In today's competitive retail landscape, forecasting demand accurately makes or breaks any business. A major global retailer relied on a slow and resource-intensive forecasting process that made it hard to keep up competitiveness. With VirtusLab’s help, the retailer now uses a new forecasting framework that allows for faster creation of forecasts and the addition of further predictions easily.
Our client faced a significant challenge with their existing forecasting process. The slow, resource-intensive process lacked the flexibility to easily add long-term forecasts. The prediction of future sales, inventory levels, and customer demand for their products lagged behind, disrupting the retailer's efforts to plan their buying and merchandising strategies. The hoped profit maximization was undermined by overstocking or understocking of products.
VirtusLab (VL) implemented a new forecasting framework that streamlined the entire process, making it faster and more efficient than ever before. By introducing parallelization with Apache Spark, we enabled our client to take advantage of multiple computers and run the process simultaneously, eliminating the bottleneck caused by using only a single computer.
The framework relied on complex time series analysis and statistical models to predict future trends better. VirtusLab processed several hundred terabytes of historical data, analyzed it, and ran validation tests to ensure accuracy. The framework VL developed includes handling external data sources and factors such as inflation rates that may impact sales forecasts.
Our client experienced several positive outcomes as a result of implementing the new forecasting framework:
Enhanced Data Science: data scientists now focus on critical business analysis rather than just performing manual calculations of forecasts.
Languages: Python, Spark
Database: Hive, S3
Infrastructure: On-prem: Yarn and Kubernetes
Cloud: Azure Databricks
CI/CD: Jenkins, Airflow
Modelling: Statistical models
