Where the clinic day goes, and the workflow that gives it back.

Think about the last patient you saw, and how much of that encounter you spent looking at them rather than at a screen. For most clinicians the honest answer has been drifting the wrong way for a decade, and the cause is rarely the medicine. It is the software, and the workflow built around it.

This matters because the documentation burden is now measurable, large, and tied directly to whether you burn out. The encouraging part is that almost none of it is inevitable. A practice management system is a set of design choices, and most of the lost time comes from choices that could have gone the other way. So it is worth being precise about where the day actually goes before talking about how a better-designed system gives it back.

Two hours of screen for every hour of care

The clearest number comes from a direct-observation study that followed 57 physicians through their working day. For every hour of direct clinical face time with patients, they spent nearly two more hours on the electronic health record and desk work inside clinic hours, and many then took another one to two hours of documentation home each night.¹ The screen was not a tool beside the work; it had become most of the work.

Event-log data tells the same story from the other side. Across a year of primary-care records, physicians spent close to six hours of an eleven-hour day inside the EHR, with clerical tasks and the inbox accounting for the largest shares, and a steady after-hours tail that clinicians have learned to call "pyjama time".² None of that is time with a patient, and none of it is why anyone trained in medicine.

The cost is paid one click at a time

That lost time is not spent thinking; it is spent clicking. An emergency-department analysis that gave the genre its name counted close to 4,000 mouse clicks across a single busy ten-hour shift, with physicians spending more of their time on data entry than in direct contact with patients.³ Each click is trivial on its own, which is exactly why the total goes unnoticed until someone counts it.

What makes this maddening is that the click count is a design decision, not a law of nature. When researchers timed clinicians completing the same ordering tasks across different systems, the number of clicks varied as much as eight-fold from one implementation to the next, and error rates on some tasks reached one in two.⁴ The same order, the same clinician, a fraction of the friction: the difference is entirely in how the software was built.

Usability is not a soft metric; it predicts burnout

It is tempting to treat "the system is clunky" as a complaint to endure rather than a problem to fix. The evidence says otherwise. When physicians rated their EHRs on the System Usability Scale, a standard instrument used across consumer software, the technology scored in the range that earns a plain letter grade of F.⁵ These are tools that would fail as a phone app, used all day for the most consequential work there is.

And the usability score is not just an irritation; it tracks who leaves. In the same national study, every single point of better usability was associated with measurably lower odds of burnout, a clean dose-response relationship rather than a vague correlation.⁵ Better software is not a perk. It is, on this evidence, a retention strategy.

The inbox became the second clinic

Even when you are not documenting, the system is generating work for you. In one multispecialty practice, automated, system-generated messages made up nearly half of a physician's weekly in-basket, outnumbering messages from actual colleagues and patients combined; clinicians who received more of these algorithmic messages were significantly more likely to report burnout and to plan on cutting their hours.⁶ The inbox stopped being communication and became a queue the software fills.

The hours add up quietly. Audit-log analysis of more than a thousand primary-care physicians found they spent roughly an hour each workday on the in-basket alone, and more than a third of that time fell outside scheduled work hours.⁷ An hour a day, much of it at night, on a part of the job that did not exist a generation ago.

The alerts you have already learned to ignore

Decision support was meant to make all this safer, and in principle it can. In practice, it interrupts so often that clinicians tune it out. A 2024 meta-analysis found that physicians override roughly nine in ten drug-drug interaction alerts, a rate so high that the warnings stop functioning as warnings at all.⁸ When a system cries wolf on the safe ninety percent, the dangerous ten percent vanishes into the same reflexive dismissal, and a safety feature quietly becomes a safety risk.

Notes that say less the longer they get

The documents themselves have suffered too. When researchers traced where the text in progress notes actually comes from, they found that in a typical inpatient note only about a fifth had been written fresh; nearly half was copied and pasted, and the rest imported by template.⁹ A note assembled mostly from yesterday's note is longer, harder to read, and quietly less trustworthy than the short one it replaced.

This is partly cultural and partly a property of the tools. Comparing health systems on one shared platform, US clinicians generated a far higher proportion of automatically created note text than their counterparts elsewhere, and an earlier comparison found their notes ran about four times longer.¹⁰ Length is not thoroughness. When the next clinician cannot find the one line that matters, the bloated note has failed at the only job a note has.

What workflow that respects the day looks like

Put those findings together and a design brief falls out of them. If clicks are the cost, the system should reach the common task in as few of them as possible. If the inbox is the second clinic, results and correspondence should arrive structured and routed rather than as a pile to triage. If copy-paste corrupts the record, the system should make entering data once, cleanly, easier than cloning it. The goal is not a prettier screen; it is a shorter path between the clinician's intent and the record.

This is the brief Lumen is built against, and in a single specialty it is a tractable one. Because the system is designed only for ophthalmology, the eye exam is a structured workflow rather than a blank text box: visual acuity, refraction, pressures and the rest are captured as data once and then plotted as a trend across visits, so reviewing a glaucoma patient means reading a chart, not re-reading five notes. Coded diagnoses come from a clinician-curated favourites list instead of a search through the full code set, and billing items are prepared alongside the visit with the bilateral and multi-operator rules already applied, which removes a whole category of after-hours reconciliation.

The same logic runs through the parts that usually force a clinician out of the record. Scans open in the browser beside the note rather than in a separate viewer, so an OCT is one place, not two systems. Inbound referrals, pathology and correspondence are designed to arrive through Medical Objects into the record itself, which is the difference between an inbox and a fax tray. Notes are versioned rather than overwritten, so the integrity that copy-paste erodes is preserved by the system instead of by the clinician's discipline. None of this is exotic; it is the ordinary consequence of treating every click as a cost.

Where AI helps, and where it does not

Ambient documentation is the most promising recent development, and the early evidence is genuinely good. In a large health-system rollout, an ambient AI scribe was associated with reduced documentation and EHR time within the first weeks of use,¹¹ and a 2025 multi-site study found that thirty days of ambient AI was associated with a substantial drop in clinician burnout and after-hours documentation.¹² Used well, it can return the minutes that note-writing takes away from the conversation.

The caveat is that a scribe drafting into a system that still takes 4,000 clicks to act on the note has fixed only the easiest part. Ambient AI removes typing; it does not remove the ordering, the coding, the billing or the inbox. That is why the durable answer is workflow first and AI inside it, not AI bolted onto a system that was the problem. The profession has put a number on the ambition: the AMIA 25x5 initiative sets the explicit goal of cutting documentation burden to 25% of its current level within five years.¹³ That target is only reachable by addressing the workflow, not just the keyboard.

This is now the reason people leave

None of this is abstract in Australia. In the RACGP's national survey, administrative workload has become the single most-cited reason general practitioners give for planning to leave the profession, with the great majority reporting concern about the burden.¹⁴ The paperwork is no longer a grumble at the edge of the job; it is pushing clinicians out of it.

Which is the real reason workflow deserves the attention usually reserved for clinical features. Every click you do not have to make, every result that arrives where it should, every note you can actually read is time and attention handed back to the person in the chair. Good practice-management software is not measured by how much it can do. It is measured by how much of your day it gives back to the patient, and to you.