Why Healthcare AI Stalls on Data and Trust, Not Models

A senior healthcare AI leader sees AI as inevitable and already valuable, but says regulatory data friction, unproven ROI, and organizational fear, not technology, are what hold it back.

This director of AI at a med tech firm treats AI as real and already delivering clinical and productivity value, not hype. The firm uses AI on-device to tell surgeons whether cancer was fully removed, and to denoise images so acquisition stays fast. Yet the binding constraint is data. Deployed devices generate de-identified images daily, but consent, IRB approval, and per-institution data agreements make most of that data unusable, turning data access into a multi-year, costly burden. Internally, the operator names three frictions: cost, organization, and fear. AI carries recurring token and subscription costs that mount quickly for smaller firms, and this operator cites that nearly 90% of AI initiatives fail to produce permanent change because ROI is never realized. The discipline that matters is picking few high-value projects and proving wins. Organizationally, the operator warns that firms bolt AI onto existing structures instead of becoming AI-native, and wishes for a published blueprint for that transition. Workforce fear of replacement must be reframed as force multiplication. Finally, the operator wants responsible-AI frameworks and federal data policy, not to slow progress but to preserve public trust against deepfake-driven backlash and to keep pace with faster-adopting global markets. The throughline: technology is not the problem. Data access, cost discipline, organizational design, and trust are.