Machine learning based estimation of delivery windows: A suggestion of how to implicitly mitigate the last mile problem

Abstract

During the past 15 years there has been a rapid growth in internet users. Currently more than 2 billion people worldwide are connected to the internet, with half of these located in developing nations. More than 330 million of the remaining 1 billion are located in India and this number is expected to grow to 770 million by 2020 (Gnanasambandam et al. 2012; NASSCOM 2016) Ranade, & Rao, 2016). Alongside the growth of internet users there is also an expected growth of online purchases. Indeed, by 2020 it is expected that 200 million Indians will partake in online purchasing activities totalling USD 34 billion (ibid.). This rapid growth will naturally present online retailers with a wide array of challenges. One of the most prominent issues felt by online retailers not only in India but also worldwide is related to the so called Last Mile which can be defined as “the last link in the supply chain to the home” (Edwards, McKinnon & Cullinane, 2009, pg. 104). Traditionally companies have relied on transport of scale in order to reduce costs however with a wide range of products being delivered to a wide array of different addresses that is increasingly difficult to achieve. Therefore, several online retailers have researched how to optimize transport especially in the last mile. Several logistics providers have adopted various forms of machine learning in order to improve service levels. These include route optimization, stock management and other cost saving measures. However, little attention have been put on customer satisfaction and expectation matching. Through an instrumental case study approach, this paper has explored the different aspects, which practitioners have faced in relation to the last mile problem, and has sought to leverage the features/strengths inherent in different ways to apply machine learning based applications to solve one of the observed issues - long delivery windows.