Beyond Innovation Theater: Building the Conditions for Innovation in Health Care

Boon-How Chew, MD, MMed, PhD, is an academic physician and professor of family medicine at Universiti Putra Malaysia. His research increasingly explores the intersection of AI innovation in medicine and the reform of scientific writing workflows within legacy academic infrastructures. This is the final article in his three-part op-ed series on health care transformation and innovation (readpart 1andpart 2).

Health care has a familiar image of innovation: a technology that did not exist a few years ago now improving care. The image is not wrong, but it is incomplete. Health care nonetheless equates innovation with visible technology, and a sophisticated tool dropped into an untransformed workflow changes very little. In the wrong setting, technology can add cost and complexity without producing durable improvement.

Innovation is also broader than breakthrough technology. The literature distinguishes disruptive or radical innovation from incremental and frugal approaches, and by how far the innovations redesign the work. Point solutions place a new tool into an unchanged workflow; system solutions rebuild the interdependent decisions around it. Much digital health investment has been of the first kind, which is part of why so much of it changes so little. The most useful digital tool may not be the most advanced one, but organizations often lack mechanisms to identify, evaluate, and spread them. A useful local fix can therefore disappear when its creator rotates departments, changes role, or leaves.

Innovation Myths Leadership Must Confront

Such persistent myths about innovation matter because they shape what leaders notice, fund, and permit. When innovation is equated with expensive technology, assigned to specialist teams, or judged only by whether a pilot succeeds, frontline problem-solving becomes easier to overlook and organizations learn less from both successful and unsuccessful attempts. Three myths are especially worth confronting.

Myth 1: Innovation requires a large budget. Frugal innovation shows that constraint can stimulate workable solutions rather than prevent them. Money still matters for evaluation and scale, but early innovation also depends on permission to test, relationships that support learning, and a Teal organization able to retain what staff discover.

Myth 2: Innovation is the job of innovators. Employee-driven innovation is distributed across organizations, not confined to innovation units or unusually “creative” individuals. Institutions that want these ideas to survive need protected opportunities for staff to improve work and a process for capturing useful changes before they disappear.

Myth 3: Failed innovation is a wasted effort. A pilot that falls short can still produce valuable knowledge if the organization examines why it failed and uses that learning in the next iteration. The important distinction is between failure that is ignored and failure that is analyzed. That requires enough psychological safety for staff to report what did not work.

What Genuinely Enables Innovation in Health Care

If the myths above explain why innovation stalls, the next question is what allows it to take root and endure. The answer is not another innovation program, technology platform, or strategy document. It lies in the everyday institutional conditions that determine whether people can notice problems, respond to them, and carry useful changes beyond isolated individual effort. Three conditions recur across the organizational literature:

1. Psychological safety: staff need to be able to question a process, admit uncertainty, and report mistakes without expecting humiliation or retaliation.
2. Distributed leadership, in which formal leaders create conditions for people without positional authority to exercise judgment and lead improvement where they work.
3. Cognitive sovereignty: The capacity to understand, question, adapt, and override an adopted system rather than merely comply with it. A team that understands why a tool is being used, knows its limitations, and can challenge its output is better positioned to use it responsibly and to innovate further solutions. Implementing solutions without this critical engagement would result in operational acquiescence rather than institutional advancement.

AI as Opportunity and Diagnostic Test

AI expands what health systems can attempt, from prediction and decision support to documentation and workflow automation. It also exposes institutional weaknesses quickly. High-stakes clinical AI depends on governed data, prepared workflows, people who understand changed roles, and processes for questioning or escalating unsafe output.

‎

The Epic Sepsis Model shows why data governance must include independent validation rather than reliance on vendor-reported performance. At Michigan Medicine, independent external validation differed from that reported by the developer, whereby the model alerted on 18% of hospitalizations yet missed 67% of sepsis cases. Local data, calibration, alert burden, and monitoring are therefore part of governance, not postdeployment extras.

Duke Health’s Sepsis Watch illustrates the other side. Frontline clinicians helped design the workflow; trained rapid-response nurses; and reviewed alerts and communicated them to emergency physicians, who retained the decision to treat. Implementation required explicit roles, training, communication pathways, and trust, not simply a model placed into the electronic record. In lower-resource settings, the same principle applies: simpler tools still need infrastructure and workforce capacity proportionate to their risk.

How to Begin: Practical Guidance

Innovation does not always require institutional permission to begin, but it usually requires institutional support to last. None of this lowers the bar. Innovation is defined by the outcome it improves—including the restoration of clinician time and attention to the patient—not throughput alone nor by its cost or novelty. Stopping low-value practice is itself an innovation and often a more difficult one, because it removes visible activity rather than creating it. Four steps are practical:

• Define the problem precisely. “Our referral process is inefficient” is difficult to act on, while “referrals are delayed because required information is often missing” is testable. Clear problem definition also makes it easier to check whether someone has already solved the same problem.
• Involve the people affected by the process—patients and staff. Both hold knowledge that formal project teams do not, and patients are a source of improvement ideas rather than only a group to consult. Use cocreation and stakeholder input before implementation, while there is still scope to change the design.
• Start with a small-scale test and decide how it will be judged. A clinic testing a referral workflow might begin with one service, track missing information and turnaround time, then adapt before wider rollout. Small-scale testing or using an effective sandbox helps teams learn quickly; predefined improvement criteria support decisions to stop, adapt, or expand the change.
• Plan for ownership and continuity from the start. Name who will document the change, review the evidence, and decide whether it should be sustained, modified, or scaled. Sustainable innovations require a clear route for long-term ownership and engagement.

The real institutional test is not how many pilots are launched or technologies are deployed, but whether useful ideas can become sustained improvements in care. AI will make technological activity increasingly easy to display; what will distinguish genuinely innovative organizations is their capacity to identify problems worth solving and enable staff to question and improve solutions, learn from evidence, and preserve what works.

In the AI era, the clearest divide between innovation and innovation theater may therefore be not the number of algorithms deployed, but whether an institution has built an ecosystem in which human creativity, critical judgment, and meaningful change can endure.

‎