Predicting Footwear Demand

Predicting Footwear Demand, 3-5 Seasons in Advance

Client’s Challenge

➢ Footwear and Apparel industry experience long time to market situation where designs are created almost 3 to 5 seasons in advance.
➢ Footwear demand forecasting is a very complex problem because demand must be predicted at Store-SKU-Size level.
➢ Given the highly granular level of predicted values, there is every chance of going wrong, which can lead to either dead stock or being stocked out
➢ Ever-changing fashion trend, which manifest in terms of variety in designs, color combination and materials.
➢ The impact of events such as sports tournaments, festivals etc. and weather changes further adds to complexity
➢ Inherent latency in production and supply chain issues
➢ The company’s spreadsheet based top-down forecasting approach often leading to wrong prediction

Analysis & Solution Approach

➢ For the purposes of the project, last 4 years of demand data was taken as inputs.
➢ Month wise events data (especially sporting events and festivals) and day wise weather data for the territories was taken from know government weather data sources.
➢ Data analysis done in detail and visualize it to look for patterns across product, location and time dimensions.
➢ Specific instances of unusual demand activity were identified.
➢ Pattern established with internal and external factors such as weather, events, competition etc.
➢ The data sources were merged to create feature rich demand data

▪ Training data was split into a train/test split where 70% of data was used to train multiple Machine Learning models and 30% of the data was used to test the models by predict demand.

▪ Models were evaluated using different measures such as RMSE (Root Mean Square Error) measure.

▪ Further models were run in an ensemble formation to observe combination with least RMSE.

Benefits Delivered:

✓ Once the models were ready, they were productionized by wrapping them into RESTful APIs.

✓ The models and APIs were deployed on a cloud environment and interfaced with customer’s existing IT systems.

✓ After the integration, demand is predicted right at the source, thereby helping the customer handle unusual demand activity.

✓ AI driven prediction solution has led to 70% accuracy in demand prediction at a very granular level. This has resulted in significant reduction in losses for the customer


It’s simple.