The Challenge
247TailoredSteel had built something genuinely rare: a fully automated production environment capable of handling custom, small-batch steel orders at scale. Where traditional manufacturers struggled with the economics of low-volume, high-mix production, 247TailoredSteel had engineered their way past that constraint. Automation was their competitive edge and for a long time, it was enough.
But as order volumes grew and product diversity increased, a different kind of problem came into focus. The same flexibility that made their model work was quietly generating waste and without the right intelligence layer on top of their automated lines, material losses translated directly into margin erosion at scale.
- Structural waste in production planning: Diverse, custom orders often couldn’t be optimally arranged on single steel plates, leading to significant material waste in day-to-day operations.
- Hidden cost and margin pressure: Waste represented a multi-million euro cost item that directly eroded margins and undermined 247TailoredSteel’s cost-leadership position in highly automated, small-volume production.
- No systematic reuse of waste: There was no reliable way to forecast whether future incoming orders could still use existing waste materials, so planners were effectively making high-stakes decisions on gut feeling.
- Lack of analytical infrastructure: While there was an operational SQL database for transactional data, there was no analytical data platform to turn that data into actionable steering information for waste reduction and dynamic pricing.
- Scaling AI across plants and product types: Different steel types, order profiles and manufacturing locations required multiple models and a robust MLOps setup, something that was missing in the existing engineering landscape.
Our Approach
We became embedded partners in their existing team structure, filling the gap between their data science capabilities and engineering execution while working directly with their investors on this critical pricing optimization initiative.
- Azure-Native Data Architecture: Built a comprehensive data platform using Azure Data Factory and Databricks with Delta Tables, moving from operational databases to analytical infrastructure designed for manufacturing scale
- Implemented event-based streaming pipelines using Azure Event Hubs for real-time manufacturing line integration
- Created dynamic data flows that could handle diverse order types and steel specifications across multiple facilities
- MLOps for Manufacturing Diversity: Deployed MLflow integrated with Databricks to manage multiple prediction models tailored for different order types and steel varieties
- Each manufacturing location required customized models due to different equipment capabilities and regional demand patterns
- Kubernetes-based deployment enabled consistent model rollouts across all production facilities
- Predictive Waste Optimization: Built forecasting models that predict incoming order patterns and match current waste materials with likely future demand
- Integrated these predictions directly into their pricing algorithms and manufacturing scheduling systems
