What If the Biggest Barriers to Diagnostic AI Adoption Have Little to Do with the AI Itself?

By Indrajeet Ghatge, Head of Behavioural Science

As diagnostic AI becomes increasingly sophisticated, a more consequential question is emerging: What determines whether clinicians actually embrace it in practice?

Much of the conversation around diagnostic AI has been focused on whether algorithms can detect disease accurately. Regulatory clearances, validation studies, and vendor marketing all address this question. However, buyers are now asking a different question. Why does a tool that performs well in validation fail to move throughput, quality, or cost at the institution level?

This answer lies between clinical behaviour, workflow design, procurement logic, and reimbursement. Yet very little research addresses these factors directly, and almost none brings together the perspectives of clinicians and buyers evaluating the same tools. The answer may be more nuanced than technology performance, clinical accuracy, or even trust.

At Phronesis Partners, we're exploring these often-overlooked human and organizational dimensions of diagnostic AI adoption. Specifically, we are examining how clinical judgment, everyday workflows and institutional realities shape the relationship between clinicians and intelligent technologies.

Drawing on behavioural science and first-hand perspectives from healthcare professionals, our forthcoming study, AI in Diagnostics: Barriers to Adoption, looks beyond the familiar adoption narrative to uncover what it truly takes for technological promise to translate into meaningful clinical value.

For MedTech innovators and healthcare leaders, the implications could extend well beyond product development. Our research will explore:

• A behavioural map of trust and override. What makes a clinician follow, challenge, or quietly ignore an AI output, and what actually rebuilds confidence after a tool gets something wrong.
• A lifecycle view of cognitive load. Which parts of working with AI become automatic, which never do, and what happens to attention under volume pressure.
• The distance between installed and embedded. Why a deployed tool is not an adopted one, and what separates the two in daily practice.
• Practical implications. What manufacturers, AI vendors, and health systems can change in product design, implementation, and commercial strategy to close the gap.

Curious about what we're investigating? Or how the findings could inform your AI strategy?

Stay tuned or get in touch with us now for an early conversation.