Bayesian demand forecasting in big data: Application to the perishable goods in India

Abstract

The value of fruits, vegetables and grains wasted in India has recently been valued at Rs 44,000 crore annually (see Economic Times, November 28, 2013). Fruits and vegetables account for the largest portion, Rs 13,300 crore - accounting for 18 percent of India's fruit and vegetable production. Two of the biggest contributors to food losses are the lack of refrigerated transport and the lack of high quality cold storage facilities for food manufacturers and food sellers. While largely an issue of infrastructure, one can think of ways of mitigating wastage at the retail level. Doing so requires a careful management of inventory as well as a forecasting method that can be used to manage that inventory. In addition, how retailers manage promotions in these categories will also affect the amount that is left at the end of the day that becomes unusable. Our objective in this paper is to attempt making some progress on the pernicious problem of wastage at the retail level. We use data from a large retailer of fruits and vegetables with over 300 stores in India to first understand the extent of the problem at the retailer. Managing inventory of perishable items with only a few days of shelf life, is especially challenging. Ordering too many or too few of these items directly impacts sales, profits and increases wastage. Product proliferation and high variability in daily sales makes forecasting sales difficult, leading to ad hoc and gut-feel inventory ordering.