Distribution networks are the conduits that connect companies with their customers, so it is hardly surprising that the way these networks are designed has a critical impact on cost and customer service.
Companies commonly use mathematical optimization models to arrive at the best network design, but this approach is flawed in one key respect — it does not take into account changing market conditions during the several years it can take to complete a design project. This is particularly onerous in developing economies where markets tend to be extremely changeable.
Research carried out at the Malaysia Institute for Supply Chain Innovation (MISI) shows that supplementing mathematical models with analyses of external variables enables companies to develop the most efficient distribution networks. The research work was completed in collaboration with a leading Asian chemical manufacturer as part of a thesis project for the MIT-Malaysia Master of Science in Supply Chain Management.
Distribution network designs specify the locations of warehouses and how much product is allocated to each facility. A chemical company typically manufactures product in large plants to lower production costs by exploiting economies of scale. Product is shipped to numerous customer locations. The design of its distribution network, therefore, determines the total cost of delivering products to meet customer demand while maintaining the appropriate service levels.
There are many ways to configure a network to meet these goals. For example, a company can reduce its inventory holding cost by risk-pooling the inventory in a few warehouses. However, this option incurs higher transportation costs and longer lead times. Alternatively, a company could become extremely responsive to demand by stocking inventory in many warehouses. But such a strategy requires higher inventory volumes and hence higher carrying costs.
Mathematical models can be used to find the optimal solution, but this might not reflect real-world demands. External factors such as regional product demand, commercial real estate prices, and transportation costs can change markedly over the three- to five-year planning horizon that is common for these design projects.
The MISI research project tackled this problem in four steps.
First, an optimization model was designed to minimize total costs including the costs associated with transporting product, opening and closing warehouses in different locations, fixed warehouse operations, and maintaining inventory. The key consideration was deciding how many warehouses the company should support, and which time periods the facilities should operate within. The model also complied with various constraints such as the need to meet minimum safety stock levels.
Second, the researchers developed an exhaustive list of uncertainties that have a critical impact on the efficiency of distribution networks. Four business and macroeconomic factors were particularly relevant for the manufacturer’s operations: demand growth, oil price fluctuations, shifts in industrial real estate prices, and interest rate changes.
Third, we calculated a plausible range of values for each of these four macroeconomic factors over the planning horizon. The ranges were derived from an extensive search of industry forecasts and reports as well as expert opinion. Using these values, we ran the optimization model to create multiple scenarios based on market conditions driven by the macroeconomic factors. And we identified the optimal network design for each scenario using the mathematical optimization model. For each network design, the cost difference between each given scenario and its optimal version was calculated (known as the “regret”).
Finally, a comparison of the optimal designs — which specify which distribution centers should be operated in each period over the planning horizon — for the different scenarios suggested that one variable had the most impact on performance: the price of oil. A more detailed analysis of oil price effects was carried out.
When deciding which network design to adopt, the chemical company should look at the ones that minimize the regret for the different future scenarios.
This approach helps companies to design distribution networks that are aligned with real-world market conditions in two important ways:
- It enhances quantitative mathematical models by considering a broad range of qualitative variables, employing techniques borrowed from scenario planning. The methodology provides a clearer picture of how a distribution network design might perform. Companies can focus on the key major environmental factors affecting the robustness of a network configuration, under various quantitative scenarios.
- By using this approach, it is also possible to get a sense of which distribution scenarios are the most relevant, given the market changes that affect the way a network performs. It is possible to represent which of the scenarios considered are most likely to occur — a valuable insight for managers who are striving to develop the most efficient channels for distributing product to customers.
MIT Sloan Management Review