AI in Healthcare 2025: Excitement With Cautious Scrutiny
Healthcare leaders see clear promise in AI, but they're not buying the hype without proof. The consensus: success depends on thoughtful integration, clinically grounded tools and a focus on real problems over novelty. This year will reward rigor, not rhetoric.
Bubble or reckoning? It depends on what you're looking at
Several executives say the hype is inflated, yet the technology is durable. The short term is noisy; the long term is where value compounds. The gap between expectation and clinical readiness is the real risk.
Phill Tornroth (Elation Health) calls it a "yes and yes" moment: AI can change care delivery, but healthcare is the hard part. Building trust with clinicians and fitting tools into workflows will decide who lasts.
Matt Cybulsky (Catalant) sees less of a bubble and more of a reckoning. He flags infrastructure realities-power, water, and hardware obsolescence-as headwinds that markets can't ignore. Diligence and reality testing will separate durable companies from noise.
Regulation is a speed limit-and that's an advantage
Adam de La Zerda (Visby Medical) argues the risk isn't runaway AI, it's underuse. Privacy and strict oversight slow rollouts, but they also keep teams focused on safety, accuracy and traceability. That discipline matters in care settings.
Edmund Jackson (UnityAI) frames it simply: hype bubble, not a tech bubble. We overestimate the short term and underestimate the long term. Fundamentals win-create real client value or get left behind.
For context on guardrails, see official guidance on health data privacy from HHS HIPAA.
Where AI is already useful-and where it's overhyped
Dr. Julius Bruch (Isaac Health) reports steady gains in cognitive health-screening and workflow support in particular. Progress is incremental by design, and that's healthy in clinical fields that require evidence.
Eirini Schlosser (Dyania Health) draws a line between marketing and substance. Making sense of vast clinical text is a structural need. With clinical input, transparency and safety frameworks, this looks less like hype and more like overdue modernization.
Kara Egan (Teal Health) urges teams to ask sharper questions before they commit. Source quality, evaluation plans and failure modes matter more than claims.
What actually wins in 2025
Anu Sharma (Millie) expects a shakeout. Winners will either have defensible data moats or distribution that makes products stick. Another path: care delivery models that embed AI into clinical workflows to improve outcomes and cut costs.
Sanjay Doddamani (Guidehealth) points to structural needs-efficiency, documentation quality, population health plus precision-level management, and less administrative drag. Solutions built with clinical governance, operating models and measurable performance will outlast tools that only bolt on technology.
Emily Greenberg (Joy Parenting Club) sees the shift moving from capability to responsible application. Long-term value comes from AI that is human-literate-tone, timing, context, trust-and extends clinician capacity.
Stephen Smith (NOCD) notes we're still early. Even simple use cases like search have changed how people access information. Expect uneven execution, but continued adoption.
Practical playbook for health system leaders
- Start with high-friction, high-cost problems: documentation, prior auth, triage, care coordination, patient matching and coding.
- Co-design with clinicians. If it doesn't fit into the workflow, it won't stick. Prioritize signal over features.
- Stand up clinical governance. Define oversight, escalation paths and a change control process before go-live.
- Demand evidence. Prospective studies, bias testing, error analysis and real-world performance metrics.
- Measure outcomes that matter: time-to-diagnosis, readmissions, throughput, patient satisfaction, unit cost, and clinician time saved.
- Audit data use. Clarify PHI handling, provenance, consent and retention. Ensure models and vendors meet privacy and security requirements.
- Plan for infrastructure realities: latency, uptime, cost per inference and environmental footprint. See IEA's overview on data center energy use here.
- Avoid AI-washing. Prefer transparent model cards, documented failure modes and clear limits.
- Integrate into EHR and existing systems with minimal clicks and clear handoffs. Reduce, don't add, cognitive load.
- Align incentives. Share savings and outcome gains with clinical teams to drive adoption.
Signals you're on the right track
- Clinicians ask for the tool when it's turned off.
- Cycle times drop and quality metrics improve without extra effort.
- Leaders can trace outcomes to specific workflows, not generic "AI uplift."
- Procurement, compliance and security sign off faster because documentation is airtight.
For teams leveling up skills
If your organization is building internal capability across roles-clinical, operations, data, compliance-curated AI course paths by job can help: Complete AI Training: Courses by Job.
Big picture: AI is here, but healthcare will decide the pace. Focus on clinically meaningful use cases, design with the end user, and hold vendors to measurable outcomes. Excitement is fine; rigor will pay the bills.
Your membership also unlocks:
