Forecasting Product Demand for a Retail Chain to Reduce Cost of Understocking and Overstocking

Project Details




Konpal Agrawal, Prakash Sarangi, Rahul Anand, Raj Mukul Dave, Ramchander G, William D’Souza





Bad inventory planning can have a negative impact leading to loss of sales at the retailer end (understocking) or an inventory build-up across the chain (overstocking). This can result in potential losses to all stakeholders in the chain. Forecasting can help demand planners of retail chains make better decisions regarding the right quantity of products to stock on retail shelves.

Our forecasting goal is to manage inventory better by forecasting demand and determining the right amount of product to stock. We will also look to forecast how the demand varies based on seasonality. We studied store-level sales data for a particular SKU, to identify trends across stores with high, medium or low growth rates for the SKU. The stores were chosen to include those samples with the highest, intermediate and least annual sales. The data was taken from Kaggle. The data is relatively clean and covers daily sales for a period of 5 years and has seasonality. The methodologies used were Naïve Forecasts, Holt-Winters Smoothing (Additive/Multiplicative) and Linear Regression (Additive/Multiplicative).

Holt Winter’s exponential smoothing method provided the least MAPE value and was chosen as the model for forecasting demand for the next quarter. Holt Winter’s smoothing is used when trend and seasonality is present in the data. The appropriate Level(α), Trend(β) and Seasonality(γ) values were arrived at for each store either iteratively or by choosing the optimize function in XLMiner.

Conclusion & Recommendations: By evaluating different models and their error measures using the 2 years of training data and 1 year of validation, the Holt Winter’s model is used to forecast demand for the next 3 months. The demand from the stores for the next 3 months can be used by demand planners within the company to better manage inventory and ensure no stockouts and overstocking. The recommendations for various stakeholders within the retail ecosystem are as follows:

  1. Retailers can incentivize sales on weekdays to make demand smooth and predictable. Additionally, retailers with low sale volumes can rent out temporary storage facilities to replenish shelf space when needed.
  2. Warehouses can consolidate shipments by retail outlets to reduce shipping cost/ vehicle and in case of highly variable demand, increase order frequency.
  3. Production and manpower planning can be done with the demand forecast, shifts can be increased during peaks and SKU batch runs can be scheduled on production lines based on demand forecasts.

Application Area: