Health Systems Struggle to Deploy AI Beyond Initial Purchase

Healthcare organizations are discovering that acquiring AI technology represents only the first step in a much more complex implementation journey, according to industry executives speaking at a U.S. News Healthcare of Tomorrow conference in Washington.

The sector faces fundamental questions about integrating AI products into clinical workflows, protecting patient data, and demonstrating financial returns—challenges that remain largely unresolved despite growing AI investments across health systems.

Workflow Integration Remains Elusive

Brian Anderson, chief executive of the Coalition for Health AI (CHAI), an industry-supported nonprofit with approximately 200 health system members, described the core difficulty: taking tools like Claude Code or ChatGPT Health and embedding them into intricate clinical workflows while maintaining patient data security. These integration challenges have become urgent priorities that the industry has yet to solve systematically.

The complexity extends beyond simple deployment. Health systems must determine how to monitor AI model performance, including accuracy metrics and patient health outcomes—elements Anderson noted are frequently neglected in data governance strategies.

Security Anxieties Intensify

Recent events have amplified security concerns among healthcare technology leaders. Following Anthropic's decision to withdraw two models after a federal directive banned their use by foreign nationals due to security risks, chief information security officers and chief information officers are increasingly worried about AI models being weaponized to attack health systems and extract sensitive data.

Anderson reported receiving numerous urgent inquiries from health system CISOs asking what will happen when new versions from Anthropic or OpenAI are released. He emphasized the need for the industry to collaboratively develop defensive protocols that address the anxieties of health systems, physicians, and patients.

ROI Measurement Falls Short

Seema Verma, executive vice president and general manager of Oracle Health and Life Sciences and former Centers for Medicare and Medicaid administrator, highlighted inadequate measurement of technology investment returns. She cited ambient AI scribes as an example: some administrators dismissed them as not cost-effective without accounting for the additional patients physicians can see when documentation time decreases.

Why It Matters

These deployment challenges explain why healthcare AI adoption lags behind initial enthusiasm and investment. Without solving workflow integration, security protocols, and ROI measurement, health systems risk wasting capital on underutilized technology while missing opportunities to improve patient care and operational efficiency. The industry's ability to address these foundational issues will determine whether AI delivers on its promise to transform healthcare delivery.

Verma also pointed to emerging complications, including how to handle and document disagreements between human judgment and AI recommendations—legal complexities the field has yet to navigate.

"It's more than just a purchasing decision: It's use, it's monitoring, and it's thinking about how the human is going to interact and where those friction points may occur," Verma said.

These details were first reported by U.S. News at their Healthcare of Tomorrow conference.