From Good to Great: Deep Learning Forecasts Cuts Inventory Costs in Manufacturing
Learn how Aimpoint Digital, in partnership with a Global parts manufacturing company, developed a global neural network model to forecast inventory usage across thousands of SKUs, outperforming the most dependable statistical techniques.
The Challenge
Managing inventory of manufacturing equipment across thousands of sites requires accurate, site-level forecasting to balance carrying costs and service levels. For this large industrials company, the data science team was responsible for predicting weekly SKU utilization at each location. However, many items exhibited highly intermittent demand, limiting the effectiveness of traditional statistical forecasting methods. The team needed a scalable global modeling approach that could leverage shared patterns across SKUs, incorporating signals such as seasonality, geography, and item category to improve restocking decisions to reduce excess inventory and minimize stockouts.
Our Approach
During an accelerated 3-week engagement, Aimpoint conducted detailed segmentation analysis across SKUs, identifying distinct demand patterns and profiles across groups. The team then leveraged a state-of-the-art global neural network model that would use information like seasonality, location, and item-categories to estimate the amount of weekly demand for each SKU.
Results
The improved forecasting framework was integrated into their existing forecasting pipeline, providing immediate value to planning teams after showing improvement at 2 out of every 3 sites during development.
Demand for new SKUs, which most statistical techniques would struggle to deal with, can be forecasted 12% more accurately since the global neural model will be able to use information from other items’ histories.
Our client will also benefit from a 75%-90% reduction in model retrainings, which will lower compute costs while still maintaining strong predictive performance.
Key Takeaways
By uncovering global patterns across SKUs, our neural network model outperformed the previous solution:
When classical forecasting methods reach their limits, it may be beneficial to explore options like global neural models that can look farther and wider for key patterns. When developed methodically, these tools can take your solution from acceptable to state-of-the-art.
The client now has a global forecasting model that improves accuracy across a wide range of SKUs, including those with intermittent or limited demand history. This framework was scaled across the business and provided reduced operational overhead and compute costs, supported better inventory planning, and laid the groundwork for AI readiness.
Let’s talk data.
We’ll bring the solutions.
Whether you need advanced AI solutions, strategic data expertise, or tailored insights, our team is here to help.