3-PT clinic's AI scribe: $0 to $15k a year

Two published results about AI scribes point in opposite directions. In a randomized trial of 238 outpatient physicians, one scribe cut time spent in notes by about 9.5% and another showed no significant change (NEJM AI, November 2025). In a small preprint on simulated physiotherapy consultations, documentation time fell 35% on average (JMIR preprint, 2026). Put those two numbers through a three-clinician physical therapy practice and the yearly value of the time saved runs from $0 to about $15,000. This article shows the sums so you can replace our inputs with yours.

If you run the practice and spend your evenings finishing notes, this is the arithmetic that decides whether a scribe gives you hours back or just a monthly invoice.

The inputs, and which ones are ours

The wage is the only figure here that comes from a government source. The US Bureau of Labor Statistics reports a median annual wage of $101,020 for physical therapists as of May 2024 (BLS Occupational Outlook Handbook, checked 26 October 2026). Divided by 2,080 working hours, that is $48.57 an hour, or about 81 cents a minute. The division is our calculation, not a figure BLS publishes, and it values time at pay only. It leaves out benefits, payroll taxes and what a clinician bills.

The other inputs are assumptions we chose to keep the sums simple:

• 3 clinicians, each writing 8 treatment notes a day, 220 working days a year. That is 5,280 notes.
• 10 minutes to write a note without a scribe. This is the number most likely to be wrong for your clinic, so measure it.

We found no peer-reviewed measurement of minutes per note in real physical therapy practice. The two results below are the nearest evidence, and each has limits that we set out next to it.

So what for you: treat the 10 minutes and the 8 notes a day as blanks to fill in from a week of your own diary, not as benchmarks.

Three scenarios from two studies

Scenario 1, no saving. In the NEJM AI trial, 238 physicians across 14 specialties were randomized to one of two scribes or to usual care. Nabla reduced time in notes by 9.5% against control (95% confidence interval, -17.2% to -1.8%, p=0.02). Microsoft DAX Copilot showed a change of -1.7% (confidence interval, -9.4% to 5.9%, p=0.66), which is not distinguishable from zero (Lukac et al., NEJM AI, November 2025). A tool that does not move note time is worth $0 in this arithmetic, whatever it costs.

Scenario 2, the real-clinic result. Apply 9.5% to a 10-minute note and you save 0.95 minutes. Across 5,280 notes that is 5,016 minutes, or 83.6 hours, worth about $4,060 at $48.57 an hour. The trial measured physicians' time in the electronic record note, so it says nothing certain about physical therapy notes, and the interval around 9.5% is wide, from 1.8% to 17.2%.

Scenario 3, the simulated result. In a JMIR preprint by Krogh and colleagues, an ambient scribe cut documentation time by an average of 35% across 13 simulated outpatient physiotherapy consultations (JMIR preprint, 2026). Apply 35% to a 10-minute note and you save 3.5 minutes. Across 5,280 notes that is 18,480 minutes, or 308 hours, worth about $14,960. The preprint had not completed peer review when we checked, the cases were simulated, and 13 is a small sample. Treat it as the top of the range, not the expected value.

So what for you: the spread between the rows is larger than any difference between tools on a price list, which means your own measured minutes matter more than a vendor's headline percentage.

The break-even line

Suppose a vendor quotes $100 per clinician per month. That figure is a placeholder, not a price we found, so use your real quote. For three clinicians it comes to $3,600 a year. At 81 cents a minute, the scribe has to save at least 4,447 minutes across the year to pay for itself, which is about 0.84 minutes per note on our assumptions.

Against that line, scenario 1 loses the full $3,600. Scenario 2 clears it by roughly $460. Scenario 3 clears it by about $11,360. A small saving per note is enough to break even and a large one is clearly worthwhile. The risk is paying for a tool that lands in scenario 1, or one that saves time in the note but adds it back in review.

That second risk deserves attention. A clinician must still read and sign every note, and our arithmetic counts only the time spent writing. If checking a drafted note takes 3 minutes, the saving shrinks by those 3 minutes. We cover the accuracy side in the accuracy-testing gap in AI scribes.

So what for you: ask for the price per clinician, divide the annual cost by $0.81, and divide again by your note count to get the minutes per note the tool must beat.

Saved time is not yet money

The wage figure prices the time at what you pay for it. It becomes revenue only if the minutes are turned into something: a bookable patient slot, a finish time that is earlier by half an hour, or an evening handed back. Eighty-four hours a year is about 23 minutes on each of 220 working days, shared across three people. That is a real gain for a tired owner, but it is not a free treatment slot.

If you intend to convert saved time into extra visits, replace the wage with your net revenue per visit and recompute. We have no verified figure for that, because it depends on payer mix and your own fee schedule. This is why the case for buying rests on your numbers, not on ours.

So what for you: decide before the trial what the saved minutes are for, because a saving without a use disappears into the day.

What the evidence points toward

Frankly, do not buy on a vendor's percentage, and do not dismiss the idea because one tool in one trial did nothing. The evidence says some scribes save measurable time, at least one did not in a randomized comparison, and no one has published a clean measurement for a clinic like yours. Run a measured trial before you commit.

1. Time 40 of your own notes without the tool, from first keystroke to signature, across all three clinicians. The average is your baseline.
2. Trial one tool for 30 days with an agreed stop date and no annual contract. Confirm a HIPAA business associate agreement is in place before any patient data is used.
3. Time 40 notes again, including the minutes spent reviewing and correcting the draft.
4. Compare the saving per note with the break-even line from your own quote. If it clears the line, buy. If it does not, stop.

If you want the choice of tool narrowed before you spend a month on a trial, our comparison of Heidi, Nabla and Dragon for physio notes sets out the differences. An AI Opportunity and Growth Assessment ranks scribing against the other places time leaks in your clinic using your own numbers, or you can book a free 20-minute call.

None of this is financial or clinical advice. The tool results come from physicians and a simulated study, and the wage is a national median.

Related: The accuracy-testing gap in AI scribes  ·  Bundled or standalone AI scribe for a physio practice  ·  74 days to Jan 1: 4 Medicare calls for PTs

Sources: BLS Occupational Outlook Handbook, Physical Therapists (median annual wage $101,020, May 2024; checked 26 October 2026); Lukac et al., NEJM AI, published 26 November 2025 (238 physicians, Nabla -9.5%, DAX -1.7% not significant), as reported in secondary coverage; Krogh et al., JMIR preprint, 2026 (13 simulated consultations, 35% average reduction, not peer reviewed). The 3 clinicians, 8 notes a day, 220 days, 10-minute note, $100 monthly price and hourly conversion are our own illustrative inputs, not published figures. Figures were checked on 26 October 2026.