Eeshika Dadheech
There is a pattern that keeps repeating itself in health AI.
A new company raises a strong round for an AI tool. The technology is genuinely impressive. A sepsis prediction model, an ambient documentation tool, a clinical decision support system that captures what humans miss. The pilot results are often compelling and health system leadership is enthusiastic but then, quietly, utilization flatlines.
Clinicians find workarounds. Alerts are dismissed. The AI tool sits inside the EHR, is technically deployed, but functionally ignored. The startup company that looked like a breakout twelve months ago is now having difficult contract renewal conversations and wondering what went wrong. What went wrong is the last mile.
The Problem Isn’t the Algorithm
The venture community has invested billions into health AI on the assumption that better technology produces better outcomes. In a closed system, that would be true but healthcare is far from a closed system. It is a dense, emotionally charged, chronically under-resourced environment where even the most well-designed tool must survive contact with reality. This reality is one most AI developers have rarely, if ever personally inhabited.
The AI model can be right 94% of the time. But if the nurse receiving the alert is already managing six patients, waiting on a physician to call back, and documenting the patients’ conditions from forty minutes ago, the alert becomes a distraction. Not because the nurse doesn’t care, but because they are already operating at capacity and have learned, through hard won experience, to triage everything, including the technology that was supposed to help them.
In technology and logistics, the “last mile” describes the hardest step—getting a system all the way to the end user. In healthcare AI, that last mile is the nursing workflow. It’s not a technical problem but a human and operational one. Where most health AI investments quietly go to die.
Adoption Is Emotional, Not Just Rational
When technology companies think about adoption, they tend to think about staff training, implementation timelines, and change management checklists. However, in clinical environments, adoption is also deeply emotional.
Nurses have experienced so many new technology introductions, all with the promise of making their jobs easier. Yet in reality the opposite is true–extra clicks, additional documentation fields, the IT systems that talk to each other in theory but not in practice. The result is bedside-users practicing forms of protective skepticism. It means that a new AI tool, no matter how technically sound, enters the unit carrying the weight of every failed promise that came before it.
Companies that understand this design differently. They build for trust before they build for efficiency. They know that the first question a nurse asks is not “is this accurate?” but “will this make my night harder or easier?” and that the answer to the second question determines whether the first one ever gets tested.
What Nurses See That the Demo Never Shows
Every health AI product has two versions. There is the one that exists in the conference room that is elegant, controlled, and built on the assumption of quiet attention and predictable workflows. And there is the one that exists at 2 a.m. on a unit running two nurses short, where a patient is deteriorating in room four and a family is waiting in the hallway. These are not the same product. A system that looks miraculous in the demo can fail quickly on the unit floor not because the technology is wrong, but because the workflow assumptions were. This knowledge is not abstract. It is operational intelligence of the highest order. Yet it remains almost entirely absent from the product development and diligence processes of most health AI companies.
Until the people who live inside the 2 a.m. version of healthcare are involved in shaping the technology meant to support it, we will keep building tools that look remarkable in conference rooms but struggle when they finally arrive in the patient care unit.
The Investment Implication
For investors, the last mile problem is a valuation problem.
A health AI company that cannot demonstrate sustained utilization is a company with significant renewal risk. It is a company whose growth story depends on selling new logos rather than expanding within existing ones. It is a company that will eventually face the question that every health system asks before a second contract: “Did your tool actually improve care?”
The companies that answer yes to that question have almost always done one thing differently. They treated the humans delivering care not as end users to be trained, but as design partners whose insight shaped the product itself. Increasingly, that means nurses.
Nurses are the highest-volume touchpoint in patient care, interacting with patients more than any other clinician. But their role is not simply proximity—it is clinical judgment. Nurses continuously interpret changes in patient condition, advocate for patients and families, and safeguard safety in real time. They are the ones who will accept, question, escalate, or work around whatever AI produces. In many care settings, nurses are the final human checkpoint where a recommendation becomes a clinical action.
A health AI company that embeds nurses across product design, clinical validation, implementation, and commercialization carries a fundamentally different risk profile than one that does not. Not because of representation, but because of what nurses know and how nursing knowledge fundamentally shapes the solution. Nurses bring a distinct clinical lens to care delivery: the continuous bedside perspective where surveillance, prioritization, patient advocacy, and early recognition of deterioration unfold moment by moment.
Physicians and nurses care for the same patients, but their clinical knowledge and responsibilities lead them to understand the care environment in fundamentally different ways. When AI tools are designed primarily through a medical perspective without incorporating nursing knowledge, the assumptions about how care unfolds can diverge sharply from the reality of bedside practice. When the primary end user is absent from design and validation, technology that appears sound in theory can struggle in real clinical environments.
Nurses Are the Key to Actual Adoption
Healthcare AI will keep producing remarkable technology, and the pace of innovation is only accelerating. But as the market grows more competitive, the founders and investors who win will be the ones who understood early that technical performance was never the hardest part. The hardest part is building under real conditions, with real constraints, and real people who have every reason to be skeptical. The competitive moat these founders are searching for was never inside the model. It was inside the minds of the people who show up for every shift, manage every edge case, and decide in real time whether to trust what the screen is telling them. Nurses have always been the make-or-break variable in healthcare AI adoption and the industry is just starting to catch on. At Nurse Capital, we invest in companies that treat nursing expertise as a core input because those are the ones building something healthcare will actually keep.
