Lita Lagom: : En kvalitativ studie om controllers tillit till AI-genererade beslutsstöd

Sammanfattning

This study examines how management accountants (controllers) experience and reason about their trust in AI-generated decision support, and which factors influence this trust in different decision-making situations. The background contains the shift of the controller role toward a more analytical and strategic business partner, where AI tools bring a tension between algorithm aversion and automation bias. The study is guided by two research questions: What factors influence controllers' trust in AI-generated decision support?How does controllers' trust in AI-generated decision support vary depending on the decision-making situation?To answer these questions, a qualitative research strategy was applied, using an interpretive design, a cross-sectional design, and an abductive approach. The material was collected through nine semi-structured interviews with controllers in both the public and private sectors, and was analysed through a thematic analysis. The theoretical framework draws on trust as a dynamic and experience-based process, as well as on the concepts of algorithm aversion, automation bias, explainable AI (XAI), and the distinction between perceived and actual trust. The results show that professional knowledge constitutes the fundamental prerequisite for trust, as controllers use their own knowledge as a filter through which the AI tool's output is tested before it is accepted or rejected. Trust is built as a dynamic process, where individual negative outcomes limit the tool's area of use rather than destroying trust entirely. Trust is higher for objective and data-driven tasks and decreases for tasks perceived as subjective, where the boundary is determined by the controller's own interpretation. The majority do not request advanced explanatory models; instead, source references function as a sufficient form of transparency. Finally, trust is influenced by sectoral context, where the public sector is characterised by institutional caution while the private sector is characterised by business logic. The study contributes by showing how algorithm aversion, automation bias, and XAI are connected through the controller's professional knowledge, and highlights a paradox in which the controllers who benefit most from AI are often those who need it least. Since protection against automation bias is tied to the individual rather than the role, organisational guidelines appear necessary.