Healthcare organisations do not benefit from technology that performs well in a demonstration but fails in routine practice. In clinical settings, time is limited, cognitive load is high and anything that adds extra work is unlikely to last. The practical value of AI depends less on novelty than on whether it addresses real obstacles in care delivery and supports work where care is actually delivered. A durable approach begins with disciplined problem definition, careful workflow observation, deliberate task allocation and trust built through evidence. That process starts before a tool is introduced.
Must read: AI Adoption in Health Systems Faces Readiness Gaps
Organisations need to understand the current experience, decide how it will be measured and identify the friction they want to remove. Without that foundation, claims of improvement are difficult to demonstrate. Adoption also depends on whether the people doing the work recognise the problem themselves. If they do not, change is likely to feel disruptive rather than useful.
Measure Before Change
Any organisation considering AI needs a clear view of the current state before introducing a new capability. Without a baseline, it is not possible to show improvement with confidence. Progress may be described, but it cannot be demonstrated in a way that stands up to scrutiny. In a healthcare environment where leadership teams, clinicians, patients and boards may be sceptical of innovation that does not translate into practical value, measurement matters.
The baseline can be built from straightforward operational indicators. Time on task, error rates, rework, completion rates, satisfaction, training time and the frequency of help requests or escalations can all help define how work happens today. These measures support later comparison, but they also impose discipline. Establishing the before picture forces organisations to examine actual conditions rather than rely on assumptions about efficiency or burden.
The same discipline applies to defining the problem. In many AI discussions, attention settles on the tool before the problem has been stated clearly. A more reliable starting point is the hurdle that needs to be overcome, the friction that needs to be removed or the outcome that needs to change. That shift matters for adoption as well. It is not enough for leaders to conclude that a problem exists. The people carrying out the work need to recognise it too. When they do not, intended improvement can be experienced as interruption. When the pain is real and shared, users are more likely to engage in testing, accept iteration and develop trust in the process.
Observe Work as It Happens
Direct observation of workflow is essential because users do not always identify what is broken. In many environments, friction becomes normalised. People adapt to inefficient steps, develop workarounds and build muscle memory around processes that no longer serve them well. As a result, interviews and surveys may not capture the full extent of the problem.
Watching real work can reveal unmet needs that would otherwise remain hidden. Extra clicks may seem too minor to mention. Repeated steps may feel inevitable. Shortcuts may have become routine despite introducing risk. Manual tracking may continue in the background even when no one wants to acknowledge it. These details shape the reality of day-to-day work and often determine whether a new tool will succeed.
Muscle memory has particular force in clinical settings. During a busy clinic day, people are likely to return to what feels familiar, fast and reliable. A tool that disrupts that pattern without a clear payoff is unlikely to last. For that reason, workflow redesign cannot depend only on asking users what features they want. Observation is more revealing when the goal is to understand where effort, delay and risk actually sit.
Measure Clinical Value and Build Trust
Task allocation matters as much as technical capability. The central design question is not simply whether AI can automate a task, generate output or make a recommendation. The more important question is where the work should live. Some repetitive administrative tasks drain time without improving care and may be suitable for automation. Other tasks involve ambiguity, judgement and context, and shifting them to automation can introduce new risk. AI supports care only when it removes burden rather than relocating it. If output requires constant correction, repeated verification or extra searching for the source, it increases cognitive load even if it appears efficient on paper.
Usability gains are often easier to quantify than clinical value. Reduced time, fewer errors, better satisfaction, fewer clicks and less rework can all be measured relatively quickly. Even so, those indicators are not sufficient on their own. The more important questions concern whether the tool improves care, improves outcomes, reduces harm or supports better decisions. These effects may take longer to appear, but they still require a measurement plan.
Trust also has an operational dimension. It depends on practical questions about where data is stored, how long it is stored, who has access, who governs it and how those answers can be guaranteed. Transparency matters as well. Users need to understand why a recommendation appears, where the information came from and whether it can be verified without adding more work. Trust also requires testing for unintended consequences under real-world conditions. Access is part of the same picture, because new technology can widen gaps when affordability, digital literacy, language, connectivity or disability access are not addressed deliberately.
A durable approach to AI in healthcare begins with the reality of work rather than the appeal of emerging technology. Effective implementation depends on benchmarking the current experience, defining the problem clearly and confirming that users recognise the friction that needs to be addressed. It also depends on observing real workflows, allocating tasks intentionally and measuring both efficiency and clinical value. Trust grows when systems are transparent, traceable, governed and tested under practical conditions. Access must also be designed deliberately so that progress does not widen existing gaps. A human-centred approach keeps attention on reducing burden, supporting care and creating digital experiences that respect the people expected to use them.
Source: HealthData Management
Image Credit: iStock
