37 757,912.98 727,983.39 $ $ 7 126.2 732,299.13 $1,488,992.34 756,693.21 $ $ 8 From a costs perspective, improving the forecasting methods does not seem to have an appreciably significant impact. Also, improving the accuracy and removing the bias has an effect of 41 increasing the amount of lost sales. It appears that the main driver of overall costs is the inventory holding costs. By not sending replenishment shipments right away, the managed truckload policy keeps the inventory at the DC at lower levels throughout the simulation. Also, because the actual customer demand sometimes goes below forecasts, having a lag in the inventory replenishment allows the DC inventory levels to naturally adjust and maintain the target level. 42 6. Conclusion in the current business scenario the dispatches happen immediately when the reorder point is reached in DC inventory. Hence there is no forced smoothing of loads through a given week of month. Our model outputs show that the bi-weekly deployment scheduling gives the best results with respect to the truckload variability. The standard deviation is lower in this deployment versus the other deployment. The deployment schedule also provides an optimum result with respect to the inventory holding at the DC. The inventory held at the DC is significantly lower than the daily deployment without allowing the stock to go below the reorder point. This inventory difference is to the tune of 2.5 days at the DC. In comparison to the daily deployment schedule a lower supervision in bi-weekly schedule assures of the desired stock service level at a lower average stock level. Although a stock-out is an undesirable event that impacts a company and its brand value negatively, CPG industry still has a balance between excess and shortage costs if we compare that with either hi-tech, pharma or heavy equipment industries. For CPG industry, this result is especially applicable since the penalty of stock out is not very large (considering daily requirement low value products). In this setting the excess inventory holding would be extremely penalising due to low margins. The impact of demand forecast accuracy on the truckload variability was not significant. This is explained by the aggregation of random bobbing of actual sales numbers around the average. With respect to the presence of bias in the demand forecast, there was only a level change in the number of truckloads but there was no change in the variability of the truckloads required from week to week. As a recommendation for our sponsor company, we propose the following. To moderate the variability of the weekly truckloads, our analysis shows that actively managing and limiting the number of loads sent may be an effective alternative to the current operations. From the simulation results, it seems that responding right away to demand fluctuations is unnecessarily driving up the 43 costs of transportation and inventory holding. Delaying and limiting the replenishment shipments allows the inventory at DC to adjust naturally to demand fluctuations while maintaining service levels. Intuitively, it makes sense that the DC is able to withstand the demand fluctuations given the target levels of inventory. Should the sponsor company decide to pursue the managed truckload option, there appears to be an opportunity to simplify its transportation spending and reduce its overall logistics costs. 7. Future Work Before moving forward with a decision on how to moderate the transportation spending of the sponsor company, more analysis is required. The analysis done in this thesis only studied a portion of the goods shipped from just one plant to one DC. The study did not take into account the entire range of goods shipped in this freight lane, and the effect on demand and shipping that the other products might have on the freight variability. Also ignored are the effects of other plants and DCs within the supply chain network of the sponsor company, and the impact that production switch over costs may have on the inventory replenishment decisions. In the real world, the deployers are reviewing inventory for their product groups across the entire network. Therefore, in addition to meeting inventory targets, the deployers are also seeking to balance the inventory across the system. This means that there are multiple plants and DC involved within the replenishment process for any given product. The inventory balancing of the system also includes shipping goods from a plant based on capacity constraints at the plant warehouse, which was not considered in the simulation model. The logical next steps would be to build a more comprehensive model that takes into account all the relevant entities in the supply chain as well as incorporate the real world constraints and incentives of the stakeholders. 44 References Carey, Nick. (2014). Expanding U.S. economy exposes rising truck driver shortage. Reuters. Cassidy, William B. (2014). Driver shortage will limit truck capacity, growth, ATA economist says. Journal of Carriers. Cetinkaya, S., & Lee, C.-Y. (2000). Stock Replenishment and Shipment Scheduling for VendorManaged Inventory Systems. Management Science, 46(2), 217. Cummings, C. R., I1. (2014). Improving the inbound supply chain through dynamic pickup windows. Massachusetts Institute of Technology Thesis. Fugate, Brian S., Davis-Sramek, Beth, & Goldsby, Thomas J. (2009) Operational collaboration between shippers and carriers in the transportation industry. The International Journal of Logistics Management, Vol. 20 Iss: 3. Kim, Y. J. (2013). Analysis of truckload prices and rejection rates. Massachusetts Institute of Technology Thesis. Lee, Hau et al. (2004). Information distortion in a supply chain: The bullwhip effect. Management Science, 50. Mentzer John T. (2000). Supply Chain Management. New York: SAGE Publications. Rivera, Edward. (2014). Long distance freight trucking in the US. IBISWorld Industry Report 48412. Roberts, Jack. (2014). The coming crunch. Commercial Carrier Journal. USDOT. (2011). Freight facts and figures 2011. US Department of Transportation. 45 Wen, Y.-H. (2011). Shipment forecasting for supply chain collaborative transportation management using grey models with grey numbers. Transportation Planning and Technology, 34(6), 605-624. Zsidisin, George A., Voss, M. Douglas, & Scholsser, Matt (2007). Shipper-Carrier Relationships and Their Effect on Carrier Performance. Transportation Journal. 46 . - - = - - - - -- - -- I -- off- - -ML- - Appendix A Weekly forecast error of all 15 products. Week Actual Demand (Pallets) Forecasted Demand (Pallets) 1637.15 1 1284.78 1637.15 896.36 2 1637.15 1853.88 3 1637.15 4 1788.64 1338.68 5 1197.58 1134.20 133868 6 1212.35 1338.68 7 1613.08 1338.68 8 1011.16 1338.68 9 1190.08 1583.97 10 1069.85 1583.97 11 1249.23 1583.97 12 1269.12 1583.97 13 14 1174.49 1516.93 1095.50 1516.93 15 1516.93 1453.85 16 1516.93 17 1514.24 1377.01 18 1302.91 1794.02 1377.01 19 1252.13 1377.01 20 1142.54 1377.01 21 969.65 1377.01 22 1482.90 1471.72 23 1482.90 1332.51 24 1482.90 1255.87 25 1482.90 1077.53 26 1476.53 1101.87 27 1476.53 28 1375.48 1039.11 1476.53 29 1476.53 1237.18 30 1349.38 31 1138.53 1349.38 1250.28 32 1349.38 1169.75 33 1349.38 34 1051.92 1349.38 1211.52 35 1574.83 1082.85 36 1574.83 1264.87 37 1574.83 1224.88 38 1574.83 1111.16 39 1581.32 40 1362.60 1362.60 1571.49 41 1331.10 1362.60 42 1362.60 1179.16 43 1040.44 1327.33 44 1040.44 1442.95 45 1040.44 1650.58 46 1040.44 1272.19 47 1040.44 1200.25 48 1491.09 1172.84 49 1491.09 1269.27 50 1491.09 1285.27 51 1491.09 1441.66 52 APE 27% 83% 12% 8% 12% 18% 10% 17% 32% 33% 48% 27% 25% 29% 38% 4% 0% 6% 23% 10% 21% 42% 1% 11% 18% 38% 34% 7% 42% 19% 19% 8% 15% 28% 11% 45% 25% 29% 42% 14% 13% 2% 16% 22% 28% 37% 18% 13% 27% 17% 16% 3% 47 I - -Aw- -4 - - -,tw