For a decade the direction of travel in health IT was consolidation. Suites absorbed point solutions, health systems absorbed practices, and the independent clinic was told the future involved picking one large vendor and living inside it. That story is now running in reverse at the small end of the market — not because anyone won an argument about architecture, but because the enterprise model never fit a two-person office and the administrative burden got bad enough to force the issue.
This piece is about what the evidence actually supports, including the parts that are inconvenient for people selling ambient AI. I have a commercial interest here — I build one of these tools — so everything below is sourced, and where the research undercuts the sales pitch I've said so.
First, the burden is real and it is measured
The documentation burden is one of the better-quantified problems in medicine, which makes it unusual among things vendors complain about on your behalf.
Average total EHR time per eight hours of scheduled patient time among ambulatory physicians, from a study of more than 200,000 physicians published in the Journal of General Internal Medicine. Documentation specifically accounted for 2.3 of those hours.
That benchmark comes from AMA-funded research using national EHR metadata. Two things in it deserve attention. The first is that total EHR time and documentation time are different measures, and they get conflated constantly in vendor material — if a claim about "hours in the EHR" doesn't say which one it means, treat it as marketing.
The second is the variation underneath the average. Infectious disease sat at 8.4 hours per eight scheduled; primary care at 7.3; dermatology at 4.3; orthopaedics at 3.3; anaesthesiology at 2.5. A tool's value proposition is not the same for a dermatologist and an endocrinologist, and any vendor quoting you one number for all of medicine hasn't looked at the distribution.
Underneath the time data sits a usability problem. In comparative System Usability Scale scoring, EHRs have been benchmarked around 46 out of 100 — in a landscape where a score in the 70s is considered acceptable software. Studies of click behaviour across hospital networks have found ranges on the order of dozens of discrete clicks to complete a single routine order. This is not a training problem. It's an interface designed around billing and compliance requirements rather than around the person using it at 6pm.
What the 2026 JAMA study actually found
Until this year, most public claims about ambient scribe time savings traced back to vendor-run pilots or single-site studies. In April 2026, JAMA published the largest controlled look at the question so far, co-led by investigators at Mass General Brigham and UCSF.
The design matters for how much weight to give it: 8,581 ambulatory clinicians across five academic health systems — Mass General Brigham, Emory Healthcare, UCSF, Yale New Haven Health, and UC Davis — of whom 1,809 adopted an ambient scribe, tracked over more than two years, using tools from Ambience, Nuance DAX Copilot, and Abridge alongside Epic.
Reductions in total EHR time and documentation time respectively, per eight scheduled patient hours, associated with ambient scribe adoption. Relative decreases of roughly 3% and 10%.
Adoption was also associated with about 0.49 additional patient visits per week. Benefits concentrated among primary care clinicians, advanced practice clinicians, female clinicians, and clinicians who used the tool in at least half their visits — that last group seeing considerably larger reductions.
The findings vendors don't quote
Three results in the same study complicate the sales pitch, including mine.
- After-hours EHR time did not significantly decrease. Time recovered during the day appears to have been absorbed by other clinical work rather than removed from the workday. If your pitch — or your hope — is that ambient AI ends evening charting, this study does not support it.
- The authors call the effect modest. Senior author Rebecca Mishuris noted the reductions are unlikely to fully account for the large burnout improvements reported elsewhere (STAT coverage), which means something other than raw minutes is doing that work. We come back to what that something is below.
- Only about a third of adopters used the tool in half or more of their visits. The largest benefits went to consistent users. A licence that sits idle produces nothing, and most licences were not being used consistently.
Read that first bullet twice if you're evaluating vendors. Almost every ambient scribe on the market, including ours, sells on the promise of getting your evenings back. The largest study to date found daytime documentation time fell and after-hours time didn't move significantly.
That doesn't make the tools worthless — 16 minutes a day of reduced documentation time is real, and roughly half an extra visit a week is real. It means the honest claim is narrower than the common one, and a vendor who won't tell you that is telling you something about how they'll behave after you've signed.
What the study measured, and what it didn't
Before drawing conclusions for a two-physician practice, it's worth being precise about the population studied — because this is where a lot of vendor commentary is currently overreaching in both directions.
The study looked at ambulatory clinicians at five large academic health systems, all running Epic, using three enterprise ambient platforms. It did not include solo practices, small independent groups, or any of the lightweight EHRs those practices actually run.
That's a limitation on generalisability, and it cuts one specific way: we don't know what these tools do in a two-person office, because nobody has measured it at this scale. That is a genuinely open question, not a hidden advantage.
Where some commentary is going wrong. A line of argument circulating since April holds that the JAMA findings are simply wrong for independent practice — that enterprise systems are where good software goes to die, and a solo doctor would close the laptop and go to dinner.
That's a hypothesis, not a finding, and it isn't what the study shows. Absence of evidence about small practices is not evidence that small practices do better. Any vendor telling you otherwise — including one selling to small practices, which is what we do — is asking you to accept a claim nobody has tested.
What can be said honestly is that the composition of after-hours work differs between settings, and that difference is mechanically relevant. In a large system, documentation is one input among several into the evening workload; in-basket volume, institutional messaging, and cross-departmental routing all compete for the same hours. Reduce documentation time and the other inputs remain. In a solo practice the after-hours pile is weighted differently, and unfinished notes are a larger share of it.
Whether that difference is large enough to change the result is untested. It is a reasonable basis for a hypothesis and an unreasonable basis for a promise.
The dose-response finding, and the catch inside it
The most actionable result in the study isn't the headline average. It's the gradient underneath it.
Only about a third of adopters used the tool in half or more of their visits. Benefits concentrated in that group, along with primary care clinicians, advanced practice clinicians, and female clinicians — with reported additional savings for the heaviest users reaching roughly 25 minutes a day.
Share of ambient scribe adopters in the 2026 JAMA study who used the tool in 50% or more of their visits — the threshold associated with the largest reductions in documentation time.
The practical implication for a buyer is straightforward: a licence used occasionally produces close to nothing, and any ROI projection built on the study average is describing a population where two-thirds of users were inconsistent.
The catch is one that enthusiastic readings tend to skip. This is observational data, and clinicians who find a tool useful use it more. Some of the gap between heavy and light users is dose-response; some is selection. The study design can't fully separate those, so treat "use it more and save 25 minutes" as an association rather than a lever you can pull.
Why burnout falls further than the clock explains
The most interesting question the time study raises is one it can't answer: why do burnout improvements reported elsewhere look so much larger than 13 to 16 minutes would predict?
Two 2025 studies in JAMA Network Open, both involving Mass General Brigham investigators, point at an answer — and they're worth separating, because they're frequently conflated.
You and colleagues surveyed more than 1,400 clinicians at Mass General Brigham and Emory Healthcare. Mass General Brigham saw a 21.2% absolute reduction in burnout prevalence at 84 days; Emory a 30.7% absolute increase in documentation-related wellbeing at 60 days. The authors note low survey response rates at both sites, which is a real limitation on how much weight to put on it.
The second study is the more mechanistically useful one. Olson, Meeker, Troup and colleagues found burnout among participants fell from 51.9% to 38.8%, and measured something the time studies don't capture at all:
Note-related cognitive task load, on a 10-point scale, before and after ambient scribe use — a 2.64-point reduction. The same study found a 2.05-point improvement in clinicians' reported ability to give patients undivided attention.
That measurement is the crux. Writing a clinical note from nothing is a composition task: recall the encounter, impose structure, generate language, check completeness, all at the end of a day when the cognitive tank is empty. Reviewing a draft is an editing task. Editing is not free, but it is a different kind of work, and the evidence suggests it costs less.
This reframes what ambient documentation is for. If the product were only a stopwatch, 13 minutes would be a thin proposition. If what it actually changes is the character of the task — from generating under fatigue to reviewing under fatigue — then the time figure was never measuring the main effect.
Worth holding alongside that: not every study finds significant burnout movement. A separate Sutter Health evaluation published in JAMA Network Open found mean note time per appointment fell from 6.2 to 5.3 minutes, while its burnout change (42.1% to 35.1%) did not reach statistical significance. The cognitive-load mechanism is well supported and not unanimous.
Why enterprise deployment models don't survive contact with a small practice
Notice where that study was run: five large academic health systems, all on Epic. That is the environment nearly all published ambient AI evidence comes from, and it is not the environment most independent practices live in.
A health system deploying an ambient scribe has an informatics team, an Epic analyst, a vendor implementation manager, a security review process, and a budget line for all of it. A two-physician podiatry practice running Office Ally or Practice Fusion has a front desk, an outside biller, and whoever is least afraid of computers.
The mismatch shows up in four places:
- Contract structure. Enterprise agreements assume seat counts and multi-year terms that make no sense at three providers.
- Deployment. An integration project measured in months is not a project a solo practice can staff.
- EHR coverage. Vendors build integrations where the users are, which means Epic, Cerner, athenahealth. The long tail of lightweight EHRs that independent practices actually run gets served late or never.
- Maintenance. An integration is not a one-time cost. Someone has to care when the API changes.
The integration tax nobody puts on the quote
Here is the part of the unbundling story that gets missed, and it's the part that matters most for anyone buying at the small end.
When a documentation tool integrates with your EHR, you are not buying one relationship. You are buying three: with the tool vendor, with your EHR vendor, and with whatever agreement governs the connection between them. That third relationship is where the cost hides.
| What integration adds | What it looks like in a small practice |
|---|---|
| Vendor approval | Your EHR decides whether the tool you chose is allowed to connect, on their timeline, not yours |
| Platform or per-provider fees | Some EHRs charge a flat annual developer fee; others bill per provider per month, which scales against you as you grow |
| Coverage risk | If your EHR isn't on the supported list, the tool doesn't work, no matter how good it is |
| Maintenance exposure | API changes, credential rotation, and version upgrades become your problem the day they break |
| Switching cost | Changing EHR later means re-litigating the whole stack |
The genuinely interesting move at the small end of the market is not building a more specialised model. It's removing the integration requirement altogether — the tool drafts the note on the clinician's own device, and the clinician places the finished text into whatever chart they use. No approval, no supported-systems list, no third relationship.
That is a real trade-off, not a free lunch. You give up automatic filing into the chart, and you accept a manual step. What you get is a tool that works on day one regardless of what EHR you run, and that can't be switched off by a vendor decision you don't control. For a practice with no IT function, that trade is often the right one — and it's the actual shape of unbundling at this end of the market.
Documentation and the revenue cycle, without the overclaim
Any honest discussion of documentation in independent practice eventually reaches money, because denials are a genuine operational problem. Industry reporting has put initial denial rates in the low double digits, with a substantial share of providers reporting more than one in ten claims denied, and analyses of denied-claim behaviour consistently find that a large proportion are never reworked at all.
It is tempting — and increasingly common — to draw a straight line from those numbers to a product claim: capture better notes, prevent denials, protect revenue. I'd encourage buyers to be sceptical of that line, and vendors not to draw it.
Where the line sits. Documentation records the care that was delivered. That's its function and its legal standing. A tool that shapes what appears in a chart according to what clears a payer's review is no longer recording care, and the compliance exposure from that lands on the practice, not the software company.
There's a growing genre of advice encouraging clinicians to write notes engineered to satisfy automated payer review. Federal enforcement history with documentation software vendors — including multiple False Claims Act settlements in the last decade — is a reasonable argument for treating that advice carefully.
The defensible version is narrower and still worth something. An appeal is assembled from the note. A note reconstructed at 11pm from memory contains less than the encounter did. Details that were never recorded cannot be used later, for an appeal or anything else. Capturing accurately at the point of care means more of what actually happened survives. Whether that changes any particular payer decision is not something software can promise.
What to actually ask a vendor
If you're evaluating ambient documentation for a small practice, these questions separate the serious products from the well-marketed ones:
- Does it require an EHR integration, and is mine supported today? Not on a roadmap. Today.
- What does the integration cost, including anything my EHR vendor charges? The number on your quote may not be the whole number.
- What is the evidence for your time-savings claim, and is it yours or independent? A vendor-run pilot is not a study.
- Does it write to my chart automatically, and can it do so without my sign-off? The answer to the second half should be no.
- Does it suggest codes or modifiers? Understand exactly what you're taking on if it does.
- Who holds the BAA, and what happens to audio and transcripts? Retention periods, training use, subprocessors.
- What happens if I stop paying? Whether your past notes are portable matters more than anyone expects it to.
Where this leaves independent practice
The unbundling thesis is broadly right, but not for the reason it's usually argued. It isn't that small vendors have better models than large ones. On raw transcription quality, the enterprise tools in that JAMA study are excellent.
It's that the enterprise delivery model — integration projects, approval cycles, seat minimums, IT dependency — is a poor fit for practices that have none of the infrastructure it assumes. The opening at the small end isn't sophistication. It's the removal of prerequisites.
And the honest expectation to set, for buyers and vendors both, is the one the data supports: a modest, real reduction in documentation time, larger for clinicians who use the tool consistently, with no guarantee about your evenings and none at all about your denials.
MediAttune is built on that trade deliberately. It runs on the clinician's own iPhone or iPad, drafts the note from what was said during the visit, and hands it back for review, editing, and signature. There's no EHR integration to wait on, which means it works the same whether a practice runs Office Ally, Tebra, Practice Fusion or something nobody outside the building has heard of — and it means the clinician moves the finished text into the chart themselves. It has no specialty templates and no clinical logic. It doesn't suggest codes, check modifiers, or shape a note toward any billing outcome, and it won't add a finding the clinician didn't state. For practices with no IT function, that set of constraints is the point rather than a limitation.
How much time do ambient AI scribes actually save?
The April 2026 JAMA study of 8,581 ambulatory clinicians across five academic health systems found adoption associated with 13.4 fewer minutes of total EHR time and 16.0 fewer minutes of documentation time per eight scheduled patient hours — relative decreases of about 3% and 10%. The authors describe the effect as modest.
Do ambient scribes reduce after-hours charting?
That study did not find a significant reduction in EHR time outside scheduled hours. Time freed during the day appeared to be redirected to other clinical work rather than removed from the day. It's one of the most important findings for anyone evaluating these tools, and one of the least quoted.
Do the JAMA findings apply to a solo or small practice?
Unknown. The study covered ambulatory clinicians at five large academic health systems, all on Epic. It did not include solo practices, small independent groups, or the lightweight EHRs they typically run. That is an open question rather than a demonstrated advantage either way.
If time savings are modest, why do burnout studies show larger effects?
The most likely explanation is that the main effect isn't clock time. Olson and colleagues measured note-related cognitive task load falling from 7.10 to 4.46 on a 10-point scale — a 2.64-point reduction — alongside a burnout decline from 51.9% to 38.8%. Composing a note from nothing and reviewing a draft are different kinds of work, and the second appears to cost less even when it takes similar time.
Why is integration a problem for small practices?
Integration typically requires EHR vendor approval, sometimes a partner agreement and fees, and IT capacity to deploy and maintain. A solo practice has none of that, and the timeline belongs to the EHR vendor rather than the practice.
Can documentation software prevent claim denials?
No software can promise that. Payment depends on the care delivered, how accurately it was documented, the payer's coverage policy, and administrative factors outside the chart entirely. Documentation records the visit; it doesn't control payer behaviour.
Sources
Time and visit volume: Changes in Clinician Time Expenditure and Visit Quantity With Ambient AI Scribes, JAMA, 1 April 2026. Co-led by Mass General Brigham and UCSF across five academic health systems. See also the Mass General Brigham newsroom summary and STAT's coverage.
Burnout and wellbeing: You et al., Ambient Documentation Technology in Clinician Experience of Documentation Burden and Burnout, JAMA Network Open (2025), 1,430 clinicians at Mass General Brigham and Emory Healthcare.
Cognitive task load: Olson KD, Meeker D, Troup M, et al., Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout, JAMA Network Open. 2025;8(10):e2534976. doi:10.1001/jamanetworkopen.2025.34976.
EHR burden benchmarks: Holmgren et al., Journal of General Internal Medicine (2024), national EHR metadata across more than 200,000 physicians, summarised by the AMA in Five physician specialties that spend the most time in the EHR; and Melnick et al. (2021), collected in the AMA's EHR use research index.
Denial-rate and rework figures referenced generally are drawn from published industry reporting; specific rates vary by payer, specialty, and year, and practices should rely on their own remittance data rather than national averages.
Disclosure: MediAttune.AI builds ambient clinical documentation software and has a commercial interest in this topic. This article is general information, not clinical, billing, or legal advice.