Real-time SKU optimization and bottleneck prediction using machine learning.

Developed a predictive machine learning model (Llia) to forecast real production time per SKU, enriched with process, quality, and supply chain data, improving factory planning accuracy by 65%.

01. Problem

Suspensys faced challenges in production planning driven by oversimplified historical data that failed to capture SKU-level cycle time variability on the shop floor. This led to inaccurate production planning, fluctuating cycle times, and unplanned material shortages and bottlenecks, ultimately reducing operational efficiency.

02. Assessment

DB’s analysis concluded that a simple linear forecasting model would not be sufficient due to high industrial variability. The solution required a multidimensional data approach, connecting legacy systems with real-time data from the shop floor and supply chain.
The key was building a robust predictive model capable of accurately forecasting actual production time per SKU.

03. Solution

DB built and trained a robust machine learning model, named the Llia Virtual Assistant. The model was enriched by integrating and cross-referencing three critical data domains: production processes (machine sensors), quality (inspection data), and supply chain (inventory and lead times).
This holistic data approach enabled refined SKU-level predictions. A continuous retraining cycle was implemented to ensure model relevance amid operational changes and new SKUs.
🇪🇸 Español:

Benefits achieved

Significant improvement in planning accuracy

Increased production forecasting accuracy by 65% per SKU compared to the previous method, resulting in more stable and reliable production planning.

Bottleneck anticipation and mitigation

The model predicts operational bottlenecks before they occur, enabling factory managers to proactively reallocate resources.

Reduced material stockouts

Integrating supply data with SKU forecasting significantly reduces unexpected inventory stockouts and optimizes material flow.

Continuous retraining and adaptation cycle

An automated data pipeline enables continuous model learning from new data, ensuring long-term relevance and performance.