AI Patient Triage Cutting Emergency Wait Times in Half
A regional health network with 12 hospital sites had emergency wait times averaging over four hours. Manual triage could not keep up with patient volume. We built a conversational AI triage agent connected to their Epic EHR. It structured patient intake, scored acuity, and routed patients faster. Average wait time dropped from 4.2 hours to 1.5 hours.
The health network runs 12 acute care facilities across the region and handles over 400,000 emergency department visits a year. Nursing shortages and rising patient volumes pushed manual triage past its limit. Clinical leadership had tried traditional workflow optimization. It wasn't enough.
The bottleneck had several causes. Manual data entry from registration to EHR took 12–18 minutes per patient. Acuity scoring varied across sites and nurses. High-acuity patients weren't getting flagged fast enough. Nurses were spending 30% of their shift on admin work instead of patient care.
12–18 minutes of manual data entry per patient before clinical assessment begins
Inconsistent acuity scoring across sites leading to suboptimal patient routing
High-acuity patients not identified quickly enough — 8% of critical patients waited >60 minutes
Nurses spending 30% of time on administrative intake rather than clinical care
Patient satisfaction scores in the bottom quartile of the health network's peer group
We built a conversational AI triage agent on tablet kiosks in waiting areas. It connects to the network's Epic EHR via HL7 FHIR APIs. The agent runs structured symptom intake, applies the Manchester Triage System protocol, and uses ML to score acuity. It generates a pre-populated clinical note for the nurse. Their job becomes validation and sign-off — not data entry.
Clinical workflow mapping across all 12 sites — identifying variation in triage protocols and data requirements
Conversational AI agent built using OpenAI Assistants API with custom guardrails for medical terminology and escalation pathways
ML-enhanced acuity scoring model trained on 3 years of triage outcomes from the client's own EHR data
Epic EHR integration via certified HL7 FHIR APIs — pre-populating clinical notes, not replacing them
Tablet kiosk deployment with accessibility accommodations and multilingual support (7 languages)
Clinical governance process: every AI acuity score reviewed by a registered nurse before any routing decision
Six months after full deployment, average ED wait time across all 12 sites fell from 4.2 hours to 1.5 hours. Nurses went from spending 30% of their shift on admin to 12%. Patient satisfaction moved from the bottom quartile to above the peer group median. Zero safety incidents were linked to AI-assisted triage.
Average ED wait time: 4.2 hours → 1.5 hours (64% reduction)
Nurse administrative time: 30% → 12% of shift
Critical patient identification speed: 8% waiting >60 min → 1.2%
Patient satisfaction: bottom quartile → above peer median
Zero AI-related safety incidents in 6 months of operation
“We've seen measurable improvements in both patient satisfaction and staff workload from week one. The AI doesn't replace clinical judgment — it gives our nurses the time and information to exercise it better.”
Dr. Amara Osei
Chief Medical Officer, Regional Health Network
Client identity is withheld under NDA. The figures reported here were verified against the client's own internal reporting at project close.
Clinical governance has to come before technology. The nurse review requirement wasn't a limitation. It was the feature that earned clinical trust.
Training on the client's own outcome data beat generic medical models every time on acuity accuracy.
Building multilingual support from the start — not as a retrofit — was the deciding factor for adoption at two sites with large non-English-speaking patient populations.
Key Results
Services Engaged
Technology Stack
Facing a Similar Challenge?
Let's discuss your specific context and what results are realistic for your organisation.
Get in touch