Making AI actually perform in acute care settings

Ambient AI is exploding in popularity because charting overhead is growing and clinicians want time back. However, acute care adoption significantly lags ambulatory settings: KLAS estimates in their Ambient Speech 2025 report that 95% of adoption is in ambulatory care, with only 5% in acute care.

Why is this happening?

• More complex encounter workflows: Triage, initial assessments, test and order results, re-examinations, consultations and more complex problem lists demand multi-step encounters over several hours, with real-time changes in decisionmaking. PIT, split-flow/fast track, consult loops, re-evaluations during boarding, and critical patients that are crashing are all part of the job.
• Intricate documentation requirements: Acute care notes are not a simple reformatting of an ambiently captured conversation. SEP-1, stroke door-to-CT, critical care capture, MIPS measures are just a handful of the thousands of criteria defined by Quality, Risk, and Revenue Cycle Management teams:.
• Need for site-level and clinician-level customization: No two sites are the same – local stroke triggers and APP cosign rules differ hospital to hospital.

Acute clinical leadership and CMIOs need solutions that can model actual Emergency Medicine and Hospitalist Medicine workflows, absorb local guidelines and requirements, and integrate seamlessly into their EMR systems. Speech-to-text transcriptions alone are insufficient: you need clinical reasoning in workflow, and a note that stands up to Quality/RCM/CDI requirements.

Key considerations for successful deployment of AI in acute care

1. Model current clinical workflow

Successful clinical AI solutions must respect current practice workflows to drive utilization. Solutions must deeply understand the reality on the ground: does the ED leverage a Provider in Triage (PIT) model? Is there a low-acuity fast track? What privileges do Advanced Practice Providers (APPs) have? Are HM clinicians staffed on admit-only shifts? The clinician experience needs to support each workflow, and note outputs need to reflect these considerations.

2. Capture clinically-defensible medical decisionmaking

Acute care decisions hinge on baseline risk and comorbidities. Clinical AI solutions must help clinicians make their reasoning explicit. Documentation should surface how CKD shapes contrast use, how anticoagulation changes head-injury management, and how diabetes with atypical symptoms moves a borderline troponin toward admission. Clearly distilling a clinician’s clinical reasoning generates outsized impact on documentation time, Quality/Risk misses, and coding outcomes.

3. Drive Quality and Risk improvements

Quality and Risk guidelines shouldn’t be seen as checklists. They’re important steps in ensuring every patient has the best outcomes possible. Stroke alerts, SEP-1 sepsis documentation, and MIPS measures need to be surfaced in real time, to ensure clinicians can risk mitigate before arriving at a final disposition and plan. Medical directors need control over recommended pathways and standards of care to ensure medicine is practiced as recommended in their groups –

4. Customize while aligning with RCM and CDI requirements

Every minute matters in acute care settings. Clinicians want their notes to reflect their clinical decisionmaking, but they don’t have time to review thousands of guidelines related to procedure capture, EKG interpretations, E/M leveling, and hospital admission criteria. AI can help drive more compliant notes, resulting in fewer downstream denials and queries, while still allowing charts to reflect individual doctors’ tone and charting style.

How to measure AI impact in acute care

Acute care leaders should partner with vendors to run tightly defined pilots that measure both quantitative and qualitative impact. Measure what matters:

• Clinician adoption and satisfaction: documentation time savings, % of clinicians using solution for full shifts, % retention 90 and 180 days after onboarding
• Operational efficiency: Improvement in patients per hour, Time to Provider, Turnaround Time to Admit or Discharge (TAT-A/TAT-D)
• Quality/Risk adherence: Compliance with Quality Measures like SEP-1 compliance and MIPS
• Financial ROI: Reduction in coder and CDI queries, change in procedure and critical care capture, trends in E/M leveling and specificity of ICD-10 diagnosis capture

Ambient transcription is table stakes. Success will be driven by the quality of in-workflow clinical reasoning, and creation of defensible notes. Launch at your sites, grade vendors on the pilot rubric above, and let the results decide.

The post Making AI actually perform in acute care settings appeared first on Becker’s Hospital Review | Healthcare News & Analysis.