Visualization
Change Management & AI Adoption
+
Systematic approach for sustainable AI adoption: adoption barriers (trust gap, automation bias), success factors and measurable KPIs for organizational change.
Technology alone does not transform an organization. Successful AI implementation in healthcare rarely fails because of technology - but rather because of trust deficits, a lack of digital literacy, cultural resistance and unclear governance. This visualization and the associated concept systematically address this gap.
The special thing about healthcare: The consequences of poor AI adoption are directly relevant to patients. Alarm fatigue due to uncalibrated systems, automation bias[1] in medical decisions and unclear liability for AI errors are real risks - not theoretical ones.
Digitalization means, above all, cultural change – that is my main practical experience. The most common reaction to impulses for change: "Yes but..." I know this resistance well - and I know how to address it constructively, take it seriously and transform it into real adoption.
(Clinical) employees do not accept AI systems if their functionality remains opaque. Explainability (XAI) is a prerequisite for adoption - not an additional feature.
Without basic AI knowledge, employees cannot assess strengths and limitations. Training is not a nice-to-have, but a safety requirement.
AI introduction changes workflows and role models. Involve clinical champions early on – change management begins before the first pilot.
Even well-trained employees adopt AI recommendations uncritically.[1] Override rate monitoring and calibration training are mandatory.
Start with LOW and MEDIUM risk applications that quickly show added value. HIGH-risk systems only after a proven compliance structure and human-in-the-loop processes.[4]
Explainable AI (XAI) builds trust – especially in clinical settings. Transparent models are used more frequently and more responsibly than black box systems.[2]
Internal multipliers accelerate adoption and increase acceptance. Change management begins before the first pilot - not after the go-live.
Override rate, alarm fatigue, frequency of use and clinical outcomes must be systematically recorded. What is not measured is not controlled.[3]
Roles, responsibilities and escalation paths must be defined before deployment. For high-risk AI systems, the EU AI Act requires human oversight — and it is ethically imperative.
Own representation and conceptual framework: Kawaschinski K., Karolinska Institutet, Spring 2026. Sources checked as of: April 2026.
This material was prepared with the greatest care. It does not replace legal, medical or professional advice. No warranty for completeness or timeliness.
Parts of this document were created with the support of generative AI and editorially reviewed. This content is not a substitute for legal, medical or professional advice. (EU AI Act Art. 50)