How Machine Learning Will Help This Russian Steel Giant Prevent Energy Fraud and Stay No. 1 in Profitability

Steel and mining giant Severstal may rank fourth among fellow Russian steel producers but it is the most profitable company in the industry globally.

Last year, the Cherepovets-based company reported $2.5 billion earnings before interest, taxes, depreciation, and amortization (EBITDA) and almost $8 billion in revenue for 2017. But the company still struggles to minimize energy costs and combat electricity fraud, a common problem across Russia.

Andrey Kostenko is a Research and Development director at Severstal’s utilities security project, called SKIF. He says, “You may ask why this is our main business challenge — energy represent 10 percent of our total annual costs.” While Severstal generates 80 percent of its electricity, it depends on suppliers for the remaining 20 percent.

Kostenko explains that Severstal can’t accurately assess how much electricity it consumes in real time, so forecasting is very difficult: “We have to give a forecast for every hour of the next day to our supplier. We pay fines if we exceed that forecast or if we don’t reach the amount forecasted.”

The result? According to Kostenko, Severstal spends $12 million a year because of extra costs. “With electricity imbalances, we don’t know how much we really spend or consume.” The company pays an additional $1 million in fines due to incorrect energy forecasting.

As it turns out, energy fraud — and the lack of a real-time fraud detection system — has made accurate insight difficult. Fraud can include energy theft, “fake” energy companies asking for upfront payment without providing electricity, or utility companies charging more for energy than they’ve actually supplied.

Severstal believes that technology innovation will help prevent fraud and improve energy management. That’s why Severstal turned to SAP for help. Already an existing SAP customer for ERP and mining solutions, Severstal partnered with SAP to develop a prototype using Internet of Things (IoT) technologies to harness meter data, as well as machine learning and advanced analytics solutions from SAP Leonardo, a set of groundbreaking technologies, industry expertise, and services that help business become more intelligent.


Click the button below to load the content from YouTube.


In a highly competitive industry, this project exemplifies Severstal’s strategy to explore technology that will help bolster its profit leadership. According to its 2017 Annual Report, last year the company invested $27 million in IT and digital projects and “developed a strong digital team dedicated to trialing and introducing the latest international technologies.”

SAP started the project with a design thinking workshop to help Severstal clearly articulate its energy management problem and identify who would use the technology. Discovering the problem was straightforward — improve forecasting and stop fraud — but pinpointing the end user led to some surprises. The team originally thought it would be used by electricity planners and technicians but ultimately, as Kostenko explains, “our real key users are security officers because they are responsible for detecting imbalances.”

Using the SAP Leonardo design-led approach and intelligent technologies, the companies created a system that monitors real-time energy use and analyzes energy consumption disparities. For example, one team may report that they used 100 kilowatts but department-wide usage shows that team used 400. The solution identifies the discrepancy and triggers an alert to security officers that there may be electricity fraud.

Kostenko reports, “This tool has started saving us money already. Even being a prototype, we’ve detected several core problems and started solving them. I would guess that we’ve already saved one or two million dollars. And we hope real-time monitoring of imbalances and electricity consumption will let us save millions more.”

SAP Leonardo machine learning algorithms have been trained to recognize patterns in electricity use — and detect and flag deviations. SAP Analytics provide security officers with transparency into where the aberrations exist, highlighting problem zones.

Going forward, Kostenko and his team believe those machine learning algorithms could be applied in other use cases. For example, machine learning could help Severstal understand discrepancies in liquid natural gas, oxygen, and inert gas consumption. It could also be applied to business processes like accounting and procurement. Additionally, the company plans to add Big Data and blockchain capabilities to the fraud detection solution.

According to Kostenko, “As we become more intelligent, we will be able to manage more of our problems.”