Exploring the Swedish AI Adoption Gap : Generative AI Integration in SMEs from Experimentation to Organisational Use through Reskilling and Organisational Adaptation

Sammanfattning

Generative AI is becoming part of everyday business life, but for many smaller firms, its use in organisational practices remains challenging. While tools such as ChatGPT and Microsoft Copilot are easy to access, the harder question is how SMEs move from individual experimentation to more stable and useful organisational use. This thesis focuses on that in-between stage, where GenAI is already present in daily work, but its role and value are still being shaped. The purpose of this study is to understand how Swedish SMEs experience the transition from early GenAI experimentation to more regular organisational use. The study follows a qualitative and exploratory design based on seven semi-structured interviews with participants from micro, small, and medium-sized firms in Jönköping County. The analysis was conducted abductively, connecting the interview findings with organisational learning theory, the 4I framework, while allowing new empirical themes to emerge from the data. The findings show that GenAI integration in SMEs depends less on access to technology and more on how people learn and adapt to these tools in everyday work. A central finding is the presence of a learning flow gap, where individual experimentation does not automatically move into shared routines or organisational practices. Participants learned mainly through trial and error, output checking, informal sharing, and task-specific experimentation. Managerial support helped legitimise experimentation, but integration remained selective and was limited by time pressure, trust concerns, unclear responsibilities, limited structure and the lack of dedicated roles for AI implementation. Overall, the study shows that GenAI adoption in SMEs is gradual, uneven, and selective. Rather than being fully embedded across the whole organisation, GenAI becomes useful in specific tasks or workflows. This thesis concludes that the main challenge is not simply adopting GenAI but creating the internal learning conditions that allow experimentation to become shared and repeated in practice.