OPTIMIZING TRANSPORTATION MODES IN A DISTRIBUTION NETWORK OF PHARMACEUTICAL GOODS Hendrik Van Landeghem Ghent University, Industrial Management Technologiepark 903, 9052 Gent, Belgium [email protected] Abstract: The paper describes the results of a study to optimize the transportation modes in a distribution network. The company involved is a producer of pharmaceutical consumables, and its European distribution network is the focus of this paper. First we describe the data collection and validation phase. Next a MILP model is formulated, that optimizes the allocation of goods to markets through selected Lane Modes of transportation. Several scenarios are studied. We show that by optimizing the selection of Lane Modes we obtain up to 36% of cost reduction. Finally the results are commented upon. Keywords: Transportation systems, Operational Logistics, Distribution Planning

1

The pharmaceutical distribution network

The research was performed for a large company, producing pharmaceutical consumables and instruments, as part of a Master’s thesis (De Smet and De Vuyst 2005). The focus was on the European distribution network, served from one central European DC (EDC) in Belgium. The network handles some 79 different product types, sending them to about all countries in Europe. Total transportation volume amounts to about 377.900 m3 per year, for a cost of roughly 21 million €. We aggregated the customers into 739 zones (see table I) in order to obtain a fine mesh of demand, that would model both transportation destinations and distances faithfully. This was necessary because the optimization model had to be allowed to Figure 1. Distribution network with major clients group different product flows into shipments, and then select the optimal transportation mode for this flow.

Country Austria Belgium Germany Danmark Spain Finland France Great-Britain

Country Greece Ireland Italy Luxemburg Netherlands Norway Portugal Sweden

# zones 11 9 86 9 52 99 97 174

# zones 1 2 92 1 10 10 3 83

Tabel I. Number of customer zones per country in the study

The research questions we studied were: 1. What is the optimal mode of transportation for inbound and outbound flows? 2. How will this evolve under the growing business for the next 5 years? 3. What will be the effect of adding a second echelon of local DC’s, specifically for the German region.

2

Obtaining realistic transportation costs

One of the hardest parts of any study is data collection. In logistic models the unit costs for transportation and storage are very important, since optimal results are very sensitive to variations in these costs. In practice, the cost of transporting one m3 over 1 km is very hard to obtain. The carrier landscape is of a staggering complexity, with an abundance of potential suppliers, and a bewildering array of different fee structures for transportation. Obtaining transportation costs for this study was split in two parts: unit costs for the current transportation links, and costs for links that would appear in the above scenario’s, but were not used today (hence no costs were available inside the company). 120.00%

40

35 100.00% 30 80.00% Frequency

25

20

60.00%

15 40.00% 10 20.00% 5

0

0.00% 0.00

12.28

24.56

36.84

49.11

61.39

Figure 2. Actual shipment volumes at FTL cost

70.00

2.1

Inbound transportation

The inbound flow to the EDC originated from a total of 25 plants (or aggregated suppliers). Its main modes were air, sea and road, consisting mainly of containers. Analysis of the size of shipments revealed a global utilization rate of 77% for truck shipments, indicating a first possibility to reduce costs by increasing utilization. Cost figures were fairly easy to obtain from the company’s books. 2.2

Outbound transportation

Due to the sheer number of outbound shipments, we first analyzed the data in bulk. In figure 2 we show a histogram of Full Truckload (FTL) shipment volumes for one particular link, yielding an average truck load of 36,2 m3 or 52% utilization. The FTL cost of this (existing) link was about 1050 euros, regardless of utilization. Reducing transportation costs will therefore mean increasing utilization of FTL shipments. However, this is clearly a non-linear cost, so in order to keep modeling simple (given the large problem size) we introduced a simplified set of transportation choices, called Lane Modes. They are listed in the left side of figure 3. The total cost of a FTL was thus allocated to the actual load, yielding a unit cost of 15€/m3 for FTL_100, increasing to 20€ for FTL_75 and 30€ for FTL_50. Using this model we then defined a total of 22 outbound Lane Modes (figure 3, right side), and 3 Inbound, based on the above 5 classes, but augmented with different frequencies of the FTL mode (every day, every 2 days, every 5 days, every 15 days, …). This approach is an extension of (Zäpfel and Wasner 2002), who use only 2 truck types. To determine a correct cost per km for the different modes on new links, we investigated the costs of multiple carriers (both current and prospective ones), and using linear regression, obtained a fairly reliable result as shown for the “Pallet” Lane mode in figure 4.

Full Truck Load

FTL_100 = 70m³

FTL_75 = 52,5m³

FTL_50 = 35m³

Pallet

Small package

10m³ < x