The challenge
Before automation, the manufacturer faced growing product variability and increasingly complex production planning. Every order included numerous parameters that had to be evaluated correctly to avoid downtime and lost efficiency. With high order volumes, even small inaccuracies translated into significant time and capacity losses.
-
A growing portfolio: hundreds of products defined by combinations of layers, materials, and print types had to be classified manually.
-
Dependence on specialists: assigning orders to production compatibility groups relied on senior planners.
-
Limited scalability: expanding the product range risked slowing production down and driving up costs.
-
Risk of errors: manual assignment led to inconsistency and delays.
The solution
Aspecta designed an AI classifier on AWS (SageMaker, Lambda, Glue, S3, API Gateway). This approach was chosen over a traditional rule-based system or an on-premises application because it supports rapid model training, scales with demand, and integrates easily into the client’s existing processes.
The AWS architecture keeps maintenance costs low and allows the model to improve continuously through a feedback loop. The model uses historical order data, layer parameters, print types, and materials to determine the correct production compatibility group automatically.
Main components of the solution:
-
Data layer: centralized storage (S3), ETL pipeline (Glue / Lambda).
-
Modeling: SageMaker with XGBoost / LightGBM / CatBoost; metrics: accuracy, F1, confusion matrix.
-
Deployment: SageMaker Endpoint or Lambda with API Gateway for real-time predictions.
-
Integration: API connected to ERP and PLM.
-
Monitoring: CloudWatch, SageMaker Model Monitor, feedback loop.
-
Security: IAM, encryption, audit logs.
Results and benefits
The deployment delivered measurable results that confirm its value in practice. The key indicators show improvements in accuracy and speed, and a substantial improvement in overall planning efficiency.
-
Classification accuracy: 96.2% (target 95% or higher).
-
Processing time: from 5 to 10 minutes down to around 7 seconds.
-
Operational impact: faster planning, fewer errors, more consistent assignment.
-
Scalability: no manual rule updates required.
“Automated decision-making sped up our planning and reduced our dependence on expert know-how.” – Head of Planning
Conclusion
The project confirmed that a well-designed AI model can change how manufacturers handle growing portfolio complexity and the demand for fast decisions. Combining domain expertise, AWS infrastructure, and integration with ERP systems produced a solution that is accurate, stable, and sustainable in day-to-day operation.
Wondering how AI could improve the efficiency of your decision-making and planning? Get in touch for a free introductory consultation.
