The question in most boardrooms is binary: adopt AI or don't. That is the wrong frame. The real governance problem is whether your organization can actually deploy what it buys.
By Dr. Sarah Matt, MD, MBA | March 24, 2026
Last month Amazon launched two healthcare AI products in one week. A purpose-built agentic platform for health system administrative workflows. A consumer-facing health assistant now available to every Amazon Prime account in America. The coverage was predictable: breathless on the consumer side, technical on the enterprise side, short on operational analysis for the people who actually have to make decisions about this.
Hospital boards read the same headlines. And then, typically, the wrong question lands in the next strategy session: "Should we be doing more with AI?"
It is the wrong question for a specific reason. It treats AI as a category decision rather than an operational one. And it puts the board in a position of approving a direction without the framework to evaluate whether the organization is built to execute it.
When a board asks "should we adopt AI," there are only two answers: yes or no. Both are wrong.
Yes, without a deployment framework, produces what I call the Pilot Trap: a successful, well-resourced proof of concept that never scales because the infrastructure that made it work in the pilot does not exist in the broader organization. The board approves the investment, the pilot meets its benchmarks, the vendor celebrates the case study, and eighteen months later the tool is running in one unit while leadership tries to explain to the CFO why enterprise ROI is still flat.
No, as a blanket position, is its own kind of failure. Amazon just put a health AI assistant on 200 million Prime accounts. OpenAI launched ChatGPT Health in January. Anthropic launched Claude for Healthcare the same month. The patients sitting in your waiting rooms are already using these tools to decide whether to call you, what to ask, and whether the diagnosis you gave them sounds right. The board that votes to "wait and see" on AI is not managing risk. It is outsourcing the patient's first clinical touchpoint to Amazon's product team.
Neither binary answer addresses the actual governance question, which is: what is our deployment architecture, and do we have the organizational capacity to make it work?
The conflation of Amazon Connect Health and Amazon's consumer Health AI in most coverage is a governance problem in miniature. They are structurally different, and treating them as one story produces a muddled strategic response.
Amazon Connect Health is a B2B administrative platform: five AI agents targeting appointment scheduling, patient verification, documentation, and EHR integration. At $99 per user per month for up to 600 encounters, it is priced aggressively against Microsoft's Nuance DAX Copilot, which has been in the ambient documentation space since 2022. UC San Diego Health is the reference customer. For health system boards, the relevant question is not "is this useful?" Reducing administrative burden for clinicians is a real and documented problem. The question is whether your organization has the change management infrastructure to deploy it without replicating the same implementation failures you have already lived through with your EHR and your last three digital front door vendors.
The consumer Health AI is a different strategic signal. Amazon now controls the patient's intent layer at scale. A patient searching for a specialist, managing a prescription renewal, or evaluating whether their symptoms warrant a visit is increasingly doing that through a platform optimized for Amazon's commerce model, not your care model. The referral loop between "Health AI suggests you might need X" and "Amazon sells X" is the business architecture. ECRI naming AI chatbot misuse the number one health technology hazard of 2026 the same week this launched is not irony. It is the entire tension in one news cycle.
For hospital boards, the strategic question about the consumer product is not "how do we compete with Amazon." It is: "How do we ensure that the patients Amazon is now pre-screening are arriving at our care settings with accurate information, appropriate expectations, and a clear path to completed care?" That is a care navigation design question. It is not an IT question.
The boards that will navigate this well are the ones that replace the binary adoption question with a set of operational ones. Here is a starting framework.
1. What is our current administrative AI layer, and what is it not covering? If you do not have a clear answer, you do not have a deployment strategy. You have a collection of point solutions. Amazon Connect Health and its competitors are targeting the gaps in that layer. A board that cannot map its own administrative AI coverage cannot evaluate whether a new platform fills a real gap or duplicates existing infrastructure.
2. What is the change management infrastructure supporting our current AI deployments? Not the approved budget. The actual human infrastructure: dedicated project management bandwidth, a clinical champion with protected time, escalation pathways when the tool generates an alert nobody knows how to act on. Every AI implementation I have assessed in the last two years has had a well-resourced pilot and an underfunded scale plan. The board cannot see this in a quarterly update. It requires asking directly: what happens to this deployment when the vendor's implementation team leaves?
3. How does AI interact with our care navigation design? This is the question that does not appear in most board packets but determines whether AI investment produces clinical outcomes. A health system that invested $40 million in its digital front door still has a 34 percent no-show rate in specialty clinics if the downstream patient journey is fragmented. AI at the front door does not fix architecture problems behind it.
4. What is our governance structure for algorithm updates? Amazon, Microsoft, and every other AI vendor reserve the right to update their models post-deployment. The FDA's Predetermined Change Control Plan framework permits significant algorithm changes without new regulatory submission. If your governance charter does not include a notification requirement and a clinical impact review process for material model updates, your board has approved a deployment that will change without your knowledge.
Amazon can schedule the appointment. It cannot fix the whiteboard in the patient's room.
Last month I was rounding with a CMO at a 600-bed system. Strong digital front door metrics: appointment starts up 22%, digital registration climbing. He walked me to a patient floor and showed me a whiteboard in a patient's room. Seventy-one years old, diabetes, COPD, heart condition, five-day stay. On the whiteboard: four phone numbers she had written herself. Cardiology scheduling. Pulmonary clinic. Endocrinology. The main hospital line. Nobody had given her a care navigation plan. Nobody had told her the portal she used to check in could also schedule her follow-up appointments.
That gap between the technology performing and the patient knowing it exists is not an AI problem. It is a care design problem. No platform from any vendor closes it. The board that approves AI spend without approving the care coordination infrastructure to make that spend matter is making a predictable mistake, and it will show up in the same place it always does: outcomes data and HCAHPS scores that do not move despite a well-built tech stack.
The Amazon news is worth taking seriously. It is not worth panicking over. The organizations that will use it well are the ones that already have answers to the four questions above. For everyone else, the relevant investment is not a new AI platform. It is the operational infrastructure that makes any AI platform actually work.
If your board is working through AI strategy and your team would benefit from an independent clinical operator's perspective, the conversation starts at calendly.com/sarahmattmd.
Strategy, governance, and clinical operations. No hype. No recap journalism. Just the practitioner's read.
Subscribe to The Sarah Matt Briefing