BaFin - Navigation & Service

Erscheinung:16.07.2018 | Topic Fintechs Big Data and Artificial Intelligence: “The machines can’t be allowed to assume responsibility, even in automated processes.”

Interview with BaFin President Felix Hufeld, Gerold Grasshoff (Senior Partner, The Boston Consulting Group), Professor Stefan Wrobel (Director, Fraunhofer IAIS) and Claus Wechselmann (Managing Director, PD – Berater der öffentlichen Hand)

Mr Hufeld, why did BaFin commission this study? What were the main objectives of the study?

Hufeld: One of the effects of digitalisation is that the companies we supervise now have completely different opportunities for storing and analysing data. This will fundamentally change the way financial services are provided. We need to understand these trends to allow us to discuss the resulting implications for financial regulation and supervision.

Mr Grasshoff, Professor Wrobel, Mr Wechselmann, you worked together intensively with BaFin on this study for several months. What was your approach?

Grasshoff: A holistic approach was pivotal for developing the project. Based on an in-depth technological analysis by the Fraunhofer Institute IAIS, it allows us to infer well-founded strategic, opportunity- and risk-related implications for the banking, insurance and capital markets. In turn, we then developed the supervisory and regulatory consequences from this. All three dimensions of supervisory and regulatory activities were analysed, starting with financial stability and market supervision, through firm supervision, down to collective consumer protection.

Wrobel: From a technological perspective, at the moment we can see that core technologies in the context of digitalisation – above all Big Data and Artificial Intelligence – are being used in an increasingly wider range of areas. However, not all of these approaches are viable or sufficiently resilient to withstand the challenges in the financial industry. We have therefore concentrated on separating the wheat from the chaff and identifying technologies that combine a high level of potential application and social responsibility in the long term.

Wechselmann: Together with BaFin, we assembled a project team in a very short time that combines the broad-ranging expertise and experience needed for the holistic approach underpinning the study. Continuous interaction and in-depth discussion of the perspectives relating to technology, economic strategy, supervision and regulation were the critical success factors for the project and the results of this study.

What challenges did you face?

Hufeld: The widespread involvement of almost all areas at BaFin and several external project partners certainly demanded considerable commitment and effort from everybody involved. In addition to project management, BaFin involvement was not just limited to banking, insurance and securities supervision, but also extended to our experts in consumer protection, risk models and IT, for example. As far as our external partners are concerned, the Boston Consulting Group contributed its market and sectoral expertise, Fraunhofer IAIS was in charge of the technical aspects, and Partnerschaft Deutschland supported us in setting up and managing the project. My thanks go to all of them.

What do you think are the most important results?

Hufeld: From BaFin’s point of view, it is of fundamental importance that the machines can’t be allowed to assume responsibility, even in automated processes. Responsibility definitely remains in the hands of management. That’s why it is also imperative for the processes to be embedded in a proper business organisation, regardless of how automated they are. We must at all times be able to track how a fully or partly automated process arrives at a decision. This is particularly important because it is the only way that we – as supervisors – stand any chance of detecting errors in the analysis process at an early stage and of intervening accordingly. Of course this also applies to the companies themselves.

At a glance:Explainability of complex models

Generating transparent, understandable models is currently one of the most important fields of research in machine learning in order to understand and analyse how intelligent systems arrive at a decision at any point in time. In contrast to “black box models” – purely statistical, data-driven learning models – transparent models also include expert knowledge for explanatory purposes so that the logic or the individual system outputs can better be understood. One of the most famous scientific approaches is the LIME (Local Interpretable Model-Agnostic Explanations) algorithm, which provides a local explanation model for the individual case to be explained and similar data points using simpler methods. Other approaches generate directly understandable models, for example as rules. Overall, even with very complex models, new approaches enable at least insights into the functioning of the model and the reasons behind decisions.

BaFin additionally has a special role in the field of consumer protection. Genuine data sovereignty is an increasingly important regulatory goal. If customers decide to pay for certain ostensibly free services on the internet using their data, that is their right. However, it is important that they understand the value of their data and above all know who will have access to their personal information. If customers can assume that their data will be handled responsibly and transparently, this can only increase confidence in the financial sector.

Mr Grasshoff, what implications do you consider to be particularly relevant?

Grasshoff: Big Data and Artificial Intelligence open up additional competitive opportunities in the financial sector – for both existing and potential new market participants. There is one overriding reason for this: the new technologies reinforce the trend towards the disaggregation of the value chain. In particular, we are expecting a higher degree of decoupling between the customer interface and core processes.

For example, tech companies have huge amounts of customer data that they can use to develop new personalised offerings in the banking and insurance sectors. Competition at the interface to the customers is driven by their expectations. Companies that use Big Data and Artificial Intelligence to develop innovative solutions will be the winners in this competition. Another factor is that the Second Payment Services Directive has opened up access to the underlying customer interface in payment transaction systems for companies that are not traditionally active in the financial sector.

In addition, we see tremendous potential in the core technical processes, for example in loan processing, securities settlement and payment settlement, as well as in bank control processes, thanks to Big Data and Artificial Intelligence. Speech recognition, speech synthesis or data mining, for instance, open up entirely new approaches to designing operating models.

Big Data and Artificial Intelligence enable overall more efficient and more effective operating and business models for financial services providers. But only those companies that manage to grow the necessary technical and specialist expertise will be able to develop and implement such models. Financial services providers are confronted with the challenge of using these new technologies in this environment – and especially under the influence of the General Data Protection Regulation (GDPR) – while at the same time maintaining a fundamental level of customer confidence.

Professor Wrobel, do you share this view?

Wrobel: Yes, from where I stand, the study clearly reveals the tremendous potential that the evolution of Big Data towards the topic of Artificial Intelligence, or AI for short, entails for the financial and insurance industry. Companies and institutions will be able to further optimise their products and services by strategically deploying AI technologies, make processes more transparent and more effective, and hence reinforce the trust of their customers. Using Big Data and AI is therefore of crucial, strategic importance for the competitiveness of the industry. I also consider three aspects that have already been mentioned to be particularly relevant and noteworthy from a technological perspective: data sovereignty, transparency and credibility.

Would you like to explain this briefly?

Wrobel: .The study underlines in particular that the successful, socially accepted integration of Big Data and Artificial Intelligence in the financial and insurance industry can only work if the companies incorporate these three aspects into data-driven products and services, and implement them technically with a high level of professionalism. This insight is evident in many ways in the different development phases of a value chain of data-driven offerings – starting, for example, with data gathering or data generation.

In this context, the study underscores the fundamental importance of data sovereignty for the entire value chain in order to reliably offer and maintain data-driven products and services, verify the reliability of the underlying data at any time and also safeguard sovereignty over information security and data protection.

In addition, the study suggests indispensable standards that must be applied when artificial intelligence technologies are used. It is clear that a high degree of transparency of the AI methods used is just as essential as the reliability, plausibility and traceability of the algorithms used. Transparency in particular plays a crucial role, not least because of data protection or, for example, preventing algorithms from discriminating against individual customer groups. In this case, the certification of algorithms or data-driven products and services can create trust at both economic and social levels.

Mr Wechselmann, do you see this similarly from your perspective?

Wechselmann: .The sort of fundamental changes that are driven by the increased use of new technologies pose challenges not only to companies, but also to public administrations. BaFin is therefore quite rightly addressing, at an early stage, the potential opportunities and risks not only for the companies that it currently supervises, but especially also for consumers.

Besides the numerous opportunities described in the study, the risks involved with the more intensive use of BDAI must be analysed and addressed by appropriate measures. Although the focus of the study is on the financial sector, some findings can certainly also be applied to other areas.

Mr Hufeld, how do you rate the results?

Hufeld: nThe most important thing is for the regulatory regime to keep pace with the speed at which innovations are emerging in this area. It is a challenge to create a regulatory framework that is forward-looking and sufficiently flexible to be applicable to other rapid innovations in the field of Big Data. However, the supervisory principles must not be so pliant that they cannot offer any security to the companies operating in this area.

Our principle continues to be “same business, same risk, same rules”. Supervision must follow an approach of technological neutrality. A more principle-based regulatory approach is particularly suitable in this environment. For us at BaFin, this all means developing new specialist and technical expertise.

What steps will BaFin take next?

Hufeld: We have now issued the study for consultation. We want to use it as the basis for in-depth dialogue with the industry, as well as other national and international supervisory authorities. This sort of open exchange in the national and international environment has proven its worth – including in other major issues – and forms the basis for reassessing regulatory requirements where necessary.

Please note

This article reflects the situation at the time of publication and will not be updated subsequently. Please take note of the Standard Terms and Conditions of Use.

Additional information

Did you find this article helpful?

We appreciate your feedback

Your feedback helps us to continuously improve the website and to keep it up to date. If you have any questions and would like us to contact you, please use our contact form. Please send any disclosures about actual or suspected violations of supervisory provisions to our contact point for whistleblowers.

We appreciate your feedback

* Mandatory field