Sales Forecasting with Deep Learning

Challenge

In the retail environment, sales forecasting is one of the most well-known data science use cases for the application of statistics and machine learning. Our client, an international retail corporation, faced the challenge of better planning the procurement of products in a specific merchandise category. The general goal was to test the applicability of forecasting models. In the past, there were frequent out-of-stock situations, negatively affecting sales and customer satisfaction.

Approach

To forecast the required quantities, checkout data at the receipt level was aggregated on a daily and weekly basis. In addition to product, stock, and promotion data, various external influencing factors such as holidays, seasonal effects, and aggregated sales information were modeled and incorporated into complex deep learning models for forecasting. These models were meticulously optimized to maximize forecasting accuracy and then implemented into the client's IT infrastructure. Here, the models can now generate automatic, daily forecasts for the products in question. These forecasts are automated and archived to ensure traceability of the results over time.

Result

Based on the daily generated, product-specific forecasts, our client can now optimize the procurement and logistics of the products in question. With a forecast horizon of three months, both short-term and medium-term actions can be planned and executed based on this information.