Strategically simulating proper distribution
By Homero H. Contreras, Pablo Nuño, Eric Porras and Eduardo Zelaya
Figuring out how to get products or services to their endpoints is at the heart of everything for a business that wants to continue operating. While the owner of a small, retail store seems to have it easy – the customers come in, pick up their wares, pay and leave – things become more complex as organizations grow larger. But integrating simulation into supply chain management issues can simplify the decision making process, particularly in developing countries where cost strategies often are the best option.
Companies face a complex business environment in the era of globalization, but these factors mean that there always are opportunities to implement new strategies and create value for clients and suppliers. Supply chain management can play a vital role in these opportunities because it encompasses many parts and subsystems, such as inventory, the distribution network and manufacturing, and provides the way to link suppliers with their final clients.
In his seminal “What is Strategy?” paper for Harvard Business Review, Michael Porter considered strategy as a way to achieve a unique and valuable position by making a trade-off in activities and organizing the entire company around that position. He also suggested that there are two generic strategies, cost leadership or differentiation. Differentiation might be the best way for companies to provide a high value to its suppliers and customers; however, there are situations where cost leadership must be prioritized. Developing countries are facing a constant market that is sensitive to price, so in places like Mexico, the cost strategy usually is the proper one to pursue.
Strategy and simulation in distribution networks
A distribution network is a complex system that embraces a lot of operations linked by mathematical calculations and stochastic behavior; in particular, the integration of all the activities within a distribution network provides opportunities to create value, reduce costs, raise productivity and maximize profit models. It is also a dynamic system that continually evolves as long as the business is working to serve its customers.
Some decisions must be taken within the operation of the network to focus on long-term, midterm and short-term time frameworks. Managing all the issues included in the network is a great challenge. Industrial engineering tools have been developed to aid in managing the distribution network, and there are two main approaches to analyze it: analytical models and simulation models. Analytical models are based on a set of equations that define the model and are solved by a closed form or an algorithm. On the other side, simulation models are dynamic models that resemble the behavior of a system and include the system’s main relationships in the model.
Simulation is considered a better tool because it provides a way to integrate the stochastic conditions faced by the distribution network along with its dynamic behavior. Any distribution network includes strategic, tactical and operational issues.
It is important to note that operational models deal with the day-to-day operations of the network and, therefore, are more detailed, require a complex structure and are more suitable to low-level management positions.
On the other hand, strategic models are focused on decisions that are relevant for the company in the long term. They look for value creation through specific investments while considering the reduction of cost. These models are suitable for helping make decisions that affect the competitive position of the company, and they are used by top executives.
So a strategic model must be designed with top executives in mind while fitting in with the strategy of the company. A distribution network’s model must provide information to configure the framework of the network without focusing on operational issues.
The business case
Health services are a key issue in the strategy of any country, and production and distribution of medicines face strong competition in the marketplace, especially in the fight between generic drugs and patented drugs.
Some people in Mexico are uninsured, and the public health system has faced problems with the supply of medicines, thus forcing these uninsured people to buy them at their own expense. On the other hand, private drugstore companies are facing a great amount of competition, especially among the three or four largest drugstore companies in the country, because most of the products they sell are patented, or name brand, drugs. These products cost a whole lot more.
But because the Mexican government has authorized a considerable number of generic drugs, some pharmaceutical companies focus on these kinds of products. The lower costs of these drugs help draw a lot of clients, so some drugstore companies are offering generic products under their own trademark as a way to attract and retain customers.
One Mexican corporation that owns a drugstore company with operations across the entire nation (31 states plus a federal district) decided to embark upon a complete redesign of its distribution network. The changes were to focus on the network’s strategic configuration. The tactical and operational issues were not considered simultaneously, and members of top management, including the CEO and the vice president of logistics, were interested in a way to configure and evaluate the network’s possible configurations based on the minimum cost.
The actual distribution network was based on the following structure:
- One master distribution center
- Nine regional distribution centers (RDCs)
- 34 local warehouses
- 2,000 drugstores (some owned by the company; others were franchisees)
About 20 percent of the products are manufactured at a facility the corporation owns that is located next to the master distribution center. From there, the products are sent to the RDCs. Regional centers also receive products from external suppliers. In these instances, the distribution cost from suppliers is included in the product’s price.
RDCs send products to all drugstores that the corporation owns, as well as to big franchisees and local warehouses located within an RDC’s associated geographical area. Small franchises are not served directly by the corporation. They must buy and pick up their products from local warehouses; therefore, there is no distribution cost associated with them. RDCs, warehouses and stores request products based on their inventory levels. The safety stock is defined heuristically as 30 days. Distribution is carried out once per week, and a strategic aggregation is based on regions served by each RDC, as shown in Figure 1.
All transportation is performed by an external company that uses different kinds of trucks, except for some specific places, mainly around coastal areas, that are served by ferries. The system includes designed truck routes. Delivery routes span some states and serve a different number of points of sale; therefore it was a good idea to design a specific methodology to assign a unitary transportation cost. Each RDC operates independently from the others with no overlap in operations or transportation between RDCs.
The company was interested in evaluating the suitable number of RDCs that should be included in the system, along with the regions served by each one. Options included creating new RDCS, closing some and merging others. If needed, a second master distribution center could be opened, and the system used to define the safety stock in the inventories through the complete network could be revamped as well.
Two key performance indicators were defined to evaluate the minimum cost desired:
- The total transportation cost of the network
- The total cost of holding inventory in the network
Data from one complete year was available. Some information that related to special demands, such as specific sales for the government or for other companies, were not included in the model because they weren’t common. They also weren’t significant, representing less than 2 percent of the total. Such orders are carried out under a make-to-sell approach. In order to provide support for the company’s decision making process, a simulation model was designed with a strategic approach.
The simulation model
A simulation model to evaluate the strategy to reconfigure the distribution network was developed under a standardized framework. According to Dayana Cope, Mohamed Sam Fayez, Mansooreh Mollaghasemi and Assem Kaylani in their 2007 Winter Simulation Conference paper “Supply Chain Simulation Modeling Made Easy: An Innovative Approach,” standardized simulation models are those that can be applied to a broad range of systems and, at the same time, be precise enough to be adjusted for different scenarios and performance criteria. Such models become specific when data for a particular system is loaded.
Authors like Guruprasad Pundoor and Jeffrey Hermann in “A Hierarchical Approach to Supply Chain Simulation Modeling Using the Supply Chain Operations Reference Model,” published in the International Journal of Simulation and Process, suggest that there are always some common processes within distribution networks that can be reused. Qing Wang’s “A Discrete Event Modeling Approach for Supply Chain Simulation” in the International Journal of Simulation Modeling suggested that simulation must be focused on the specific elements of the supply chain.
In this drugstore company’s case, the simulators based the distribution operations model on a two-echelon framework for the supply chain where there is a supplier serving “n” number of clients through a matrix of data. This structure will form the basic unit of the model where the complete logic and common processes are carried out, and its replication both forward and backward can create the complete network, just like a series of steps where a supplier becomes a client of another previous supplier, as shown in Figure 2. Furthermore, due to the fact that the distributive processes could yield some subnetworks, the structure also can be replicated in a parallel way, converging in a point-of-origin of the network.
The simulation model considered the code of the basic two-echelon structure, and this logic then can be replicated in a serial way, starting from the final end of the distribution network (the final consumers) and going back up to the main supplier, thus forming the complete network.
It must be remembered that the model is based on a strategic approach; however, some tactical issues must be included to evaluate the network based on key performance indicators. The main characteristics of the model are:
- Aggregation: An aggregate approach has been included, adding all the pieces and products to be demanded in a single aggregated demand without making a distinction among individual items. This approach provides an insight to the total items in inventory through the whole system; however, the model standardization allows an easy way to be modified to include families of items or individual products in a tactical manner.
- Unitary transportation cost: There are several delivery routes spanning several states, and even the types of vehicles vary. Trucks are the main delivery vehicles, but ferries serve some coastal regions. Therefore, a methodology to integrate the unitary transportation cost for each region has been developed and included as an input data to the model.
- Discrete operation: Because most events occurring during normal operation of the distribution network occur at a specific point in time, discrete operation is used. This approach leads to the fact that all variables are considered as discrete, allowing for a fast execution of the model. The model is based on a single entity flowing through the model and executing the common logic previously stated.
- Time framework: Considering the strategic approach and the aggregation, the model does not consider small gaps of time, so the entire cycle of operations are carried out in a discrete time scope, which can be one week, one month, etc.
- Simulation language: In order to achieve a standard model and its recursion both in serial and parallel, a typical graphical simulator was not used because of the restrictions for handling these kinds of models. A simulation language was used to prepare this logic and provide an encapsulated model, where all the information is included without using any external references to databases, spreadsheets or any other software.
Operating the model and getting results
First, the standardized simulation model was validated by simulating the actual distribution network against one year of historical data. Then, the validated model was used to evaluate different scenarios for the new configuration of the network.
Simulation was carried out based on a steady-state analysis, and the differences in the total inventory level were about 2.7 percent versus historical data. In the case of transportation costs, the difference of the simulation versus historical data was about 3.2 percent.
Considering these differences and a target error of 5 percent, results from the simulation model were within tolerances. Therefore, the researchers decided that they could use the model with confidence to evaluate new alternatives to inventory policies and other aspects of the distribution network. Some scenarios were designed to improve the distribution network, and several options were considered, including:
- Opening, closing and/or merging regional development centers
- Reassigning regions to different RDCs
- Increasing delivery frequencies
- Opening a second, new master distribution center All improvement scenarios also considered a new calculation of safety stocks using a multiechelon inventory system. More than 20 scenarios were designed and tested.
The model only required the recalculation of input data and the activation, via a simple logical switch, of the multiechelon system. No more changes to the model were required, and the scenarios were evaluated in a simpler and faster way.
Some alternatives were considered promising, but according to results from the simulation, these options provided no improvements to the total cost of operating the distribution system. Once all the alternatives were analyzed, the best distribution network was defined as one having eight RDCs and one master distribution center, shown in Figure 3. The new distribution system merged four of the original RDCs into two new ones and kept five of the original RDCs in the network. These five are to use the new safety stock levels. One new RDC was opened to absorb part of the regions previously served by other RDCs.
The entire new configuration is based on increasing the frequency of delivery to twice per week in the metro areas where RDCs are located, along with using multiechelon inventories. Increasing deliveries might increase costs while reducing inventories, so the trade-off suggested by Porter’s Harvard Business Review article was considered. However, by reassigning the various regions, the new distribution system will lead to reduced transportation costs.
The suggested distribution network would reduce the total system cost by about 20 percent, as shown in Figure 4.
Not too good to be true
Using a simulation model provided insights about the distribution network and allowed the integration of stochastic conditions occurring in the real operation of the network. Furthermore, the design of the model, based on the input-data matrix, boosted the design and evaluation of the scenarios. This was possible because some of the scenarios were similar, involving identical regions and/or regional development centers, while others just merged the data. Therefore, some information was reused.
The nature of the standardized model supported the fast execution of the analysis. Although any scenario had its own structure, the main logic was identical, and the only parameter required was the number of areas and/or RDCs to be considered. So depending upon the scenario to be run, the researchers just had to adjust the matrix input data.
The final savings of 19.3 percent of the total cost is important for the company. Once translated into U.S. dollars, this money can be used to reinforce the corporation’s competitive position through a small reduction in prices and expanded marketing to get more clients.
Finally, the risk associated with configuring a new network was reduced drastically. Corporate officers were worried about the suggested cuts in inventory level, fearing that the comparison versus the actual heuristic method was looking “too good to be true.” So the reductions were implemented gradually in one of the regional development centers. However, after it was demonstrated that the new way to calculate inventories was providing the same service level to its customers, the company deployed the new inventory levels to its complete network.
Homero H. Contreras has spent 15 years in the textile, automotive and retail industry. The business consultant holds a bachelor’s degree in industrial engineering and is a Ph.D. candidate at Universidad Popular Autónoma del Estado de Puebla (UPAEP University) in Mexico.
Pablo Nuño is a member of the Mexico System of Researchers and the director of graduate and international affairs at UPAEP University. He has more than 20 years of experience in academia. He holds a Ph.D. in industrial engineering from Oklahoma State University and has served as a business consultant for companies in the United States and Mexico.
Eric Porras is a full-time professor at Escuela de Graduados y Dirección de Empresas (EGADE) at Instituto Tecnológico y de Estudios Superiores de Monterrey (ITESM) in Mexico City (Santa Fe campus). He holds a Ph.D. in operations research from Erasmus University of Rotterdam (Netherlands), has more than 15 years of experience in consulting and academia, and has published papers in several international journals.
Eduardo Zelaya is a part-time professor at EGADE Business School at ITESM-Campus Santa Fe and has more than 20 years of academic experience. He holds a Ph.D. in management science from the University of Nottingham and Stirling in England. His experience in consulting has been focused on strategy and competitiveness, logistics and optimization.