Inventory Management in the Era of Big Data

The explosion in the availability and accessibility of machine-readable data is creating new opportunities for better decision making in applications of operations management. The swell of data and advances in machine learning have enabled applications that predict, for example, consumer demand for video games based on online web-search queries (Choi and Varian 2012) or box-office ticket demand based on Twitter chatter (Asur and Huberman 2010). In the context of inventory management, demand is the key uncertainty affecting decisions and such works suggest a potential opportunity to leverage large-scale web data to improve inventory decisions, for example, for stocking video game titles or allocating cinemas of varying capacities. There are also many other applications of machine learning, including Da et al. (2011), Goel et al. (2010), Gruhl et al. (2004, 2005), Kallus (2014), that use large-scale and web-based data to generate predictions of quantities that may in fact be of interest in operations management applications. By and large, however, these applications and the machine learning techniques employed do not address optimal decision-making under uncertainty that is appropriate for operations management problems and, in particular, for inventory management. We study how these data, leveraged appropriately, can correctly and successfully inform inventory management decisions and provide a competitive edge. We focus on a particular case study of the distribution and manufacturing arm of a global media conglomerate (henceforth, the vendor), which, as a distributor of multi-media, is among the three largest in the world. The vendor, which shall remain unnamed, is a direct customer of Silkroute, a provider of analytics platforms for managing manufacturing, distribution, and retail operations. The vendor, which ships an average of 1 billion units in a year, as well as the media retail industry at large, is under increased pressure to improve operations and lower costs in the face of increasing digitalization, declining sales, and diminishing shelf space. The heightened importance and consequence of good inventory decisions provide an excellent case study of the use of large-scale data for achieving a competitive edge in a squeezed industry. We consider the vendor’s VMI (vendor-manage inventory) operations in selling over half-a-million entertainment titles on CD, DVD, and BluRay at major European retailers with over 20,000 locations. To inform VMI decisions, we leverage transactional records collected and organized by the Silkroute platform, data we harvested from public Internet sources including IMDb.com (International Movie Database) and RottenTomatoes.com, and search query volume data provided by Google Trends. To leverage these data, we employ recent data-driven optimization techniques developed by Bertsimas and Kallus (2014) that address the conditional stochastic optimization problem: