Systems based on various flavours of artificial intelligence (AI) are increasingly being used in investment banks to provide clients with financial advice, manage investments and automate customer interactions in ways that are both scalable and cost-efficient. According to a joint report, published in November 2024 by the UK’s Financial Conduct Authority (FCA) and the Bank of England, 75% of firms are already using AI, with a further 10% planning to use AI over the next three years. There are, however, ethical concerns around how AI systems interact with clients, especially when it comes to ensuring the systems do not act against their best interests – even if this does not optimise the bank’s profitability and / or efficiency.
By GreySpark’s Rachel Lindstrom, Senior Manager
The ethical use of AI in client interactions in investment banking is not just about improving the efficiency of systems but also meeting optimal outcomes for clients. AI systems that include client touchpoints must carefully balance personalisation, transparency, data protection and fairness. Ethical considerations must be built into their design, implementation and ongoing operation to ensure that AI is a tool for positive change rather than something that undermines trust or harms clients’ interests in the long run.
The EU Artificial Intelligence Act (EU AI Act), which is the first comprehensive regulatory framework for AI in the world, has significant implications for how investment banks and other financial institutions use AI, especially when interacting with clients. The entry of the Act into force has been staggered, beginning in December 2024 and ending by July 2026.
The Suitability of AI for Client Interactions Across the Bank
Client touchpoints with banking systems or personnel are found across investment banks. This means that the possibilities for AI to manage those interactions are vast. Indeed, as firms seek to streamline their processes and increase efficiencies the introduction of AI to manage many interactions is highly likely. Currently, client interactions often involve a mix of human advisors, automation and some already implemented AI-driven tools.
Design of AI-based Systems
Many AI systems are designed to optimise outcomes based on specific metrics — such as increasing the firm’s revenue, pushing high-commission products or generating more trades. This can create a situation where the AI might recommend a course of action to a client that benefits the bank more than it does the client. The question is how to ensure AI systems adhere to a hierarchical list of priorities – namely, how they would deal with a scenario where considerations on the prioritisation list are in direct conflict with one another. Many AI systems, particularly those using complex machine learning models, operate as ‘black boxes’ and clients may not fully understand how the AI comes to its recommendations. Transparency is crucial, therefore, for trust and clients must feel confident that the AI is not making decisions based on opaque criteria.
Use of Data and Algorithms by AI-based Systems
When clients interact with AI-powered financial tools, it is important that they understand exactly how their data is being used and the risks involved with providing the data for the tool to use. AI systems in investment banking depend heavily on personal data to function effectively. Such data must be handled so that it is not misused or exposed to unauthorised parties.
Informed consent is a fundamental principle in ethics and AI systems must ensure that clients fully understand how their data will be used and what outcomes to expect from AI-based decisions in order to avoid negative outcomes. For example, AI systems often use broad algorithms that aim to personalise advice based on limited data. The degree of personalisation may vary as a consequence. Additionally, if the system is too rigid, it may deliver advice that is not fully tailored to the client’s requirements.
Beyond just regulatory compliance, ethical concerns about privacy are critical. The risk of cyberattack is faced across the financial services, but the newness and complexity of AI-based systems means that clients need assurance that their personal financial data will be not only protected, but also only used for the purposes to which they have agreed. More than ever, investment banks must ensure they are transparent about data usage policies and security measures.
Avoiding, Overcoming or Identifying Intrinsic Biases
If an AI system is trained on biased data or uses certain patterns in the data that inadvertently discriminate against certain groups, inequality could be perpetuated. An AI system that recommends loans or investment opportunities could be biased against individuals from certain socio-economic backgrounds, races, or genders if the historical data it was trained on reflects past discrimination in those areas. It is vital that AI systems are carefully monitored and tested for biases that could result in discriminatory outcomes. If AI systems are unintentionally reinforcing societal inequalities, the financial services industry could be perpetuating or worsening existing disparities.
As such, it is vital that the data upon which the AI model is trained is cleansed and verified, which will often require some human oversight. Those designing or purchasing AI for use within the bank must take mitigating strategies and these are illustrated in Figure 1.
Figure 1: Pyramid of Protection Enveloped by Regulatory Oversight
Source: GreySpark analysis
(Click image to enlarge)
While AI systems must be designed with explainability and interpretability when it comes to eliminating biases, regulation must be robust. Regulators and lawmakers must understand the technologies and potential impacts to ensure investors are protected and markets remain efficient. In 2024, the EU Council introduced the world’s first comprehensive framework for the operation of AI within its borders. The EU AI Act is not financial services-specific, but is highly relevant for financial services firms. The question disrupting the sleep of AI users and designers everywhere is what long-term this regulation will have on their business.
The EU Artificial Intelligence Act
The EU Artificial Intelligence Act (AI Act) applies to all investment banks operating in the EU or providing services to EU clients, as well as those who may develop or deploy AI technologies for financial services within the EU market. Essentially, any financial institution that uses AI systems in high risk areas such as client-facing financial services, trading, risk management, or regulatory compliance will be impacted by the regulation. The categorisations of areas for firms to consider are shown in Figure 2.
Figure 2: Key Aspects of the EU AI Act
Source: GreySpark analysis
(Click on image to enlarge)
The AI Act classifies AI systems into different risk categories: unacceptable risk, high risk, limited risk and minimal risk (see Figure 3). Investment banks are primarily concerned with high-risk AI systems, which include any AI systems that have a significant impact on the safety, rights or freedoms of individuals. If an investment bank uses AI in ways that directly influence clients’ financial decisions, those AI systems will fall under the high-risk category and be subject to more stringent regulations designed to protect consumers.
Figure 3: AI-based Risk Levels
Source: GreySpark analysis
(Click on image to enlarge)
The AI Act classifies AI systems into different risk categories: unacceptable risk, high risk, limited risk and minimal risk (see Figure 3). Investment banks are primarily concerned with high-risk AI systems, which include any AI systems that have a significant impact on the safety, rights or freedoms of individuals. If an investment bank uses AI in ways that directly influence clients’ financial decisions, those AI systems will fall under the high-risk category and be subject to more stringent regulations designed to protect consumers.
1. Transparency and Explainability
The AI Act places strong emphasis on transparency and explainability for high-risk AI systems, which is crucial in the sense that clients need to trust that AI-driven decisions are fair and based on clear, understandable criteria. Investment banks using AI to make trading decisions, for example, must be able to explain how recommendations were created. If an AI-powered bot were to recommend an investment strategy, the bank must ensure the system can explain why that recommendation is suitable for that client and its financial goals, risk tolerance and market conditions. Clients will have the right to understand the logic behind recommendations.
2. Data Usage and Privacy Protection
The AI Act requires that AI systems, particularly those in the high-risk category, adhere to strict data governance standards. This means ensuring that any personal data used by the AI system is handled in accordance with data protection laws such as the GDPR. The AI Act strengthens protections around how data is collected, processed and used in high-risk AI systems, ensuring clients’ privacy rights are upheld. Investment banks must ensure that clients give explicit, informed consent regarding how their data is used in AI systems. Clients will also have more control over their data and can request to see what data is being used to train or inform AI models.
3. Accountability and Liability
The AI Act introduces a framework for accountability, placing responsibility on the developers and operators of AI systems. Investment banks that deploy AI technologies will be held accountable for any harm or negative consequences caused by their AI systems, including those affecting clients’ financial wellbeing. If an AI system makes a wrong decision (for example, recommending an unsuitable investment strategy or approving a loan based on biased data), clients will have avenues for recourse. Banks will need to ensure that there is accountability for AI decisions, and they may need to offer remedies or compensation in cases where clients are harmed by these systems.
4. Risk Management and Continuous Monitoring
High-risk AI systems must be subject to ongoing risk assessments and continuous monitoring to ensure they are functioning as intended and that they do not harm clients or lead to unfair outcomes. Investment banks will need to demonstrate that they continuously monitor the performance of their AI systems, including conducting regular audits to ensure the AI is operating fairly and in line with ethical standards. If a problem is detected (such as biased decision-making or inaccurate recommendations), the system must be updated or corrected to prevent client harm. Part of this monitoring involves ensuring that the AI does not produce biased or discriminatory outcomes. If so, the bank must take corrective action.
5. Human Oversight
Even with high-risk AI systems, human oversight is mandatory. The AI Act mandates that AI systems should not be fully autonomous in the high-risk categorisation and there should always be the possibility of human intervention. For AI systems used in client interactions, clients will have the right to request human intervention or review if they feel uncomfortable with the AI’s recommendations. This helps ensure that human judgment is applied in critical financial decisions, especially in cases where the AI system might be making complex or high-stakes choices.
6. Regulatory Compliance and Penalties
Finally, the AI Act establishes penalties for non-compliance, including significant fines for non-compliant organisations. This provides a level of consumer protection, ensuring that AI systems cannot be used recklessly or in ways that harm clients. Clients benefit from the increased regulatory scrutiny and the incentive for banks to develop AI systems that adhere to ethical standards.
Regulatory Protection at the New Frontier
For investment banks, the EU AI Act aims to ensure that AI systems are transparent, explainable, accountable and fair. The consequences of the Act are enforceable for non-EU clients in an EU bank as well as for EU clients in a non-EU bank. The Act places responsibility on banks to ensure that their AI systems respect clients’ privacy, provide clear reasoning for decisions, avoid biases and are continuously monitored for risks. The Act means that clients now have the option to consent or decline participation in these AI systems with full knowledge of the potential risks and benefits. In addition, they benefit from stronger protections against AI-driven harm, with clearer avenues for recourse if something goes wrong.
When clients decline to take up services that incorporate AI models, they may miss out on the value add provided by the AI. However, for some, accepting the risk associated with a greater reliance on AI is a better approach, even if it does come at the expense of quality of service.
Ultimately, the Act is designed to ensure that AI technologies in the financial sector are used in a way that prioritises client safety, fairness and trust. For investment banks, complying with these requirements will be essential, while also providing an opportunity to build client trust by demonstrating a commitment to ethical AI practices.
