Skip to content

Field report 02 · September 2026

Healthcare AI 2026: from pilots to an operating system

A practical decision guide for provider executives, clinical leaders, information technology, finance, and compliance—with adoption data, real deployments, competing evidence, economics, failure patterns, and a 90-day portfolio plan.

Open the 23-page PDF
Abstract hospital operations network

81%

of AMA physician respondents reported awareness or professional use of AI in 2026; the survey instrument changed across waves.

71%

of non-federal acute-care hospitals reported predictive AI integrated with the EHR in 2024.

1,614

rows in the FDA AI-enabled device table counted for this report; 1,230 were radiology-led.

Start where work is reviewed

Documentation and administrative processes have owners, work products, and starting measures.

Test locally

Randomized studies show meaningful differences by product, specialty, setting, and comparison group.

Measure displaced work

A local time saving is not value if the work moves to a clinician, patient, appeal, or correction queue.

Keep a reliable stop

Every deployment needs version records, a shutoff control, a downtime process, and a removal plan.

The uncomfortable evidence

Ambient documentation works differently by product and setting

In a 238-physician randomized trial, Nabla users saw a 9.5% time-in-note reduction versus control while Microsoft DAX users did not show a significant reduction. Both products produced modest improvements in burnout-related measures, and clinically significant inaccuracies were still reported occasionally. In a separate Mayo emergency-department comparison, AI scribes required more note-section time per patient than human scribes. The practical conclusion is a local trial design—not a vendor leaderboard.

Real deployments

Named stories, with the caveat attached

Kaiser Permanente × Abridge

The system announced availability across 40 hospitals and more than 600 medical offices after a year of implementation work. Scale proves deployment capacity; it does not supply a universal effect size.

Providence × Microsoft DAX

A randomized step-wedge study reported 2.5 fewer hours per week of off-hours documentation, with the vendor-license relationship disclosed in the published study.

UPMC × Abridge

UPMC moved from an early cohort to describing Abridge as its primary ambient AI tool, while openly disclosing its financial interest in the company.

Stanford Health Care

In 8,740 eligible emergency-department encounters, use was only 11.2% and concentrated among a small group and lower-acuity, noninterpreted visits—an important selection warning.

Inside the field edition

Designed to support a decision now

  • Eight-part healthcare AI market map
  • Value and reversibility decision matrix
  • Ambient-documentation study comparison
  • Local trial design and stop rules
  • Food and Drug Administration device review card
  • Vendor evidence checklist
  • Worked economic model
  • 90-day rollout and balanced scorecard
  • Eight common failure patterns
  • Clinical incident scenario exercise
  • Board approval questions
  • Methods and linked sources

Selected primary sources

Put the report next to your pilot plan

No form wall. The PDF always opens in a new window.