Is the Robodoc in? : a qualitative study on the state of AI implementation in Swiss hospitals

Background: The number of publications dealing with the utilization of artificial intelligence (AI) in healthcare is exploding. Hopes are high that this will be a contribution towards increasing quality and efficiency. Yet little is known about current implementation levels in Swiss hospitals and practitioners’ needs. Research question: This paper aims to understand the current state of the leading Swiss hospitals on their AI implementation journey. Furthermore, we want to shed light on practitioners’ utilization of and expectations regarding AI. Method: We conducted eight interviews with hospital AI experts. The interviews were conducted using semi-structured guides and focused on (a) the current AI implementation maturity level, (b) the current and future potential of AI, and (c) enablers and hurdles along the transformation path. Findings: (a) The Swiss hospitals that took part in our study are still in a very early stage of AI implementation, despite having a reputation of being among the leading Swiss healthcare organizations. (b) The interviewees saw very high potential for AI to increase outcome quality, a rather high potential for reducing the administrative burden, yet only a moderate potential for reducing costs in the next 10 years. (c) The experts confirmed the typical 2 hurdles to implementation like lack of funding, technological issues and a lack of clearly distributed roles in their organizations to drive the implementation forward. Discussion and research propositions: Three main research propositions were identified. (I) As the urgency to implement AI is currently still very low, technology providers need to better understand the actual needs of practitioners. (II) As practitioners are only moderately positive about the contribution of AI to reducing costs, we need solid economic evaluations of AI-implementations to quantify their effective impact. (III) Research on the ideal organizational setup for AI implementations is still scant, and this should be explored in more detail in the future