Case Studies

Rebuilding Clinician Trust in an AI Triage Tool: A Behavioral Approach to Restoring Adoption

Background

This case study examines how a mid-size regional health system in the northeastern United States struggled to achieve adoption of an AI-assisted clinical triage tool designed to support emergency department decision-making. Despite strong pilot performance and high model accuracy, real-world adoption stalled at 43% six months after deployment. Clinicians used the tool but frequently ignored its recommendations, relying instead on their own judgment and experience.

Leadership initially assumed the issue was a training gap and prepared additional onboarding efforts. A behavioral diagnostic revealed a different problem: clinicians understood how to use the tool—they did not trust it. The barriers were rooted in concerns about professional autonomy, accountability, and the ability to evaluate AI-generated recommendations.

Intervention

A behavioral audit using the AI Trust Axis identified several trust breakdowns, with two factors driving the majority of resistance.

  • Autonomy Safety was low: Clinicians worried that following AI recommendations could create uncertainty around responsibility if outcomes were poor. Because the system did not clearly position AI as a supporting tool rather than a decision-maker, clinicians defaulted to maintaining full personal control by ignoring recommendations.

  • Fairness Comprehension was limited: The tool provided priority scores and recommendations but offered little insight into how those outputs were generated. Clinicians did not need access to the underlying model—they needed enough explanation to evaluate whether the recommendation aligned with their own professional judgment.

Based on these findings, the health system redesigned the clinician experience by reframing AI recommendations as decision support rather than instructions, adding a “Why this recommendation?” feature that surfaced relevant patient factors, and simplifying the override process to make human judgment easier and more accepted.

Results

The redesign produced significant improvements:

  • AI adoption increased from 43% to 79% among nursing and physician staff;

  • Override rates decreased from 61% to 34%, reflecting greater engagement with recommendations rather than avoidance;

  • Clinician confidence scores increased from 51% to 79%; and

  • Staff perceptions shifted from viewing the tool as “unreliable” to describing it as a “helpful second opinion.”

Following the improvement, the health system expanded the AI tool to additional departments and began evaluating further AI applications.

Real-World Application

This case demonstrates that successful AI adoption requires more than technical accuracy or user training. Organizations implementing AI systems can apply these insights by:

1. Designing for human agency

  • Position AI as support for human judgment, not a replacement for expertise;

  • Make accountability and decision ownership explicit; and

  • Ensure users feel empowered to accept, question, or override recommendations.

2. Improving AI explainability

  • Provide explanations that match users’ decision-making needs;

  • Surface relevant factors behind recommendations; and

  • Help users develop accurate mental models of how AI contributes to decisions.

3. Reducing psychological friction

  • Make human override simple and accessible;

  • Normalize disagreement with AI outputs; and

  • Treat human feedback as a valuable input for improving systems.

These principles apply across healthcare, finance, insurance, and other professional environments where AI must work alongside human expertise.

Why It Matters

AI adoption failures are often treated as technology or training problems, but many are fundamentally trust problems. People resist AI not simply because they lack knowledge, but because they have legitimate concerns about autonomy, accountability, and whether they can rely on systems they do not fully understand.

This case study demonstrates that behavioral science can identify the specific psychological barriers preventing AI adoption and enable organizations to design systems that people are willing to trust, use, and integrate into their decisions. The key lesson is simple: successful AI adoption depends not only on building intelligent systems—it depends on building trusted human-AI relationships.