Reclaiming Time: Designing AI That Gives Back the Day

When asked what is wearing them down, most clinicians will tell you that it is not patient care. Care is what they trained for and what they still enjoy. What they cannot endure is everything about the care that has to be done surrounding it; the overflowing inboxes, the incessant clicks, the rapid fire messages that no one has the time to respond to. We developed digital systems to streamline care, and instead we turned the attention of care staff into the scarcest resource on the team.

Burnout is a justified response to the current system. It is not a lack of resilience. It is "time theft" by blindly adding unnecessary steps, clicks and approvals that restrict the time available for the most important tasks.

How We Got Here

It is important to recognize how we reached this point to understand that there was no malicious intent to create systems that would result in an excessive burden of clicks for clinicians. Each system solved a real problem. Billing required a framework to include documentation. Safety required additional controls, thus more notifications. It became easier to message, which led to an explosion in message volume. Each step was justified in isolation. The end result was a workday where care was continually pushed to the periphery, dominated by work related to care. No one designed the work burden and, as a result, no one takes responsibility for it or measures it.

Measuring Time

Time connects almost everything we focus on in healthcare. With enough time, we can think and improve quality. If we are rushed, we lose safety, and we can miss the important things. Time also affects whether a provider stays or leaves. With all the factors, we measure time remarkably little.

Instead, we measure volume, like the number of visits, the amount of throughput, and the productivity units. It is a counting exercise, and it is entirely incorrect, because a provider can meet every productivity target and still not feel like they did their job because of the volume of time that was not spent with patients. The most important measure, the one that is rarely measured, is time. It is not like we cannot measure it. Most of the information is in the system logs. The system knows when a note was finished and how long someone was on the inbox. It is just that we did not choose to track it.

Costs are incurred downstream from decisions made upstream during the workflow. Clinicians do not get just the emotional impact of a difficult day. Clinicians become more emotionally detached from the work and make decisions to cut short the conversations that take the most time to prepare for. Missing the signal in the noise is what they do as they work under severe time constraints. The patients who need the most time, the ones who do not fit the clinical template, are the ones who suffer most from a fully booked schedule. This lost time becomes a problem for the patient, and not just the staff.

What Giving Time Back Really Means

AI becomes useful when it is deliberately designed to save time. The time of a clinician is wasted by writing notes well after the appointment has finished, and if clinicians have to do it for many patients, it will add up to a massive backlog. AI has the potential to save clinicians time by summarizing lengthy notes, and by clearing routine messages before they ever reach the clinician, which saves time across thousands of communications.

But this is where things start to fall apart. Returning time is a claim, and a claim that is unmeasured is a claim that is untrustworthy. Most healthcare AIs are marketed on the promise of time savings and are sold with the first model accuracy as the only measure, which are not the same thing at all. A tool can create a perfect model and still make each day as burdensome as all the others.

Measure the Time, Not the Throughput

So measure the time. The metrics are not intricate. Minutes. Time wasted waiting for the chart to finish. The work that seeps into your evenings and weekends. Time lost to the void of the inbox. The percentage of time a clinician looks at a screen instead of at the patient. Establish a real baseline before the tool is implemented, because in the absence of a baseline, every tool appears to work and no one can prove otherwise.

When time is the measure, the conversation changes. It no longer speaks to the cleverness of the technology, and speaks instead to whether the day was made lighter for those living it. It brings to light the trap that insidiously sabotages most of these initiatives. If the time savings from AI are consumed with another patient visit, you may capture a time savings on a slide, but the clinician never realized it. Time that is returned must be protected to count as time that is returned. Otherwise you have measured productivity and called it relief.

Who Gets to Keep the Time

Knowing how time gets measured creates a problem that does not need to exist. Once the measurements show how much time a tool returns, someone has to assign value to that time, and that value is almost never consciously assigned. In the absence of a conscious choice, the time falls to whoever owns the schedule, and the schedule always wants more volume. Therefore, the time gets assigned before anyone asks if it should. If returned time is to have meaning, the time must be assigned to a purpose of care, and that purpose must be assigned by someone who can protect it. It is a decision of leadership rather than engineering, and it is the difference between relieving a clinician and simply raising the quota.

Validate the Experience, Not Just the Model

Most teams overlook a basic discipline that underlies all of this. Validating a model is not sufficient. It is also necessary to validate that the experience improved. A right model can still add a click, interrupt the flow, and break the rhythm, thus making the day worse. The model is right, but the experience is a failure. Only one of them was measured.

We see the same pattern. A summarization tool will score pretty high on accuracy, but then adds a step because now the clinician is reading the summary along with the source to verify. The model completed its task. But now the days are longer. This was undetected, because no one was tracking the length of the days.

To close this gap, we have to include the clinical perspective in the design process from the very beginning. The sign-off process can be cut completely. The people who will use the tool can determine in a matter of minutes whether the tool is usable within the context of the actual workflow, and they are almost always correct. If clinicians participate in the design process, they are more likely to trust the output. If clinicians trust the output, use of the tool becomes part of the workflow rather than an exercise in change management to get people to use the tool. The same principle works in the opposite direction. A tool that is designed for clinicians without their input will almost always be ignored, no matter how good the underlying model is.

What We Are Really After

Time is the invisible thread connecting quality, safety, and the simple willingness to keep doing this work. If we want AI to improve care, the first thing it has to do is give time back to the people delivering it, and we have to be honest enough to measure whether it actually did. It does not need to be flashier or faster. It needs to be measurably lighter, in a way the person at the keyboard can feel at the end of a shift.

Previously, clinicians were expected to adapt to the innovations, to endure the challenges, and to learn the nuances of the system. The better approach is the opposite. Create innovations that adapt to clinicians, and then assess them based on the only real metric that counts here: the amount of time it saves them.

Christopher Hutchins Founder and CEO, Hutchins Data Strategy Consultants