Skip to main content Skip to footer


Steelmakers need new tools for intelligent pricing


January 3, 2022

Steel is the most traded industrial commodity and acts as the fuel for the growth of both developing and developed nations around the world. But, like other sectors, the steel industry is experiencing supply chain disruption and sharp price fluctuations. As demand has grown after the initial pandemic-related shock—prompted in part by fiscal policies enacted by major countries—price increases have become the norm, but uncertainties abound.

In the second half of 2021, we saw prices hit new peaks, due not only to cost-push factors (such as increases in prices for raw materials including iron ore and coking coal) but also to increased demand from downstream sectors such as the automobile and construction industries. Companies’ pricing strategies, however, are typically reactive in nature, putting steel companies at risk of financial loss as markets fluctuate. Forecasts, whether internally generated or purchased from third-party vendors, are often influenced by “expert bias” and are often wrong, putting additional pressure on margins.

Steel companies need to become better at predicting and setting prices, and should do so quickly. Fortunately, new technologies could help steel companies forecast steel prices more accurately, on a near-real-time basis.

Rethinking the traditional approach

Steel companies gather supply, demand, trade, steel capacity, production cost and pricing data from multiple sources, then apply various macroeconomic and industry-specific forecasts to come up with future price trends. While price prediction has always been data-driven, steel producers, traders, buyers and financiers have a huge opportunity to leverage big data and analytics-driven algorithms.

Typically, steel companies focus on the direct variables impacting their prices and use the outcomes of basic predictive algorithms for decision-making. To realize more value out of pricing programs, steel companies need sophisticated algorithms that process more than just demand-supply variables—such as global GDP, industrial production and currency exchange rates—and simulate probable price behaviors more accurately.

Price prediction projects should be initiated by defining the business context in which a steel company operates, such as capacity restrictions imposed by local governments, varying levels of market demand, evolving customer needs and global events optimizing trade contracts. After defining the context, the company can conduct research to study various factors impacting steel prices in the short and long run.

With this knowledge in hand, companies can use cutting-edge technologies, such as machine learning and artificial intelligence (AI), to provide the ability to forecast steel prices. These technologies also eliminate human biases and selective judgments, which are among the biggest issues in price prediction.

Companies can use cutting-edge technologies, such as machine learning and artificial intelligence, to provide the ability to forecast steel prices.

We see a four-step approach to developing a new pricing strategy:

1. Obtain and organize the right data. Steel companies need data on the factors affecting steel prices in the short and longer term. That data should be cleaned to help analysts understand and treat data anomalies and missing values.

2. Create new models. Companies will need advanced multi-variate models to understand the causal relationship between steel prices and raw materials prices, demand and supply drivers, global and domestic steel production capacities, and macroeconomic indicators such as gross domestic product, personal consumption expenditures and industrial production. Companies also need close analysis of industry-specific factors such as lead time, shipments, imports and exports.

3. Test the models. During testing, companies can observe the performance of univariate/time series models used for predicting the value of factors, technically called regressors (such as scrap, coking coal and iron ore prices). These models can perform well in normal periods but may struggle to forecast trends during times of market volatility. To overcome this challenge, data scientists can:

  • Improvise univariate models’ performance through regular accuracy tracking.
  • Plug the external accurate forecasts of regressors (provided by 3rd party sources) into the price prediction model to get better steel price forecasts, instead of using a time series method.
  • Explore the possibility of leveraging commodity markets’ future contract prices for regressors such as scrap, coking coal and iron ore.
  • Deploy multivariate models to forecast the regressors mentioned above.

4. Incorporate “what if” capabilities. In highly dynamic market conditions, even advanced models may have difficulty understanding and anticipating prices. Sophisticated steel price prediction algorithms can be coupled with “what if” capabilities such as scenario managers to better equip businesses with the ability to translate market and business intelligence into pricing realization. Analysts can run different scenarios to visualize the impact on steel prices of force majeure events as well as market conditions.

Benefits of a new technical architecture

In addition to these steps, steel companies should focus on developing a flexible, cloud-based technical architecture that could help compute model outputs at speed, and help simulate multiple scenarios which may mimic upcoming market realities. Cloud aids in the rapid development, deployment and maintenance of machine learning models that can deliver near perfect forecasts and predictions over the long term.

With modern architecture, data analysts can focus solely on model development without having to worry about infrastructure related issues. Data engineers can leverage cloud to collect, transform and manage data through data pipelines. Cloud can also help in creating automated workflows to execute independent tasks in a connected manner.

Multiple divisions across the company can use analytics-based steel price forecasting. For example:

  • Commercial divisions can build better customer satisfaction indices by providing dynamic and competitive offers, creating near perfect revenue forecasts and achieving improved margins.
  • Procurement groups can efficiently control inventory and stock levels of raw materials through improved long-term hedging strategies.
  • Production planners can optimize and allocate production capacities effectively.
  • Project managers can cross-leverage pricing analytics to facilitate strategic decisions, such as green field or brown field project investments.

Steel will likely remain an essential component of economic growth. However, emerging trends like the push for sustainable production processes (such as scrap-based production) will create additional disruption in markets. There will be many new determinants involved in understanding supply, demand and pricing trends. Steelmakers will need all the tools at their disposal—including AI, machine learning, cloud, analytics, modeling and other innovative technologies—to understand these factors and offer intelligent pricing to their customers.


Special thanks to Sean Keenan and Somya Ranjan Shaw for their contributions to this blog.


Mithilesh Kumar

Director – Technology