The Challenge
The retailer’s trusted demand forecasting model could support only a fraction of its store network because the underlying infrastructure was built for sequential, pilot-scale processing.
The Solution
Nineleaps redesigned the forecasting infrastructure using intelligent data filtering and distributed Apache Spark processing. Irrelevant sales and promotional records were removed before computation, while product groups were processed concurrently across the cluster. The modernized platform expanded forecasting from 829 to 3,316 stores and from 26 to 676 product groups. Validation across 876,253 output points confirmed that the new infrastructure preserved the existing model’s forecast results while completing a full production run in approximately four hours at a cost of $1.15.
Services
- Data Engineering
