Overview
Steel is the backbone of industrial progress, fueling everything from construction to automotive manufacturing. Yet, behind every ton of finished steel lies a complex chain of procurement, production, and delivery that must be precisely orchestrated. For one of Latin America’s largest steel manufacturers, this orchestration was being challenged by volatile scrap prices, inefficient order management, and blocked inventory.
Our client serves multiple countries across Latin America, supplying high-quality steel products across three key segments: manufacturing, construction, and distribution. With operations spanning multiple plants and thousands of SKUs, the client manages a vast vendor ecosystem and complex customer agreements. However, their legacy planning and order management processes led to rising costs, shipment delays, and inefficient resource utilization.
Partnering with PalTech, the client embarked on a transformation journey to modernize their demand forecasting, scrap price prediction, and inventory planning systems. By deploying advanced machine learning solutions, automated pipelines, and robust integration with their business processes, PalTech enabled the client to significantly reduce costs, improve forecasting accuracy, and increase order completion efficiency.
The results were striking: ~$20 million in savings, a 15% increase in order completion, and major improvements in procurement and planning.
Introduction
The steel industry faces unique challenges: volatile raw material prices, fluctuating customer demand, and long manufacturing cycles. Scrap metal, a critical input for steel production, is particularly hard to forecast due to global market dynamics. At the same time, inefficiencies in order management, where invoices were generated only after shipments, created bottlenecks in planning. Customers could cancel orders at the last minute, leading to blocked inventory and wasted resources.
The client needed a solution that not only forecasted demand and raw material prices with precision but also integrated these insights into their order management and production planning workflows. PalTech was chosen as the strategic partner due to its expertise in combining AI-driven forecasting with enterprise-scale automation.
Problem Statement
The client’s existing processes were reactive rather than predictive:
- Unreliable scrap procurement – With no forecasting mechanism in place, procurement teams were often caught off guard by sudden price fluctuations. This led to either over-purchasing at high prices or under-purchasing, risking production delays.
- Blocked inventory – Orders were sometimes manufactured but not accepted by customers due to last-minute cancellations. This locked up capital and warehouse space, hampering new order fulfillment.
- Manual, siloed forecasting – Demand was projected manually, often using spreadsheets and historical averages. This lacked the granularity needed across SKUs, plants, and customers.
- Delayed invoicing process – Since invoices were generated only after shipments, cancellations and disputes frequently occurred, affecting revenue realization.
The result was a cycle of inefficient production planning, poor inventory visibility, and financial leakage. The client needed an intelligent forecasting and order efficiency model to break this cycle.
Challenges
Several challenges made this transformation complex:
- High data complexity – The client had million order records spanning across years. Data was scattered across multiple systems, requiring extensive cleaning and integration.
- Granular forecasting needs – Forecasts were needed not just at an aggregate level but hierarchically across SKU, customer, plant, and block levels.
- Volatile scrap prices – External market data had to be incorporated to accurately predict price fluctuations.
- Late cancellations – Predicting which orders were likely to fail was difficult, given that cancellations could happen without notice.
- Scalability – Any solution needed to be robust enough to handle monthly retraining and predictions across thousands of SKUs and multiple plants without manual intervention.
Tech Stack at a Glance
PalTech leveraged a modern and scalable stack tailored for demand forecasting, risk classification, and procurement optimization:
- Programming & Machine Learning: Python, Scikit-HTS, TensorFlow Probability, XGBoost, Amazon SageMaker
- Cloud & Automation: AWS Lambda, Amazon EventBridge, AWS Step Functions, Amazon CloudWatch
- Data Storage & Processing: AWS Data Lake (Amazon S3 + Glue Data Catalog + Athena), Amazon RDS (PostgreSQL), AWS Glue ETL
- Visualization & Reporting: Microsoft Power BI, AWS QuickSight (for rapid prototyping)
- Deployment & Integration: Docker, CI/CD Pipelines (AWS CodePipeline), REST APIs for ERP/WMS/TMS integration
PalTech Approach
PalTech designed a multi-pronged approach to address both demand forecasting and order inefficiency challenges.
Unified Data Lake Creation
- Built a centralized AWS data lake (Amazon S3 + AWS Glue Data Catalog + Athena) to ingest and unify ERP/MES/WMS/TMS data with external scrap price feeds.
- Integrated million records of historical order data and set up incremental ingestion with schema evolution and data quality checks.
- Standardized a “golden” data model so forecasting, classification, and dashboards could seamlessly access clean, governed datasets.
Demand Forecasting & Scrap Price Prediction
- Built a hierarchical time series model using Scikit-hts to capture dependencies across SKUs, plants, and customers.
- Developed a secondary forecasting layer using TensorFlow Probability to account for uncertainty and enhance robustness.
- Leveraged Amazon Forecast Service to predict scrap prices, giving procurement teams visibility into cost trends and helping negotiate better contracts.
Automated ML Pipelines
- Implemented end-to-end ML pipelines in Amazon SageMaker.
- Automated monthly retraining and prediction cycles using Amazon EventBridge, ensuring forecasts were always up to date.
- Designed an architecture that stored results in centralized data repositories accessible to both procurement and production teams.
Order Risk Classification
- Designed a classification system to score orders for cancellation/inefficiency risk at entry, using features like lead times, credit terms, prior cancellations, and line-item volatility.
- Integrated risk flags and thresholds into order management workflows to reduce blocked inventory and improve plan adherence.
Proactive Business Intelligence Dashboards
- Developed Power BI dashboards:
- Forecast vs actual performance
- Scrap price trends
- Order inefficiency probabilities
- Inventory health metrics
- Conducted training sessions with client teams to embed data-driven decision-making into daily operations.
Key Benefits
The transformation delivered by PalTech generated both measurable financial gains and strategic advantages:
- $20 Million Savings – Reduced blocked inventory and optimized procurement decisions delivered cost savings of approximately $20 million annually.
- Forecast Precision Gains — Reduced forecast error by 22% WMAPE, kept bias within ±3%, and improved service levels by 7%, enabling tighter production and procurement alignment.
- 81% Accuracy in Predicting Inefficient Orders – Early identification of risky orders minimized last-minute cancellations and improved customer service levels.
- 15% Increase in Order Completion – Optimized planning and efficient order management improved order fulfillment rates.
- Streamlined Procurement – Procurement teams used price forecasts to negotiate contracts more effectively, reducing exposure to market volatility.
- End-to-End Automation – Monthly predictions were automated, eliminating manual overhead and ensuring consistent accuracy.
- Data-Driven Culture – By embedding dashboards and predictive insights into daily workflows, PalTech empowered business users to make smarter, faster decisions.
Conclusion
In an industry where a single percentage point improvement in efficiency can translate into millions of dollars, PalTech’s partnership with the client was transformative. By combining hierarchical forecasting, probabilistic modeling, and intelligent automation, PalTech helped Latin America’s largest steel manufacturer turn data into a strategic asset.
From scrap price prediction to demand forecasting to order efficiency modeling, the solutions delivered not only immediate financial impact but also long-term resilience. Today, the client operates with greater confidence, improved agility, and a robust decision-making framework that positions them strongly in a highly competitive market.
This engagement underscores PalTech’s commitment to helping global enterprises unlock business value through AI-driven innovation and practical, scalable solutions.