Ambient scribes are sold on giving you your evening back. The revenue case underneath the pitch is richer coding — one vendor advertises roughly $13,000 in incremental annual revenue per clinician. So here is the actual 2024 Medicare office-visit ledger for 42 specialties, and the up‑coding rate each one would have to sustain to earn it. Drag the rate. Watch nobody arrive.
The scribe business case has two doors. Door one is volume: see more patients in the time you saved. Door two is coding: document the visit you were already having well enough that it bills a level higher. Vendors mostly walk through door two, because door one sounds like what it is.
Door two is arithmetic, and the arithmetic is public. Below, every dot is a specialty. Horizontal: office E/M visits per clinician per year, Medicare fee‑for‑service, 2024. Vertical: the incremental Medicare dollars per clinician per year if the scribe moves the slider’s share of that clinician’s 99213s up to 99214 and the same share of 99214s up to 99215. Dot area is how many clinicians are in the specialty.
At the default — a 25% up‑code rate, meaning one in four of your established office visits gets documented a full level higher than it used to be — zero of 42 specialties reach $13,000 from Medicare office visits. The median specialty lands at about $3,532.
Hold that number next to the best measurement anyone has taken. Holmgren et al., 1,565 physicians and 1.2 million encounters at UCSF, found about 1.81 additional wRVUs per physician per week — roughly $3,044 a year. The CMS ledger says a 25% up‑code rate would produce $3,532. The measured effect of a real ambient scribe deployment is the arithmetic of shifting about one visit in five up a level. That is not nothing. It is also about a quarter of what is being advertised.
Push the up-code slider until a specialty finally touches the line. For Family Practice — 372 Medicare office visits per clinician per year — you need to move 83% of every remaining 99213 and 99214 up a level to reach $13,000. Not 83% of borderline visits. 83% of all of them. For Obstetrics & Gynecology, at 92 Medicare office visits a year, no rate under 100% gets there at all, because there aren’t enough visits to up-code. The coding door does not open wide enough. Which means the money, if it is going to appear, comes through the other door.
Here is the uncomfortable part, and it is why this is a contract problem rather than a model problem. Take the shortfall between what the scribe actually produces in coding and what the system paid expecting, and divide it by the specialty’s blended allowed amount per visit. For most specialties the gap is worth one to two additional visits per week.
One to two visits a week is not a re-organisation. It is a slightly tighter template, a 20‑minute slot that becomes 15, two add-ons on a Thursday. It is small enough that nobody has to announce it — which is exactly why it is the likeliest way the difference gets found. Human scribes ran this experiment for twenty years: a meta‑analysis of 39 studies found throughput up about 0.3 patients per hour and RVUs up 0.55 per hour. Jevons did the rest.
The scatter is almost a straight line, and that is the finding, not a flaw. Turn on colour by 99213 headroom and look for a specialty where the level-up is worth meaningfully more per visit. There isn’t one. The 99213→99214 spread across all 42 specialties runs from — to — — a national fee schedule does not vary by how clever your documentation vendor is. So the vertical axis is essentially visits × a constant. There is no specialty where coding is worth more. There are only specialties with more visits.
Which reframes the whole pitch. A scribe cannot make your visits worth more than the fee schedule says. It can only make more of them clear a threshold, and the supply of visits sitting just under a threshold is finite and shrinking — see the companion piece, where the 99214 share of Medicare office visits climbed from 41% to 55% between 2013 and 2024 with no ambient AI in the room at all.
The Medicare-share slider is the other thing to distrust. Everything above is Medicare fee‑for‑service only. Drag it to 35% — a plausible Medicare share for a general adult panel — and specialties start clearing the line, because the model then assumes your commercial and Medicaid volume behaves exactly like your Medicare volume and pays at Medicare rates. Commercial pays more, so that is conservative in one direction; but commercial panels skew younger, healthier and lower-acuity, so the up-code headroom is smaller in the other. Any $13,000 figure quoted without a stated Medicare share and a stated up-code rate is not a number. It is a brochure.
| specialty | visits/clin/yr | $ at rate | rate needed | extra visits/wk |
|---|
This is Medicare Part B fee‑for‑service, office place of service, established-patient E/M only. No Medicare Advantage — roughly half of all beneficiaries. No commercial, no Medicaid, no new-patient codes (99202–99205), no prolonged-service add-ons, no hospital or telehealth E/M, and no G2211 complexity add-on, which some systems are pushing scribes to capture and which would change these numbers. Clinician counts are distinct NPIs billing at least one of these four codes, so a part-time biller counts the same as a full-timer — visits-per-clinician is therefore an underestimate of a full-time clinician’s real load, and the specialties with the most part-timers look the thinnest.
It also cannot tell you whether up-coding is wrong. If the visit was always a level 4 and the documentation never said so, the scribe is fixing an accuracy problem and the revenue is owed. That is the strongest version of the vendor case, and this chart is consistent with it. What the chart does say is that even the strongest version does not reach $13,000 — and that a buyer who models $13,000 and collects $3,044 has an $10,000 hole per clinician to find, in a year, from somewhere.
Efficiency that meets a volume incentive gets absorbed as volume. The question a builder can answer is whether the dashboard reports minutes saved, which finance will convert, or minutes protected, which somebody has to defend.