This piece was co-written with Rik Renard, who shares weekly insights on healthcare AI and strategy on LinkedIn.
AI scribes increased physician productivity by... $58 per week.
That's what the new JAMA Network Open study actually found. UCSF tracked 1.2 million encounters across 1,565 physicians and found AI scribe adopters saw 1.81 more wRVUs - translating to roughly $3,044 annually.
For context, AI scribe tools cost health systems $1,200-3,600 per physician annually. So we're looking at roughly break-even economics based purely on physician productivity.
What the study found is that, physicians who use an AI scribe have:
- 0.80 more patient encounters per week (very modest!);
- 1.81 RVU increase per week; and
- No change in claim denials (…yet)
The relatively "meh" results with regards to productivity gains are in line with the NEJM AI randomized trial that showed a tool like Nabla saves +- 23s per note while DAX saves 5s per note (not even statistically significant). And as an aside to this finding, we’d be curious to see this metric broken out by engagement cohort. For instance, a Nabla super user is almost certain to save significantly more time than a non user.
But here’s what the study is missing. Measuring AI scribes purely on physician productivity misses the significant positive downstream impact and associated externalities ambient scribing creates.
The real value isn’t in “doctor sees one extra patient.” It’s in what happens three departments down the line that nobody’s studying. We recently read a good example from Dr. Jared Dashevsky (my guy!) which emphasized this point:
- Ambient AI listens to the conversation
- Patient mentions they've smoked for 20 years, they're 55, still smoking
- The AI automatically pends an order for a low-dose CT chest scan
-
This gets done within the health system (they get reimbursed for that)
- They find a nodule that's large enough
- They refer them to their own pulmonologist and oncologist
-
Everything stays in the system
(By the way, he also just wrote a great overview of how he uses Doximity AI tools in his workflows - good timing given the Doximity vs. OpenEvidence clown fiesta going on).
Same logic for suggesting preventative screenings, flagging specialist referrals before patients leave, better coding that captures higher reimbursement, quality metrics for value-based care contracts.
(I know this sounds cynical because it's all about revenue cycle management, but that's actually the game that needs to be played in healthcare.)
tldr: AI scribes get sold through the finance department on boosts to physician productivity when in actuality these investments deliver returns through workflow improvements three departments down the line - downstream effects that we're not studying.
