Rosa Del Mar

Daily Brief

Issue 63 2026-03-04

Adoption And Utilization As The Gating Metric

Issue 63 Edition 2026-03-04 7 min read
General
Sources: 1 • Confidence: Medium • Updated: 2026-03-08 21:18

Key takeaways

  • Many AI healthcare vendors have underwhelming real-world adoption and utilization once deployed despite strong vision marketing.
  • A health system has projected about $30M in net new margin attributable to Ambience's platform.
  • AI tooling is changing the engineer profile Ambience hires toward deep architectural thinkers and product engineers who embed in clinical settings for requirements gathering.
  • Decision traces are crucial training data for clinical AI, and mutable EHR data structures can destroy these traces, implying a need to rethink data capture architecture.
  • A near-term path toward more autonomous care is using virtual care team members to pre-compute visit summaries, gather upstream answers to likely clinician questions, and perform post-visit follow-up and monitoring.

Sections

Adoption And Utilization As The Gating Metric

  • Many AI healthcare vendors have underwhelming real-world adoption and utilization once deployed despite strong vision marketing.
  • In some large academic medical center deployments, over 75% of clinicians use Ambience daily.
  • Many enterprise AI deployments have low adoption (about 15–20% of doctors) and partial usage (about 20–40% of visits).
  • Durable AI purchasing decisions in health systems should prioritize tools that achieve high clinician adoption and clearly improve operating margin.
  • AI tools in hospitals only matter if clinicians adopt and use them at high rates.
  • Ambience has achieved broad clinician adoption by owning the window of care in front of clinicians, and this adoption level is table stakes.

Economic Validation Requires Cfo-Grade Measurement And Attribution

  • A health system has projected about $30M in net new margin attributable to Ambience's platform.
  • One Ambience customer is projecting over $30M in net new margin driven in part by revenue cycle improvements and increased throughput/access after attribution debate.
  • Credible AI ROI for health system CFOs requires rigorous measurement and attribution capabilities rather than vendor claims.
  • Ambience is willing to structure partnerships where its economics depend on helping a health system improve operating margin.
  • Ambience built an analytics stack by pulling customer data warehouse information to meet CFO standards for ROI validation.

Organizational Bottlenecks And Ai-First Operating Model

  • AI tooling is changing the engineer profile Ambience hires toward deep architectural thinkers and product engineers who embed in clinical settings for requirements gathering.
  • AI capability improvement is outpacing product development, requiring teams to plan around predicted capabilities about 18 months ahead and continually reinvent.
  • Ambience’s engineering productivity increased substantially due to internal use of the AI tool Opus 4.5, enabling fewer people to produce more work.
  • The main bottleneck in shipping healthcare AI is increasingly organizational capacity to understand problems and staff teams, rather than foundation model capability alone.
  • Companies built today should operate as AI-first organizations, and Ambience is building internal teams to encode and share decision context for faster onboarding and better decision-making.

Data Architecture Limits (Decision Traces) And Evaluation Difficulty

  • Decision traces are crucial training data for clinical AI, and mutable EHR data structures can destroy these traces, implying a need to rethink data capture architecture.
  • Defining and evaluating clinical quality is intrinsically difficult because clinical records can contain contradictions and require interpretation using factors like chronicity and clinician credentialing.
  • Clinical documentation often requires capturing an internal decision tree not fully spoken in the patient conversation, making note generation and downstream coding non-trivial.

Workflow Redesign Path: From Scribe To Virtual Care Team Members And Revenue Cycle Changes

  • A near-term path toward more autonomous care is using virtual care team members to pre-compute visit summaries, gather upstream answers to likely clinician questions, and perform post-visit follow-up and monitoring.
  • Embedding expert reasoning into models and delivering it at software marginal cost at the point of care could obsolete parts of revenue cycle workflows such as pre-bill correction.
  • Within about six months, marquee academic medical center customers often expand from scribe functionality to requesting Ambience’s broader roadmap.

Watchlist

  • An AI-vs-AI arms race between provider revenue cycle optimization and payer countermeasures is emerging.
  • Cascading clinical context across care settings and predicting next best actions remains difficult, and predictive modeling is not yet well solved by current models.

Unknowns

  • What are independently verified clinician adoption metrics (active user rate, encounter coverage, retention at 90/180 days) across multiple Ambience customers and multiple specialties?
  • How is net-new operating margin attributed to the platform (counterfactuals, controls, time horizon, inclusion of labor changes, denial/cash timing), and has it been audited by customer finance teams or third parties?
  • What specific components make up the EHR integration/context layer (data extraction scope, normalization, identity resolution, real-time latency, failure modes), and what is the incremental integration cost per new site/EHR instance?
  • What concrete data-capture approach preserves immutable decision traces in practice, given mutable EHR structures, and how often are traces incomplete or ambiguous?
  • What evaluation framework is used to measure clinical documentation quality when records contain contradictions and implicit reasoning is not spoken, and what are the observed error categories and their rates?

Investor overlay

Read-throughs

  • Clinician adoption and encounter coverage become the primary procurement filter for clinical AI, advantaging vendors that can demonstrate sustained utilization over vision-led pilots.
  • Buyer demand shifts toward CFO-grade ROI attribution and auditable measurement, pushing vendors to build analytics, controls, and outcome-linked commercial structures as part of the product.
  • EHR mutability and loss of decision traces elevate data-capture and context-layer integration as competitive differentiators and potential cost drivers for scaling across sites and specialties.

What would confirm

  • Independently verified adoption metrics across multiple customers and specialties, including active user rates, encounter coverage, and 90 and 180 day retention.
  • Audited or finance-team validated net-new margin attribution methodology, including counterfactuals, controls, time horizon, and treatment of labor changes and cash timing.
  • Clear description of integration and context layer scope, latency, failure modes, and incremental integration cost per new site or EHR instance, plus evidence of repeatable deployments.

What would kill

  • Low sustained usage after deployment, with poor encounter coverage or declining retention, indicating shelfware dynamics despite initial rollouts.
  • ROI claims that cannot be attributed with controls or audited by customer finance teams, or that depend on assumptions about labor changes and denial or cash timing.
  • Integration complexity or EHR data mutability preventing reliable decision-trace capture, leading to incomplete training data, unclear evaluation of documentation quality, or high per-site scaling costs.

Sources

  1. 2026-03-04 a16z.simplecast.com