Last mile delivery problem: Application in an e-commerce company

The coronavirus, COVID-19, which first started to be seen in Wuhan, China, in the last quarter of 2019, has led to a rapid development in e-commerce despite its many negative effects on world trade and the living conditions of the epidemic. With the new habits brought to us by the epidemic, people worldwide have started to meet most of their needs online. Based on demand, more studies and investments have begun in logistics, an indispensable element of the increasing e-commerce volume. With the studies on last-mile delivery, which constitutes the largest part of logistics costs, various logistics models have emerged to keep costs under control while responding to high demand with limited resources. This study was prepared upon the request of a company operating e-commerce in the furniture industry abroad. The company operates through its own internet sales site. When he examined the costs with the increasing order density recently, he found that most of them came from storage and logistics steps. The company carries out its activities through a local logistics company that provides storage and distribution services. The company requested to develop a storage and distribution model that could reduce its costs as an alternative to its current operation. Based on this request, four different scenarios were developed and a delivery strategy that minimized costs was determined. In Scenario 1, the current situation is analyzed and the e-commerce company works with a local logistics company and receives storage and distribution services from this company. In this scenario, there are no other costs such as employee costs or vehicle costs. Storage and distribution service pricing did not vary between the years forming the data. The storage price is a fixed amount per month. Distribution pricing varies depending on the weight of the product and delivery address. The company whose data is analyzed manages the last step delivery by receiving storage and distribution services from the logistics company, as in this scenario. In Scenario 2, it is planned that the company will manage the last step delivery process under its own operation. In order to manage this process, it must first determine a warehouse location. Using the address data from the past orders in the data set, the p-median algorithm was developed and the warehouse location was positioned by taking into account the results obtained from the algorithm. Based on current comparable rental fees, cost calculations were made considering that warehouse space would be rented from the area above the address found with p-median. In order to provide distribution service as well as storage service, the need to rent vehicles and employ personnel to make shipments from the warehouse location determined by p-median has emerged. To meet this need, vehicle rental costs and personnel employment costs were added to the total scenario cost. A routing study was carried out to the addresses in the data set using the local search-supported Clarke and Wright algorithm, taking into account vehicle and time capacity limits, so that weekly deliveries would be made to the customers through the designated warehouse. The total routing cost was reached by taking into account the oil prices of the period in the route length data set. The total cost of the scenario was reached by adding up the warehouse cost, vehicle rental cost, personnel cost and routing cost. In scenario 3, unlike scenario 2, the relationship between regions and order densities was examined. In the examination, it was seen that Pareto analysis was applicable with the data set. Pareto analysis was applied and as a result, it was seen that 79.6% of the order density came from the 1st and 2nd regions. Thereupon, the new warehouse location was determined with the help of the p-median algorithm, using the address data in the two regions that constitute the order density. Equivalent fee values were found for the specified warehouse location and added to the storage costs. Personnel and vehicle costs that come with carrying out warehouse activities are also added to the scenario cost. Routes were created using the local search-supported Clarke and Wright algorithm to deliver the deliveries from the new warehouse location to the addresses in the two determined regions, and the cost of the total distance was combined with oil prices and added to the scenario cost. It was planned to receive cargo service for deliveries to regions other than the two densely populated regions, and the cargo cost for these regions was added to the total cost of the scenario. In Scenario 4, unlike the other scenarios, a participatory delivery model was studied. For this model, unlike Scenario 3, the deliveries of the two regions that constitute the order density are planned to be made through the most appropriate delivery points to be determined. In addition to the warehouse data and costs in scenario 3, customer address data for scenario 4 was clustered and delivery was planned from 5 delivery points. Routing was carried out from the distribution warehouse address determined with P-median in Scenario 3 to the 5 delivery points determined in Scenario 4 using the local search supported Clarke and Wright algorithm. Distribution costs were calculated by taking into account the calculated route length and the oil prices of the data set period. In order to encourage the participatory delivery method in Scenario 4, gift vouchers were defined for customers and the costs of the defined gifts were added to the total scenario costs. When the costs of the four different scenarios studied were examined, it was seen that the cost of scenario 3 was lower than the other scenarios. What makes Scenario 3 advantageous over other scenarios is the use of more than one model in solving this problem, which is classified as NP-Hard. With the help of Pareto analysis used in this solution, it was seen that the orders that constitute the majority of the cost came from regions close to the center, and the distribution warehouse was found using the data of the addresses specific to this region in order to shorten the transportation distance. Thus, although the transportation distance and therefore the transportation cost were less than scenario 4, scenario 3 was preferable when looking at the total cost table. The point that increases the total cost of Scenario 4 is the cost of convincing the customer who comes with the participatory delivery model to participate. The gift voucher reward, which makes the delivery of participatory models preferable for the customer, prevented the costs of scenario 4 from decreasing. Although the last step delivery cost was lower, the extra cost item brought by the participatory model prevented this scenario from being preferred. This study shows that when the hybrid model, which is created by using different methods together, is preferred, the costs are approximately 44% lower than the current situation, scenario 1. With the implementation of the model in Scenario 3, extra costs such as personnel employment, vehicle and warehouse rental may occur, as well as additional problems that these transactions may bring. In the study, the problems and solution costs that these new additions will bring are excluded from the scope. The scenarios that emerged as a result of the study, detailed reviews of all scenarios and cost analyzes were shared with the e-commerce company and its officials, and the decision was left to the managers of the e-commerce company.