Executive Summary
A multi-state hospital system deployed a new AI triage platform, projecting a 40% reduction in patient intake times, but saw nurse override rates hit 85% within 90 days, negating all efficiency gains. The AI, owned by an innovation group and measured on deployment speed, was clinically misaligned and lacked auditable decision trails, leading to a breakdown in trust with bedside staff. By re-platforming with clinician-in-the-loop guardrails and assigning workflow ownership to the Chief Nursing Officer, the system reduced the override rate to under 15% and cut average triage time by 28% over the following six months.
The Challenge
The core conflict in clinical AI adoption is the tension between innovation velocity and the non-negotiable requirements of patient safety and clinician trust. Our composite scenario is a 15-hospital system facing endemic emergency department (ED) overcrowding and high rates of nurse burnout. National benchmarks show average ED wait times exceeding 150 minutes, and the rate of patients who leave without being seen (LWBS) is a critical metric tied to both revenue and patient safety. The executive mandate from the board was clear: reduce patient wait times, lower the LWBS rate from a system-high of 4.5%, and decrease the administrative burden on a strained nursing staff.
The newly formed AI innovation group, a central function reporting to the Chief Technology Officer (CTO), was tasked with deploying a solution rapidly to demonstrate the value of the organization's new enterprise AI license. They procured a state-of-the-art Large Language Model (LLM)—a powerful, general-purpose AI—designed for summarizing patient intake forms and suggesting an initial triage acuity level. The acuity was based on the Emergency Severity Index (ESI), a standard 1-to-5 scale where Level 1 represents a critical, life-threatening case and Level 5 is non-urgent. The project's key performance indicators (KPIs) were speed to market and the number of patient encounters processed by the AI, metrics that incentivized deployment over adoption.
