Learning algorithms evaluate mass data extremely effectively. In an avalanche of data that humans can no longer handle, they can identify relevant information very quickly and intelligently. E.ON SE wants to use this potential for risk management and is therefore testing Horváth & Partners' Global Risk Radar. The pilot project is of particular interest to the group's credit risk management. This is because in future the department wants to identify the risk profile of business partners, who they call counterparties, more effectively and in a more targeted manner, based on automated text analysis of unstructured big data from the Internet.

Bernd Kälber, Program Manager Advanced Analytics & Artificial Intelligence at "Digital@Finance", attended the first live presentation of the newly developed application at the 2017 Corporate Risk Minds conference. He was immediately struck by the tool's potential. As a physicist with a focus on theoretical mathematics, he is convinced that "Automated analysis of big data will be indispensable in future in making competitively sound strategic and operational decisions".

The pilot project started in November 2017. Bernd Kälber reviewed jointly with the machine learning and risk management experts at Horváth & Partners to what extent the critical information for evaluating risks could be generated using the Global Risk Radar. The focus was initially on the use cases relating to the topics of solar and wind energy, innovative storage technologies and pricing in an international context. During the course of the study it became apparent that the Global Risk Radar was particular strong in detecting early warning signals for potential rating changes to counterparties.

Given increasingly decentralized generation mainly from energy sources that are sensitive to weather conditions, and the development of new digital technologies, providers need to be turning themselves into digital "green", service-oriented solution providers. "This realignment affects all levels within our group, including permanent changes in the finance area," says Bernd Kälber. "Intelligent algorithms and machine learning can potentially provide major support for this transformation."

Machine Learning enables computers to independently analyze very large, complex and unstructured data sets using specially developed algorithms, and to process the results without further input. Initially, the algorithm itself has no relevant knowledge. It needs to be trained for each separate question. Different starting points require different approaches. These can be classified into three categories:

1. Supervised learning
In this case, the algorithm is trained in a focused manner using known input-output data sets, teaching it the connections between data.

2. Unsupervised learning
The input-output context is not known in advance, and the algorithm itself classifies the data and recognizes potential patterns.

3. Reinforcement learning
An algorithm learns a specific task by attempting to improve itself using feedback from its context.

The machine text analysis capability of the Global Risk Radar is based mainly on the first approach, supervised learning. For the pilot project, the Radar analyzed several million text documents from thousands of data sources on the Internet. The semantic analysis capability of the tool is constantly improving while it works, thanks to continuous machine learning.

In E.ON's "Digital@Finance" program, the pilot project provides the central component for text analytics. In order to rate counterparties, text-based mass documents from the Internet are analyzed by the Global Risk Radar looking for early warning signals. This means that information relevant to a decision can be taken into account well before it is published in traditional sources, such as rating agencies. Following the completion of the pilot phase, the energy company will therefore review the options for operational application of the tool for credit risk management.