The consumer products sector faces volatility in demand on a high scale and level of complexity, thereby posing challenges in the area of inventory management. Economic volatility and demand variability present challenges that simple models of demand forecasts are not equipped to handle. An important method of tacking demand variability is an effective way to improve the inventory control policy, which should be designed to smoothen stocking response to demand variation arising from the customers. The problem gets compounded when we are dealing with perishable goods as shelf life is very small.
The business goal is to arrive at an inventory planning policy for two such perishable classes of goods: dahi/ yogurt and fresh milk. The inventory policy will attempt to balance the costs of under-stocking vs. the cost of over-stocking these goods.
The demand data when plotted showed a linear trend with additive seasonality and some noise. There is one cycle per month with each month having 4-5 weeks. We see some peaks in Aug 2011 and Dec 2011 which is due to high demand during festivals like Janamashtmi.
This inventory policy will be used by store managers to :
1. Determine the near optimal order quantity for different seasons, days of the week etc
2. Determine the reorder point at which the order should be placed
This will be achieved by forecasting the demand for two classes of products: dahi/yogurt and milk. The lead time for the supplier is two weeks and therefore we are forecasting for two weeks! As and when actual data is available for the next week, we roll forward our forecast to include the last week data and forecast for future two weeks.
We used a metric called 'Mean Revenue Impact' which measures the average impact on revenue taking into consideration costs of under and over-forecasting. We found that the method of using Multiple Linear Regression with Auto-Regression on residuals to be superior for forecasting the demand for fresh milk; and Naive forecast to be more accurate for forecasting the demand for dahi/yogurt. Lag-1 Naive forecast was considered as our benchmark.