Artificial Intelligence in the Luxembourg Financial Sector

This article was originally published in Agefi Luxembourg, March 2021 and is reproduced with permission from the publisher.

Artificial intelligence (AI) based technologies are nowadays used in the financial sector as part of the financial services provision in Luxembourg. These advantageous solutions are acknowledged and supported by the EU’s work towards a suitable legal and ethical framework. The World Economic Forum has published its forecasts stating that the impact of AI in the financial sector will continue to grow and develop drastically, bringing new opportunities to market players, among others, for such outside of the financial sector in respect of distribution of financial products and advice.¹ The goal of ensuring efficiency, time and cost savings, quicker decision-making process and more favorable experiences of clients in the financial sector is carried by the evolving AI.

What is AI?

AI is defined as intelligence demonstrated by machines that overtake tasks which should normally have been performed by humans.² The possibilities of AI solutions range from machine learning through the use of self-learning computer algorithms to the prediction, reasoning or perception.³

AI in Luxembourg

Luxembourg is famous for its innovative mindset, recognition of digitalization and technologies. Having its wealth of international and cross-border datasets with potential commercial value for a number of industries, it is an attractive option for cutting-edge technology developers. Luxembourg's goal is to be among the most advanced digital societies in the world, in particular in the EU, but it is already a living AI lab with global influence.⁴ Besides of developing the AI solutions in Luxembourg, it is also part of Luxembourg's plans to design an attractive and living environment for data-driven and intensive services and activities relating to AI.⁵

The governments’ approach in relation to AI is human-centric, aiming for the development of AI alongside the society and its needs by keeping in mind the importance of ensuring that AI remains understandable, transparent and trustworthy.⁶ In addition, the economic gain should be noted as it is estimated that AI could contribute to 13 point 3 trillion Euro to the global economy in 2030, while the global gross domestic product could be up to fourteen per cent higher in 2030 as a result of AI.⁷

The data-driving innovation strategy of the Luxembourg Ministry of Economy has developed a catalogue of action points for the purposes of supporting new data-driven business models in economic sectors of high priority, including the financial services. Such action points are covering, amongst others, the goal to identify innovation regulation with respect to data marketplaces in order to increase legal certainty and transparency of data economy participants, working on projects to enhance the quality and accessibility of data by supporting the data infrastructure development and attracting data-driven and data-centric services and businesses to Luxembourg by creating an innovative and trusted regulatory environment. Luxembourg is promoting the Ethics Guidelines for Trustworthy AI released by the Commission's High-Level Expert Group⁸, while Digital Luxembourg⁹ supports a number of projects contributing to Luxembourg's AI ecosystem.¹⁰ While the European Commission is stressing that AI applications should respect seven key requirements to be considered trustworthy: human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination and fairness; societal and environmental well-being; and accountability¹¹, much investigation and development needs to be performed in order to transform this idealistic concept into reality.

Opportunities

Having in mind the multiple advantages AI can bring to entities dealing with or providing financial services, it is a fact that the financial sector is making use of prediction as forward-looking analyses are essential for most of the financial entities. The accuracy of the required prediction remains subject to the receipt of data whose quality is a question of cost, noting that machine learning algorithms for example need much more data than humans in order to train for certain skills which are comparable to human's skills. AI solutions promote to lower the costs of making predictions leading to better risk management and profitability.¹² Reduced costs for both, consumers and financial institutions, can make advice more affordable for consumers, while an easy access to more products and services could attract a wider range of customers with a quality of the service that is provided by AI in a potentially more consistent way than human advice.¹³ AI can facilitate the invoicing processes and post-trade settlement or ensure fraud detection and assistance with anti-money laundering. It has also introduced possibilities on imagine recognition and automated document analysis for customer complaints or virtual agents and chatbots available 24/7 for immediate replies.¹⁴ AI can also ease the due diligence process on an acquisition. Please see below four main cases which show how interesting AI can be.

Robo-advice

One of the AI applications in place are robo-advisors that provide client advice such as for instance in relation to proposed investments and ensure an access to investment advice for a larger customer base and therefore favoring inclusion. The CSSF confirmed that there is no commonly accepted definition of robo-advice and that it is used as a wide or undefined catch-all term for all sorts of online investment advisors using computer-based technologies.¹⁵ Current versions of robo-advisors exist with only limited capabilities by being able to allocate to clients one among a set of predefined portfolio based on the data obtained via a questionnaire. However, there are also enhanced versions which are developed and are able to include dynamic portfolio optimization features relying on machine learning models.¹⁶ This AI is demonstrating a more user-friendly aspect being an at all-time accessible service, allowing for more favorable costs due to the excluded human interaction.¹⁷ There is no specific EU regulation for robo-advice at EU level. Due to the constant development and associated risks related thereto, several EU Member States have decided to develop domestic regulatory frameworks relying on EU and domestic legislation, while in Luxembourg any digital financial advice services are subject to the CSSF’s regulatory requirements.¹⁸

In terms of licensing for a robo-advisor, the CSSF refers to the different statuses available in the law of 5 April 1993 on the financial sector, as amended (LFS), all that have to be compliant with the MiFID/MiFIR framework:¹⁹

• an investment adviser, Article 24 of the LFS that also applies to any non-automated financial advisor limited to providing advisory services and which do not intervene in the implementation of the advice provided by it;
• a broker in financial instruments, Article 24-1 of the LFS that is an intermediary by either encouraging parties to be brought together with a view to the conclusion of a transaction or in passing on its clients’ purchase or sale orders without holding the investments of the clients;
• a commission agent, Article 24-2 of the LFS should the robo-advisor execute orders on behalf of the clients and in relation to one or more financial instruments;
• a private portfolio manager, Article 24-3 of the LFS that uses the technology to manage portfolios as per the client’s mandates on a discretionary client-by-client basis.

Chatbots

Chatbots are virtual assistants whose task would be to answer common frequently asked questions to the institution's clients. As long as the scope of the chatbot remains limited and controlled by a human being after the interaction of the client with the chatbot, such AI can be considered as very helpful and a low risk tool.²⁰

Fraud detection and money laundering

A very frequently used type of AI that already brought positive feedback is the fraud detection and money laundering investigation assistance in financial institutions. This AI has a supervised machine learning algorithm, can identify the past frauds and use this information to detect new frauds with a higher accuracy rate in order to be of use for the manual verification by the compliance officers which do not need to cross-check all of the possible cases. AI is used in form of unsupervised machine learning, i.e. the data is grouped into clusters according to different criteria to identify individuals or transactions not sharing the same common patterns with the rest of the population and which represent anomalous behaviour.²¹

Automated credit scoring based on machine learning

Furthermore, there is also credit scoring based on machine learning that is used for automating loan decisions for corporate and retail customers. This AI is providing better overviews on cash flows and more accurate estimations of the credit risk. This process is possible due to the fact that the Directive (EU) 2015/2366 of the European Parliament and of the Council of 25 November 2015 on payment services in the internal market (PSD 2), as implemented in the Grand Duchy of Luxembourg through the law of 10 November 2009 on payment services, as amended, requests financial institutions to share client transactional data with other financial sector players (only with the prior consent of the client).²² For the sake of completeness, it should be noted that the CSSF has updated its frequently asked questions on the statuses of "PFS" - Part II on 2 September 2019 relating to regulated professionals of the financial sector stating that any such professional must comply with its obligation of professional secrecy, and in case of a data transfer from one service provider to another service provider, must obtain the consent from its clients through the board of directors for the outsourcing of the services in scope, the type of information to be transmitted in the context of outsourcing and the country of establishment of the entities providing outsourced services pursuant to article 41(2a) of the LFS.²³ Automation in the securities, banking and insurance sector can assist in such fields for example on matters related to the Markets in Financial Instruments Directive II²⁴ when speaking about the securities sector, the Insurance Distribution Directive²⁵ on insurance products distributed or PSD 2 on banking products.²⁶

Risks / Challenges / Regulation

The successful implementation of AI solutions in the financial sector would require the regulator to leave enough space for the technologies to develop. Space can, however, also lead to legal risks and uncertainties.²⁷ Regulated entities which have to comply with many legal and reporting requirements could be at risk to become subject to additional supervisory obligations so that a regulation of the means by which the service is provided may double the regulatory requirements.²⁸ The best, but most difficult exercise would be to find a regulatory framework that ensures security, certainty and protection, but also maximizes the benefits of AI, taking into account the fast development of the technologies, the lack of thorough and systematic understanding of its impacts and associated business models, and the unpredictability of its evolution for appropriate legal provisions.²⁹ It is likely that the additional legal requirements would apply for high-risk AI, while for any low-risk AI, an option may be contemplated to establish a voluntary labelling scheme.³⁰ The risks AI tools bear have to be considered and dealt with now and in future. While chatbots, for example, are very useful to ensure an access to client's help around the clock, the related risk can become serious, for instance, where the chatbot can automatically learn new knowledge from the user interactions, like in the example of Microsoft that removed its chatbot Tay from Twitter after only a couple of hours since it learnt racist behavior from other users.³¹

Robo-advisors are considered as one of the main applications of AI in the financial services.³² They should be used in connection with precautionary mechanisms in place to suspend the provision of advice in case mistakes or bias would be detected. The CSSF is recommending the financial institutions using this AI to monitor the effectiveness and appropriateness of the provided advice to avoid a mis-selling, if for example the robo-advisor would favour investment funds with higher commissions. The wrong applicability of this AI could have serious consequences, such as leading to financial stability risk.³³

Taking into consideration automated loan decisions, and in particular the lower operational costs and faster credit decisions for the clients, bias have to be avoided by such model, such as for example were a population is not represented in a sufficient manner in the input dataset used to train the model may lead to discrimination. An example for this could be that married women in Luxembourg were not allowed to hold a bank account or loan under their name without the husband's signature which was required and due to this, historical data as representative of the credit capacity of women could lead to wrong conclusions.³⁴ The changes brought to our society and daily life during COVID-19 are that certain aspects that a machine learning mechanism could misinterpret for future transactional points.

Conclusion

AI has without any doubt a huge potential in the daily life's development of any financial institution. Nevertheless, many concerns are linked to this innovation which could lead to, amongst others, legal, ethical, reputational and financial consequences. Luxembourg supports innovation and promotes digitalisation and the use of AI subject to future observations and expected guidance. The protection of client's personal data and the arising liability need to be monitored, developed and thought-through as well as the fact that where we seek to replace human interaction by AI, there are certain types of decisions which just need to be taken further to a human judgment in order to ensure AI responsibility.

¹ The New Physics of Financial Services - Understanding how artificial intelligence is transforming the financial ecosystem, World Economic Forum (WEF) (2018); Artificial Intelligence in Europe, Outlook for 2019 and Beyond, Ernst & Young (2018); Artificial Intelligence: a strategic vision for Luxembourg, Digital Luxembourg, Vision paper, May 2019, p. 7.
² European Financial Stability and Integration Review (EFSIR) 2019, Commission Staff Working Document, Chapter 4, p. 78; Artificial Intelligence, White Paper, CSSF, p. 5.
³ European Financial Stability and Integration Review (EFSIR) 2019, Commission Staff Working Document, Chapter 4, p. 78; A Definition of AI: Main Capabilities and Disciplines, European Commission, 8 April 2019, p. 6.
⁴ Artificial Intelligence: a strategic vision for Luxembourg, Digital Luxembourg, Vision paper, May 2019, p. 5.
⁵ White paper on artificial intelligence, ILNAS, Artificial Intelligence: a strategic vision for Luxembourg, p. 4, Digital Luxembourg, Vision paper, May 2019, p. 5.
⁶ Artificial Intelligence: a strategic vision for Luxembourg, Digital Luxembourg, Vision paper, May 2019, p. 6, 9; Artificial Intelligence for Regulatory Compliance: Are We There Yet?, Journal of Financial Compliance, (3(1), 1-16, T. Butler and L. O’Brien (2019); 30 Recommendations on regulation, innovation and finance, Final Report to the European Commission from the Expert Group on Regulatory Obstacles to Financial Innovation (ROFIEG), December 2019, p. 31.
⁷ Artificial Intelligence: a strategic vision for Luxembourg, Digital Luxembourg, Vision paper, May 2019, p. 6.
⁸ An independent group mandated with the drafting of two deliverables: (1) AI Ethics Guidelines and (2) Policy and Investment Recommendations.
⁹ Digital Luxembourg is a government initiative charged with unifying and strengthening the nation's digitalisation efforts.
¹⁰ Artificial Intelligence: a strategic vision for Luxembourg, Digital Luxembourg, Vision paper, May 2019, p.18 - 19.
¹¹ Coordinated Plan on Artificial Intelligence, European Commission, COM(2018) 795 final, 7 December 2018.
¹² European Financial Stability and Integration Review (EFSIR) 2019, Commission Staff Working Document, Chapter 4, p. 79;
¹³ Final report on automation in financial advice, EBA BS 2016 422 (JC SC CPFI, p. 4; Global AI Survey: AI proves its worth, but few scale impact, McKinsey & Company, November 2019, p. 2.
¹⁴ The New Physics of Financial Services - How artificial intelligence is transforming the financial ecosystem, World Economic Forum (WEF) (2018); Ernst & Young (2018), Artificial Intelligence in Europe, Outlook for 2019 and Beyond; Artificial Intelligence - White Paper, CSSF, December 2018, p. 6.
¹⁵ Robo-advice, CSSF’s position on Robo-advice, 27 March 2018, p. 1.
¹⁶ Artificial Intelligence, White Paper, CSSF, p. 41.
¹⁷ Votre conseiller en investissement est un robot : quel impact sur la protection du client de la directive 2014/65/UE, Les services financières dans un monde digital, 2019, p. 61, Isabelle Riassetto.
¹⁸ Robo-advice, CSSF’s position on Robo-advice, 27 March 2018, p. 2.
¹⁹ Robo-advice, CSSF’s position on Robo-advice, 27 March 2018, p. 2.
²⁰ Artificial Intelligence, White Paper, CSSF, p. 40.
²¹ Artificial Intelligence, White Paper, CSSF, p. 42.
²² Artificial Intelligence, White Paper, CSSF, p. 43.
²³ FAQ statuses "PFS" - Part II, Version of 30 October 2020, Question 20, p. 11.
²⁴ Directive 2014/65/EU of the European Parliament and of the Council of 15 may 2014 on markets in financial instruments.
²⁵ Directive (EU) 2016/97 of the European Parliament and of the Council of 20 January 2016 on insurance distribution (recast).
²⁶ Report on automation in financial advice, Joint Committee of ESAs report, 16 December 2016, p. 4.
²⁷ ESMA Report on Trends, Risks and Vulnerabilities No.1, ESMA, 2019, p. 32; An executive’s guide to real-world AI – Lessons from the Front Lines of Business, White Paper, Harvard Business Review Analytic Services, p. 8.
²⁸ European Banking Federation (EBF)'s response to the European Commission's Consultation on FinTech: A more competitive and innovative European Financial Sector, EBF, 15 June 2017, p. 11; Policy and Investment Recommendations for Trustworthy AI, European Commission, 26 June 2019, p. 38.
²⁹ Policy and Investment Recommendations for Trustworthy AI, European Commission, 26 June 2019, p. 38.
³⁰ White Paper on Artificial Intelligence – a European approach to excellence and trust, European Commission, 19.02.2020, p. 18, 22.
³¹ Artificial Intelligence, White Paper, CSSF, p. 40.
³² Frontiers in Artificial Intelligence, November 2018, Volume 1, Article 1, p. 3.
³³ Artificial Intelligence, White Paper, CSSF, p. 41; What is intelligence without trust?, Ernst & Young, N. Duffy and C. Cobey, 11 April 2019, p. 2.
³⁴ Artificial Intelligence, White Paper, CSSF, p. 44; How do you teach AI the value of trust? Ernst & Young (2018), p. 3.