This week I’ve been thinking a lot about whether AI is actually making our professional lives easier, or if it is just creating more vendors to buy into with lofty ROI claims. Everywhere you look there is a new AI startup touting reduction in documentation time, improvements in trials, better billing, etc… but the hard evidence is still spotty in many areas. We’re seeing some measurable wins, especially in documentation and trials, but there is still a yawning gap between buzz and real, validated clinical impact.

Here’s what’s inside this week:
• Evidence and data on actual ROI versus marketing claims
• Research that challenges assumptions about clinician burden
• Policy and regulation shaping how real gains are measured
• Tools start to show defined value in workflows

Let’s dive in

🧪 AI is delivering real ROI across healthcare domains: NVIDIA’s just-released survey finds that healthcare organizations report measurable returns on investment from AI applications in radiology, drug discovery, and operations, moving from experimentation to execution with true financial and clinical signal in core use cases.
Why this Matters: We’re starting to see industry-wide evidence that AI can generate value beyond hype through real measurable improvements.

📉 Healthcare systems report strong gains from ambient AI scribes: Health systems using ambient clinical intelligence saw reduced documentation time and revenue gains from better coding as evidence mounts that these tools provide tangible operational benefit.
Why this Matters: Ambient AI may be one of the first widely proven functional use cases with both clinical relief and financial impact.

🏥 Ambient AI documentation tools are spreading fast: A study found that nearly two-thirds of hospitals with a given EHR have adopted ambient AI documentation tools, signaling broad operational uptake rather than isolated pilots.
Why this Matters: When adoption scales this broadly, skepticism about ROI starts to meet data on real uptake and workflow benefit.

🔍 Subjective vs objective impacts of ambulatory AI scribes: A recent evaluation showed that while most physicians feel AI scribes reduce documentation, actual measured time savings were more modest, suggesting perception may outpace quantified benefit.
Key Takeaway: Perceptions of AI benefit can differ from measurable time savings, underscoring the need for hard metrics.

🌍 OECD outlines conditions for responsible AI in health systems: A major international report highlights where AI is already delivering value and what enablers (governance, workforce readiness, interoperability) matter for scaling.
Why this Matters: Responsible adoption is as important as measured ROI.

📋 HHS clinical AI adoption policy feedback emerging: U.S. federal health leaders are gathering industry input on what would actually drive AI adoption and lower costs, revealing priorities that go beyond marketing promises.
Why this Matters: Policy feedback loops may shape incentives toward real performance and cost savings.

What it does: Executes end-to-end workflows like scheduling, follow-up, billing checks, and more.
How I use it: I prototype workflow automations starting with bottlenecks like prior authorizations and see built-in efficiency gains, while keeping humans in loop for decisions.
Prompt clinicians can try: “Automate reminder sequence for patients with hypertension due for labs.”
Link: Real-world AI agent use cases in healthcare

The narrative that AI will replace clinicians or instantly “fix” healthcare is finally giving way to a more grounded reality: AI can be useful where humans are drowning in paperwork, and where operational bottlenecks lurk, but only when we measure its impact carefully. The real ROI conversation isn’t about vendor counts or flashy demos; it’s about whether we can prove time saved, outcomes improved, and costs reduced. Keep demanding hard data and choose tools that fit real work rather than marketing claims.

Bonus: Check out this webinar with Dr. Eric Topol and Dr. Bob Wachter discussing Dr. Wachter’s new book, A Giant Leap. I’m reading it now and its everything I’ve been saying. It’s a great discussion on the current state of AI in medicine and where they think the future will take us.