How to Build Clinician Trust in AI for Healthcare Operations, part 2: Rethinking Strategy and Structure
In my previous blog, I discussed some of the common AI myths that create clinician distrust and generate friction for any potential or ongoing AI rollout. But it isn’t enough to simply communicate clearly to your clinicians and to bust myths effectively; instead, leaders also need to reimagine entire frameworks around operations, onboarding, and usage.
Shifting from reacting to real-time problem solving
Most hospitals take a reactive approach, analyzing historical data for insights, reports, and decision making. For example, a charge nurse might sit down to review staffing patterns only at the end of the quarter, when attendance data is collated and analyzed from EHRs or other devices. However, this means that this same charge nurse (or even the CNO) cannot surge extra nurses or technicians as necessary, in response to unexpected occurrences, such as mass casualty events or seasonal bouts of a virus.
To be clear, this isn’t the fault of any one leader, or even the hospital as a whole. After all, it’s only been recently that our society invented the necessary devices and data tools to attain real-time visibility and insights. But as healthcare evolves, this after-the-fact methodology falls short; for instance, when compared to 1990, hospitals have to deal with a more challenging patient population (with many more comorbidities and acute issues) and longer, more complex clinical treatments.
How roles will evolve with real-time visibility
Real-time visibility closes the gap between what leadership assumes is happening on the floor and what is actually happening. For the charge nurse, that means they can see immediately whether a shift change is working, where a new shortfall is appearing, and drill into the specific cause rather than waiting weeks for a retrospective.
That shift also changes how they spend their time doing. Instead of piecing together the floor’s status through phone calls, in-person check-ins, and EHR lookups, they can see the picture directly and spend that time deciding what to do next. For instance, a charge nurse reviews weekly or monthly reports showing hours when a unit ran below target nurse-to-patient ratio useful for spotting patterns like “nights are chronically short on Tuesdays.”
The result is a role that is both more strategic, since they can spot longer-term patterns faster, and more tactical, since she can respond to a sudden problem the moment it appears.
When compounded across multiple hospitals, real-time visibility is invaluable, enabling a health system to become more nimble. Impact that used to take years or months can now be accomplished in a fraction of the time, simply because there is a steady stream of fresh insights flowing to leaders at all levels of the organization.
Assuming that a health system rolls out operational AI at scale, this shift will occur across every level of an organization. Nurses, technicians, team leads, and executives won’t (and indeed, can’t) be replaced; instead, they simply have to spend less time wading through reams of reports and data to find inefficiencies, shuttling back and forth between units and campuses, or trying to reconcile perceptions with reality.
All that time and effort can instead be devoted towards delivering better care, whether it’s an executive removing extra, excess rental equipment, or a nurse being able to spend more time bedside.
What this means for staffing models, roles, and governance
However, unless a health system or hospital is set up properly, it won’t be able to utilize the full potential that real-time insights bring. This requires a higher degree of flexibility in workflows. As my colleague Matt Cannell wrote in his implementation article, either leaders have to deprecate current procedures and build new ones, or work their new solution into what a hospital already has.
For example, the hypothetical charge nurse needs to be empowered with enough flexibility to shift a technician, move a float nurse, or even call in additional nurses from other sources (such as staffing agencies) as necessary. Without this supportive operating model, the charge nurse is powerless to act on these valuable, real-time recommendations.
Governance has to be built in from day one. Ownership needs to be assigned at the level closest to the problem, with clear thresholds for what a charge nurse can decide alone versus what requires escalation, and a fast path for the latter. If every reallocation of staff or equipment has to wait on a manager or director, the real-time insight is wasted the moment it’s flagged. Defining that decision authority upfront, rather than retrofitting it after the technology is in place, is what turns visibility into action instead of just another dashboard that nobody is empowered to use.
Optimizing onboarding for your employees
While predictive algorithms that can forecast equipment usage or optimal maintenance periods have long been used in industrial sectors, they aren’t yet widespread in hospital operations. As a result, many clinicians may associate AI with commercially-available large language models (LLMs) like ChatGPT or Claude, neither of which were designed for healthcare operations.
Therefore, overcoming mistrust through education is critical. After end users, whether they’re clinicians, team leads, or technicians, understand that AI is not there to replace them and their judgment, it’s vital to train them in best practices around AI usage. And because these best practices will transform and change along the way, their training has to be updated, either in the form of refresher courses or small, one-day workshops that will not further distract them from the point of care.
This last point is extremely important. Clinicians already are busy, often with paperwork; studies found that they spend up to anywhere from 25–50% of their time on administrative duties alone. But rather than framing these training sessions as another mandatory duty, leaders can instead position AIs as an all-seeing, intelligent personal assistant that can not only provide recommendations, but also automate away key tasks. By using these AIs, clinicians no longer have to spend long hours on the phone calling other departments or combing through obscure records to find insights. Instead, they can return to what they do best: bedside care.
The nuts and bolts of training
Any real onboarding initiative begins with mapping out and addressing the impacts on each end user’s role and responsibilities. That means sitting down with each department before the rollout, determining what normal shifts and workflows look like, and then identifying where new tools can improve these routines. Without this step, training will remain generic and staff will remain unclear on how this new solution will fit into their existing workflows.
Training also has to be specifically tailored to each role and should also mirror actual workflows, complete with screenshots, dummy interfaces, and other tools that are as true to real life as possible. That means that technicians, nurses, charge nurses, and leadership will each need different materials.
Take the example of a supply chain AI that helps orchestrate equipment across a health system. A supply chain technician would train on their unique workflow: how they receive dispatches, whether it’s via SMS or a push notification on their work device or how to sort through tasks by order of priority. On the other hand, a VP of Supply at a health system would likely use different tools, including a real-time dashboard that displays global views with key metrics such as equipment utilization, distribution, quantity, and even device flows between hospitals within the same network.
It’s also important to be realistic about end user impacts, and not over-promise. If a supply chain AI can save a supply chain tech several minutes on a task, or help a VP of Supply coordinate the system-wide pump fleet more efficiently, then the AI should be framed as exactly that: an orchestration tool that helps everyone work more effectively and reduces spend.
Do hospitals need a Chief AI Officer?
As with every emerging technology, best practices around AI are still being established. However, one controversial question has been making the rounds: do health systems need Chief AI Officers, or are they better served by putting AI under the purview of the CIO?
Part of the controversy stems from the already-full portfolios that CIOs struggle with. After all, they are tasked not only with modernization, but also with maintaining legacy systems, upon which hospitals depend (and which is more difficult each year, as vendors go out of business or deprecate support for older products).
As a result, some health systems have taken the burden off CIOs and instead appointed Chief AI Officers to take on this task. For instance, Mayo Clinic recently named a Chief AI Implementation Officer, while Kaiser Permanente, City of Hope, and Children’s National Hospital have also created equivalent roles with different titles. All share similar responsibilities, overseeing AI governance, implementation, and oversight.
Many of these networks also share another key quality: they are all large health systems, often with multiple facilities. Given their size and complexity, creating a dedicated executive position to handle AI is a logical step, especially as AI is already in widespread use across their campuses and throughout their workforce.
Conversely, regional and community health systems may not have adopted AI on the same scale, which means that creating a formal C-suite role would be premature. Even so, these health systems will need to answer key questions around governance, such as who is responsible for each tool or workflow, who will approve new alterations, and who is tasked with identifying model drift.
Therefore, the key is to ensure that someone in leadership can answer these questions and be accountable for AI. Whether that equates to appointing a dedicated AI officer will depend on the needs of the specific hospital or health system.
Start small, scale big
Change is intimidating, and the pressure to roll out AI quickly can be overwhelming. But just as every journey starts with a single step, any AI implementation can begin with a single workflow. Select one procedure, maybe even one that already uses real-time data, and experiment: task one person with the authority to oversee the success of this altered, AI-powered workflow, another to identify which threshold triggers escalation, and one more to review the decision and efficacy of these changes afterwards.
Beginning with one workflow is a more manageable starting point than either a system-wide rollout or a new executive hire. It also gives leadership a concrete result to justify changes to stakeholders, including clinicians, insurers, and patient advocacy groups alike.
