Healthcare AI adoption is rising - but still behind other sectors
AI use in healthcare has climbed since 2023, yet it remains behind sectors like information services, finance, and professional services. That's the picture from a research letter in JAMA Health Forum analyzing U.S. Census Bureau Business Trends and Outlook Survey (BTOS) data from September 2023 to May 2025.
Across that period, mean AI use among healthcare firms was 5.9%. Adoption moved from below 5% in 2023 to about 8.3% in 2025 - real progress, just not at the pace of other industries.
Key numbers at a glance
- Healthcare mean AI use (Sep 2023-May 2025): 5.9%; growth to ~8.3% in 2025
- 2025 comparisons: Information services 23.2%; Professional, scientific, and technical services 19.2%; Education 15.1%; Finance and insurance 11.6%
- Subsector gains: Outpatient and ambulatory care rose from 4.6% (2023) to 8.7% (2025); Nursing and residential care facilities from 3.1% to 4.5%
Why the gap persists
Common barriers leaders cite include data privacy and security requirements, EHR integration friction, safety and bias concerns, unclear reimbursement pathways, and change management across distributed clinical teams. None of these are trivial, but they're solvable with a focused plan and tight governance.
Where adoption is picking up
Outpatient and ambulatory settings saw the fastest growth, likely because high-volume administrative work (scheduling, documentation, intake, messages) lends itself to clear ROI and lower clinical risk. Nursing and residential care made smaller gains, an area researchers flagged for deeper study given unique workflow, staffing, and regulatory constraints.
Governance can't wait
The research team underscored two realities. First: "Future research is necessary to understand the reasons and consequences of lower rate of AI adoption in health care, particularly in certain subsectors, such as nursing and residential care facilities." Second: the quick rise since 2023 "highlights the urgent need for active monitoring and effective regulations to ensure safe and efficient deployment of AI in patient care."
For clinicians and operators, that translates to concrete guardrails: continuous monitoring for inaccuracies and bias, strict PHI controls, clear human-in-the-loop steps, and patient communication that builds trust.
Practical use cases to prioritize
- Clinical documentation support and ambient scribing to reduce charting time
- Prior authorization and utilization management automation
- Revenue cycle tasks: coding assistance, denial analysis, and appeal drafting
- Patient access workflows: triage, referrals, scheduling, and message summarization
Procurement and safety checklist
- Security and privacy: BAA, data residency, PHI handling, encryption, audit logs
- Model transparency: versioning, training data disclosures, update cadence
- Quality and bias: baseline accuracy, subgroup performance, human override
- Clinical risk: human review points, escalation paths, incident reporting
- Integration: EHR/API readiness, SSO, logging, and rollback plan
Metrics to track from day one
- Efficiency: chart closure time, prior auth turnaround, message backlog, average handle time
- Quality: error rates, clinician corrections per note, subgroup performance drift
- Financial: denial rate, appeals win rate, cost per encounter
- Experience: clinician after-hours work, patient wait times, satisfaction scores
A 90-day plan you can run
- Weeks 1-2: Pick one high-volume, bounded workflow; define success metrics and risks; form a cross-functional team (clinical, IT, compliance, revenue cycle).
- Weeks 3-6: Vendor short list; security review; sandbox with synthetic data; draft BAA; design human-in-the-loop checkpoints.
- Weeks 7-10: Pilot with a small cohort; train staff; monitor daily; capture corrections and incidents.
- Weeks 11-13: Compare against baseline; decide go/no-go; document controls; plan scale-out with ongoing monitoring.
Bottom line
Healthcare is moving - just slower than peers. The path forward is simple, not easy: target clear-value workflows, enforce strong governance, measure relentlessly, and scale only when the data proves it.
Sources and further reading
Want a structured way to upskill your team on safe, useful AI in care delivery and operations? Explore curated AI courses by job.
Your membership also unlocks:
