Overview

We engaged Viston to reduce emergency department overload and claim denials using AI triage, clinical summarization, and claims automation.

Business Challenge

Long triage wait times, rising LWBS (left-without-being-seen), and high claim denials.

Fragmented notes across EHR modules made clinician burden high and documentation inconsistent.

Our Approach and Solution

Ran a discovery and data-privacy assessment, then delivered an AI triage assistant and clinical note summarizer integrated with our EHR.

Automated first-pass claims coding with human-in-the-loop review to improve accuracy and speed.

AI Applications and Benefits Delivered

AI Triage Assistant

Nurse-facing triage chatbot for symptom intake and acuity scoring; reduced triage times and improved consistency.

Clinical Summarization

LLM-based clinical summarization that structures SOAP notes and ICD-10 suggestions.

Claims Coding Assistant

Claims coding assistant that flags missing documentation and auto-populates CPT/ICD.

Cost of Implementation

USD 580,000 for MVP and production hardening.

Time to Implement

16 weeks MVP; 28 weeks to full rollout across 7 hospitals.

Tools and Technologies Used

• Azure OpenAI Service
• Presidio for PHI de-identification
• FastAPI
• FHIR interfaces
• Epic APIs
• Azure Kubernetes Service
• Azure Cognitive Search
• Redis
• PostgreSQL
• MLflow
• Evidently AI
• Power BI
• Terraform

Quantitative Outcomes

• 28% reduction in triage wait time.
• 12% reduction in claim denials.
• 35% faster claim turnaround.
• +18 NPS in patient digital experience.

Key Performance Indicators (KPIs) Tracked

• Triage cycle time
• LWBS rate
• Claim denial rate
• Coder productivity
• Summarization accuracy
• PHI leakage incidents

Pre- and Post-Implementation Metrics

• Average triage time: 41 minutes → 29 minutes
• LWBS: 5.2% → 3.7%
• Denial rate: 9.4% → 8.3%
• Claim processing TAT: 9.2 days → 6.0 days

Stakeholder Quotes

VP, Clinical Operations

“The AI triage assistant gave us back clinician time while standardizing acuity scoring.”

Revenue Cycle Director

“Our first-pass coding accuracy improved immediately, with fewer payer queries.”

Regulatory and Compliance Considerations

• HIPAA
• SOC 2
• BAA in place with all AI vendors
• PHI redaction at ingestion
• Prompt logging without PHI
• Audit trails
• Model output explainability for coding