The forecasting initiative, designed for perishable goods distribution across various stores, seeks to enhance inventory management efficiency and minimize excess inventory by predicting weekly sales quantities. Our stakeholders are the inventory managers of suppliers, allocate inventory to stores, and surplus items result in handling costs. Given the perishable nature of the goods, failure to predict sales quantities can lead to excess inventory and additional costs.
We acquired data from Nuqleous, a company specializing in precision retail planning, following their recommendation to aggregate daily data into weekly data for a four-week forecasting window for perishable goods. Focusing on the SKU with the highest sales quantity, 1394919, we utilize a roll-forward forecasting approach which can promote accuracy by updating the recent data. External variables like average price are also incorporated to improve accuracy. Using Building Number 514 as an example, we showed the time plots and boxplots which identify the ETS method as the best model, considering outlier reduction, overfitting prevention, and minimizing residuals.
Continuous updates and real-time data refinement are crucial for model accuracy. The forecasting
models, designed for a four-week horizon, offer a proactive strategy for suppliers to optimize
inventory allocation, mitigate excess inventory costs, and reduce perishable item wastage.
There are still limitations involving the need for regular data updates to enhance predictive precision.
Despite challenges, the forecasting models present a promising step towards efficient inventory
management and sustainability in the supply chain.