We spent three years optimising 41 seconds

Summary

The best randomised result in the ambient AI category is 41 seconds a note.1 A family physician loses a reported $28,000 to $74,000 a year to downcoding that never appears on a denial dashboard.2

We have spent three years competing on the first number. So has everyone else. We think that was a category-wide mistake and we were part of it.

Documentation that carries the Medical Decision Making, and eligibility run before the visit.

Book a demo

Two numbers

UCLA randomised 238 physicians across 14 specialties, roughly 72,000 encounters. Nabla cut note-writing time by 9.5 percent, about 41 seconds, and no other ambient scribe holds a randomised win.11Lukac PJ, Turner W, Vangala S, et al. Ambient AI Scribes in Clinical Practice, A Randomized Trial. NEJM AI 2025;2(12). 238 physicians, 14 specialties, approximately 72,000 encounters.

Separately, RCM analysis across 180 family practice groups puts silent Medicare Advantage downcoding at $28,000 to $74,000 per physician per year.22Reported MA downcoding exposure of $28,000 to $74,000 per family physician per year, from RCM analysis across 180 family practice groups, 2026. Not our data. Confirm against your own remittances. Add undercoding at 19 percent of E/M visits and $37 each.33AAPC Audit Services reports up to 19 percent of E/M visits are undercoded nationwide at roughly $37 each. Neither of those appears in a denial report, because neither is a denial.

Every vendor in this market, including us, has spent three years building for the first number.

Why the seconds won

A microphone is straightforward to ship and a note is straightforward to demo. You can put a scribe on a stage and the audience watches text appear. No vendor has ever demoed a submitted-versus-paid E/M variance report and had a room applaud.

The seconds are also measurable in a way the money is not. You can time a note. Timing a downcode requires reading remittances against submissions across a quarter, and no vendor wanted a metric that made their product look small.

So the category optimised what it could show. We did it too. We knew the two-hour figure and quoted it in our own decks, then built for the ten minutes anyway.44Sinsky C, Colligan L, Li L, et al. Allocation of physician time in ambulatory practice. Ann Intern Med 2016;165:753-760.

What we think now

The note is not the product. The note is the evidence for the code, and the code is the money. A 99214 stands or falls on whether the documentation carries the Medical Decision Making, and a scribe that produces a fluent note with thin MDM has made you faster at losing $42 to $74 a visit.

A scribe is therefore for defensibility rather than for speed, which is not how any of us have sold one.

What we built

WA\ Clinician drafts against the MDM criteria and attaches the coding at the point the decision was made. WA\ Admin runs eligibility at booking, which is where about 27 percent of denials are preventable, and works the claim after on the same record.

The two halves are one system because the submitted-versus-paid variance is only visible when both halves are.

What we are not claiming

We have no randomised evidence. Nabla does. Forty-one seconds measured under randomisation is worth more than any number we can currently show you, and we would say that if we were you.

We are not an RCM company. The downcoding figures are somebody else's analysis of somebody else's practices, and the only version that should persuade you is the one from your own remittances.

A number we would rather be judged on

Ask any vendor in this category for submitted-versus-paid E/M variance across their customers, before and after. No vendor publishes it, ourselves included.

Physicians spend close to two hours on documentation and desk work for every hour of direct patient care. That was measured in 2016 and it has not moved.4 The industry response has been to make 41 seconds of it faster.

Frequently asked questions

Do AI scribes improve medical coding?

They can, and it matters more than the speed they are sold on. The note is the evidence for the code and the code is the money. From 2026 E/M level selection turns on Medical Decision Making, so a 99214 stands or falls on whether the documentation carries the problems, the data reviewed and the risk. A scribe producing a fluent note with thin MDM has made you faster at losing $42 to $74 a visit. The only randomised trial in this category measured note-writing time and found 41 seconds, and measured after-hours work and found no change.

Why do AI scribe vendors focus on note-writing speed?

Because it demos and it measures. You can put a scribe on a stage and an audience watches text appear. Nobody has ever demoed a submitted-versus-paid E/M variance report to applause. Speed is also timeable in a way the money is not, since timing a downcode requires reading remittances against submissions across a quarter, and no vendor wanted a metric that made their product look small. So the category optimised what it could show. We did the same, and we knew the two-hour documentation figure while we did it.

What should I ask an AI scribe vendor about revenue?

Ask for submitted-versus-paid E/M variance across their customers, before and after, on consecutive encounters rather than a chosen sample. Nobody in this category publishes it, including us. Then ask whether the product drafts against the 2026 Medical Decision Making criteria or simply transcribes fluently, because those produce different notes and only one of them defends a 99214. Then ask whether anything in the product touches eligibility, where about 27 percent of denials are preventable before a claim exists.

Ask us for the variance report.

Ask any vendor in this category for submitted-versus-paid E/M variance across their customers. No vendor publishes it, ourselves included.

Book a demo See pricing

About this article. Written and published by WA\, which sells clinical AI and is arguing that the category it competes in has been optimising the wrong thing. The randomised trial cited is independent of us and tested versions current in 2025. Downcoding and undercoding figures are third-party RCM analysis and are not our data. Nabla holds the only randomised win in this category and we have said so. WA\ has no peer-reviewed clinical trials published to date and does not claim any. None of this is billing, legal or compliance advice.