3, 2, 1: Health AI Brief

AI is reshaping healthcare fast. Below are 3 key AI developments, 2 studies, and 1 takeaway to help you better lead with AI. Target read time: 5 minutes.

3 Market Signals

BVP: AI companies captured 55% of health tech funding in 2025

Bessemer Venture Partners' State of Health AI 2026 report shows AI's dominance accelerating: AI companies captured 55% of all health tech funding (up from 37% in 2024), average deal sizes jumped 42% to $29.3M, and six companies broke the two-year IPO drought generating $36.6B in market cap. What's different this time? These AI-native companies deliver "services-as-software" with 70-80% margins—reaching $100M+ ARR in under five years versus ten for the previous generation. The report covers a lot of ground including ambient scribes, clinical AI for triage and risk, payer tech, and a value-based care renaissance.

So what?

Here's the dynamic to watch: providers deployed AI for revenue cycle, and it's working—claims are higher-quality and higher-cost. Payers are now getting squeezed. BVP forecasting greater investment by payers in payment integrity, prior auth + utilization review, and member engagement.

Read the full report →

Amazon launches 24/7 AI health assistant for One Medical members

Amazon's One Medical introduced Health AI—a 24/7 assistant that accesses patients' complete medical records to explain lab results, book same-day appointments, and manage medication renewals through Amazon Pharmacy. Available to all One Medical members ($9/month for Prime members). Unlike consumer health chatbots, this one knows your full history without manual uploads.

So what?

Amazon is effectively betting that owning the patient relationship—from pharmacy to primary care, and AI for everything in between—creates a competitive moat. Could be a smart partnership opportunity for health plans.

Read the announcement →

JPM26: Health systems report concrete AI wins

At JPM26, leading health systems shared hard numbers on AI impact: Mayo Clinic's AI now detects pancreatic cancer up to 3 years before traditional diagnosis. Tampa General cut sepsis mortality by 68% using AI-powered early warning. Cleveland Clinic is combining AI with quantum computing for drug discovery. The theme: AI is moving from pilots to measurable outcomes.

So what?

Hospitals are embracing AI, especially for key use cases, and seeing very promising positive results.

Read the AHA recap →

2 Research Studies

Lancet: AI mammography detects 29% more cancers with 44% less workload

The MASAI trial—the largest randomized controlled trial of medical AI to date—screened over 100,000 women in Sweden. AI-supported screening (AI + 1 radiologist) detected 29% more cancers than standard double-reading (2 radiologists) while reducing radiologist workload by 44%. The AI-detected cancers were less aggressive and caught at earlier stages.

Why it matters

This is the evidence the field has been waiting for: a massive RCT showing AI can both improve outcomes and reduce burden. The key (for now)? AI augmented radiologists rather than replacing them—and freed them to focus on complex cases.

Read the study →

Lancet: AI stethoscopes improve detection—but only when doctors use them

The TRICORDER trial deployed AI-enabled stethoscopes across 205 NHS practices covering 1.5 million patients. When doctors used the AI stethoscope, detection improved dramatically: 2.3x for heart failure, 3.5x for atrial fibrillation, 1.9x for valvular heart disease. But overall detection rates didn't improve—because many doctors stopped using the device, citing extra steps and poor EHR integration.

Why it matters

The technology worked. The implementation didn't. This trial is a cautionary tale: clinical AI lives or dies on workflow integration, not algorithm accuracy.

Read the study →

1 Key Insight

The adoption gap is the real AI challenge.

This week gave us a tale of two trials. In Sweden, AI-assisted mammography screening detected 29% more cancers while cutting radiologist workload by 44%. In the UK, AI stethoscopes could detect heart failure 2.3x better than standard care—but only when doctors actually used them.

The UK trial's sobering finding: despite proven accuracy, many physicians stopped using the AI stethoscope over time. The reasons? Extra steps in their workflow. Poor EHR integration. The friction outweighed the benefit.

The Swedish trial worked because it reduced burden—one AI-assisted radiologist replaced two. The UK trial stumbled because it added burden—another device, another step, another screen.

Same underlying technology.

Yet—

Opposite workflow impact.
Opposite clinical outcomes.

Takeaway

When evaluating AI tools, don't just ask "will it work?" Ask "will it fit?" The best algorithm in the world fails if clinicians won't use it. Workflow integration isn't a nice-to-have—it's the difference between a pilot that scales and one that quietly dies.