Forecasting model for home shopping
- Client: EROSKI
- Country: Spain
- Year: 2022
- Applied technologies: Analytics | Artificial Intelligence | Machine Learning
Eroski Implements Advanced Machine Learning Model to Optimize Last-Mile Transport and Save Costs.
With a workforce of over 35,000 employees across the country, Eroski is a leading distribution company in Spain. In the retail industry, accurately predicting future trends such as product sales, turnover, and home deliveries is crucial. Additionally, forecasting the number of customers passing through each store’s checkout during specific timeframes is essential for determining optimal staffing levels. Achieving these goals requires leveraging a combination of models capable of capturing data trends and seasonality, enabling the generation of accurate predictions with minimal errors.
Ayesa has developed a cutting-edge data analytics system that leverages artificial intelligence and machine learning to optimize Eroski’s last-mile transport service for home deliveries.
By analyzing years of historical data, the algorithm can predict the number of daily orders, their locations, preparation centers, and time slots. The machine learning algorithms have been specifically designed to account for external factors that may impact the accuracy of the predictions, such as holidays (and during the pandemic) with restrictions on mobility.
The project was implemented using Rocket’s analytical workflows on the Stratio platform, with model tracking, persistence, and comparison facilitated by MLPojects and mlflow to ensure the quality of the machine learning models throughout their lifecycle.
The advanced data analytics system developed by Ibermática and Ayesa has numerous benefits for Eroski’s home delivery service:
Predictive modelingenables Eroski to know the number of orders in advance, their geographic location, preparation centers, and time slots. This information helps them adjust the negotiation of rates with their transport suppliers, leading to significant cost savings for both Eroski and the suppliers.
Machine learning models forecast the demand for orders that each of the transport groups will need to deliver in each time slot and day, up to two months in advance. This level of accuracy helps to optimize the delivery process, without affecting customer satisfaction. The system is particularly useful in the post-pandemic era, where home delivery services have become increasingly popular. The ability to accurately predict demand and optimize the delivery process is essential in meeting the growing needs of customers.