Optimizing logistics for Seat
Neural-network forecasts help SEAT plan vehicle configurations and shorten delivery times through its Fast Lane service.
Neural-network forecasts help SEAT plan vehicle configurations and shorten delivery times through its Fast Lane service.
Challenge
SEAT needed to shorten vehicle delivery times while managing a large number of possible configurations. Long manufacturing lead times made it difficult to pre-manufacture the right cars and consistently meet customer expectations.
Data
Each car has four configuration levels: model group, trim, TMAIMG and equipment, giving approximately 10,000 possible combinations. The project used three years of historical sales to predict orders for the following three months.
Data preparation removed unused items, certifications and fleets, and updated old TMAIMGs. This reduced the dataset from 100,000 to 70,000 rows.
Solution
Artelnics developed a forecasting software tool tailored to SEAT. It uses neural networks with a self-configurable feed, the Regularized Minkowski loss index and the Quasi-Newton training algorithm.
The Apriori algorithm identifies frequent configuration combinations. Predictions are validated at three levels—Trim, TMAIMG and Combination—by calculating error and accuracy.
Vehicle orders for the next three months.
Mean sales prediction accuracy of 88%.
Results
The predictive model achieved a reported mean sales prediction accuracy of 88%.
Before the project, 50% of orders were delivered in less than one month. With Artelnics’ predictive technology, SEAT shortened the delivery time for 50% of orders to 2–3 weeks through its Fast Lane service.
Illustrations

