Our client is a leading global fashion retailer, selling in thousands of stores in countries all over the world.
The client aimed to improve their capabilities for accurately forecasting product sales of new items, specifically within the first weeks after release. This would allow them to become more effective at allocating products and avoid overstocks or understocks.
N-iX helped the client improve the existing forecasting model that helps predict product sales of new items. The model compares the new products with similar ones sold previously (based on color, size, product group, etc.) and makes accurate forecasting based on the past sales of these items. The model also takes fashion trends into account to predict customer behavior.
We utilized MLflow to experiment with the model and track results. It allowed us to log experiments, save information about various runs, reproduce results, and share findings among team members. Our engineers used Snowflake to handle the client’s datasets and applied ETL (Extract, Transform, Load) processes to transfer data to our systems for further processing and analysis.
We experimented with various data formats and indicators, and offered new calculation methods to increase the model's success rate measured by how close the amount of supplied items is to the sold items. N-iX improved the model’s accuracy, achieving a success rate of over 50% and reducing the product deficit rate to just 5%. This helped optimize inventory management and reduce the amount of unsold items, leading to significant cost savings. We also investigated and enhanced the model’s capability to minimize the difference between predicted and actual sales.
Additionally, we introduced the practice of including sales trend data within item subgroups. This helped ensure a sufficient supply of items in specific colors and/or sizes and improved the forecast accuracy by 2%.
By enhancing the sales forecasting model with N-iX, the client obtained several key advantages:
- Optimized expenses by improving the sales forecast accuracy by up to 50% and helping avoid overstocks or understocks;
- Streamlined product management and allocation, ensuring products are available in the right quantities, colors, sizes, etc.