BaFin Issues Principles on Use of Big Data & Artificial Intelligence

To formulate the principles as precisely as possible, an algorithm-based decision-making process has been broken down into two phases—development and application. The development phase examines how the algorithm is selected, calibrated, and validated. For instance, principles for the development phase relate to the relevant data strategy as well as documentation to ensure clarity for both internal and external parties. In the application phase, the results of an algorithm must be interpreted and included in the decision-making process. This can either be done automatically or by involving experts. A functioning mechanism comprising elements such as sufficient checks and feedback loops for the development phase must be established in all cases. Aside from these two phases, certain overarching principles are important for the creation and application of the algorithm and these include:

• Clear management responsibility. Senior management must have sufficient technical expertise. Also, reporting lines and reporting formats must be structured to ensure that communication is risk-appropriate and geared to the specific requirements of the target audience, from the modeler right up to senior management. Moreover, the business-wide strategy for using algorithm-based decision-making processes should be reflected in the IT strategy. There must also be staff with the necessary technical knowledge in the independent control functions.
• Appropriate risk and outsourcing management. Senior management is responsible for establishing a risk management system that has been adapted for the use of algorithm based decision-making processes. If applications are used by a service provider, senior management must also set up an effective outsourcing management system. Responsibility, reporting, and monitoring structures must be set out clearly within this context. Measures to minimize cyber-security risks should also be adapted if required and the complexity and data dependency of the algorithm must be considered.
• Prevention of bias. Bias must be prevented to be able to reach business decisions that are not based on systematically distorted results, to rule out bias-based systematic discrimination of certain groups of customers, and to rule out any resulting reputational risks. This is a key principle since such biases may occur from the development of the process to its application. 
• Ruling out of types of differentiation prohibited by law. In case of certain financial services, the law stipulates that certain characteristics may not be considered for differentiation purposes—that is, to calculate risk and prices. If conditions are systematically set out on the basis of such characteristics, there is a risk of discrimination. Such a risk also exists if these characteristics are replaced with an approximation. This would be associated with increased reputational risks and, in some cases, legal risks. In certain circumstances, BaFin might consider it necessary to take measures to address violations of statutory provisions. Companies should, therefore, establish (statistical) verification processes to rule out discrimination.

These principles represent a milestone in the efforts of BaFin and Bundesbank to create legal and application certainty for the responsible use of big data and artificial intelligence in the financial sector. 

Related Links

• News Release
• Supervisory Principles (PDF)

Keywords: Europe, Germany, Banking, Big Data, Artificial Intelligence, Supervisory Principles, Cyber Risk, Algorithms, Outsourcing, Pillar 1, Pillar 2, Regtech, BaFin