Navigating Managerial Control and Oversight in the Age of Agentic AI : A Qualitative Study of AI Governance in the Swedish Banking Sector

Sammanfattning

The rapid development of agentic artificial intelligence is creating new opportunities and challenges for the banking sector. While AI-supported systems can improve efficiency, speed, documentation, customer communication, and analytical capacity, they also create questions about managerial control, oversight, accountability, and explainability. This is especially important in banking, where decisions may affect customers’ money, personal data, access to financial services, credit, security, and institutional trust. The purpose of this thesis is to investigate how financial institutions and their strategic partners navigate the challenges of managerial control and oversight when authority is delegated to agentic AI. The study focuses on internal and ecosystem-based professional perspectives, including banking professionals and actors with experience of AI-supported banking processes. The research question guiding the study is: How do financial institutions and their strategic partners navigate the challenges of managerial control and oversight when delegating authority to agentic AI? A qualitative research design was used, based on semi-structured interviews with seven respondents who had experience of, or insight into, AI-supported banking processes. The respondents represented different perspectives, including IT implementation, AI consultancy, customer service, pension advice, banking system project management, corporate advisory work, and digital transformation management. The interview material was analysed through thematic analysis. The theoretical framework combined Agency Theory and the Resource-Based View. Agency Theory was used to understand delegation, information asymmetry, monitoring, and accountability, while the Resource-Based View was used to explain why responsible AI governance matters for protecting trust, reputation, decision quality, and organisational capability. The findings show that AI is mainly used as a support tool rather than as a fully autonomous decision-maker in high-risk banking areas. AI is accepted for routine, repetitive, administrative, analytical, and low-risk tasks, but human oversight remains necessary when decisions involve fraud, credit, compliance, identity verification, account access, large payments, customer security, or organisational accountability. The study identifies six main themes: limited AI autonomy, human oversight and accountability, control mechanisms and traceability, explainability and trust, the changing managerial role, and the balance between efficiency and control. The thesis concludes that financial institutions and their strategic partners navigate the challenges of AI oversight and control through risk-based AI governance. Managerial control does not disappear when AI becomes more present in banking workflows. Instead, it changes form. Control becomes less dependent on direct supervision of every task and more dependent on structured governance mechanisms such as clear boundaries, human approval, escalation routines, audit trails, monitoring systems, explainability requirements, and managerial competence. Therefore, responsible AI governance should be understood not only as a technical or compliance issue, but also as a managerial and strategic capability in the banking sector.