Twenty-two minutes a day doesn’t sound like much. It’s about how much documentation time an AI scribe saved each practitioner in a 2025 randomized trial at the University of Wisconsin.¹ Over a year of clinic days, it adds up to roughly 90 hours per physician, a little more than two working weeks. For a medical group with 500 physicians, that’s about 44,000 hours, close to what 20 full-time physicians work in a year.
In the group’s next leadership meeting, they’ve just gained about 20 physicians’ worth of time without hiring anyone. Now, everyone at the table has a plan for it. The CFO wants to know how many more visits it means. The chief medical officer wants longer appointments for patients with complicated conditions. The physicians, if anyone asks them, want to stop charting at 10 p.m. Each of those is a reasonable use of the time, and the software doesn’t choose between them. What they decide will determine who pays for these products and how much, which is why founders should be thinking about it now.
As background, a 2016 time-and-motion study found that physicians spent 27% of their office day face to face with patients and 49% on EHR and desk work, then reported another one to two hours of it most nights.³ A 2017 study of 142 Wisconsin family physicians measured 5.9 hours a day inside the EHR, 1.4 of them after the clinic had closed.⁴
Someone decides what the time is for
When I used to run a technology company and a team got faster, the extra capacity didn’t automatically belong to the team. Someone decided what it was for. Ultimately, the company’s incentives will steer the choice.
Clinics work the same way, and the person deciding is usually not the physician. In most large medical groups, an operations team behind the scenes will build the visit templates. This means that the accurate version of the vendor pitch would have AI giving time back to the organization. The organization, in turn, decides how much of it the physician keeps.
As referenced, providers can spend additional time on longer patient visits, additional visits (which, in a fee-for-service practice, result in more billable revenue), or it can use the savings to lower cost. Or the physician can keep it, by going home earlier and not charting at 10 p.m. Vendor marketing implies this choice, but it is also the one with the least incremental revenue or profit (though it does have additional doctor satisfaction which will reduce attrition).
The buyer’s business model makes the choice
The outcome will depend heavily on the buyer’s business model.
A fee-for-service health system is paid per visit and per procedure. Additional time will likely result in more visits and more complete billing. If physicians are paid on productivity, some of them will fill the time themselves because it raises their pay. Salaried physicians with fixed panels have little reason to see more patients and are more likely to take shorter days — until their panel size grows.
Risk-bearing organizations are different. A medical group paid a fixed amount per patient, or an integrated system that is both insurer and provider, keeps what it saves on hospital and specialist costs. For that buyer, a longer visit that identifies a problem early can pay for itself. More complete documentation also raises risk-adjusted payments in Medicare Advantage.
And a system that simply can’t hire enough primary care physicians may care primarily about retention. For that buyer, the scribe’s value is measured against the cost of recruiting a replacement.
One core problem is that this creates an unrealistic optimization. Twenty-two minutes a day spread across twenty visits is about a minute per visit. This isn’t enough time to open a new appointment slot, and it doesn’t make a visit noticeably longer. This means the actual value is significantly less than the potential value.
Additional value after the sale
In a positive development for providers, the Wisconsin trial referenced above also found that professional coders judged that the diagnosis codes in scribe-assisted notes had improved. Some of it is improved accuracy. But payers see the same claims data. CMS already applies a coding-intensity adjustment to Medicare Advantage risk scores, and commercial payers audit coding patterns. If a product’s return on investment depends mostly on higher coding, part of that return will likely be negotiated or regulated away within a few years.
The core capability is also becoming standard. Epic has announced its own ambient documentation, and other large platforms are moving the same way. When transcription and note drafting come bundled with the EHR, pricing power erodes without unique capabilities.
Physicians also do get a say. The UCLA physicians used their scribes in about a third of visits even when the software was free to them. Adoption is voluntary in practice. If every recovered minute simply turns into different work, the retention argument for the product is significantly undermined.
What I’d tell a founder
Start by working out which of these outcomes your buyer is actually paid for, and sell accordingly. A figure for minutes saved is useful evidence, but it is hard to put in a budget. Be careful about promising an outcome the buyer can’t use. A group paid per visit has little reason to buy longer patient conversations. If your product’s value depends on a particular use of time, sell it to buyers who are rewarded for that use.
Measure where the time went as well as how much was saved: visit volume, after-hours EHR time, clinician turnover, quality scores, and coding accuracy. A customer who can see the result is more likely to renew. A customer who only sees a time-saved figure will eventually ask why the budget hasn’t changed.
If your product touches billing, its accuracy will be audited, so treat auditability as part of the product from the start.
Finally, leave some of the time with the physician. The buyer is usually an administrator, but the user is a clinician who can stop pressing the button. If utilization is too low, renewals will be more difficult.
References
- Afshar M, et al. “A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-Being.” NEJM AI, 2025. PMID 41625485.
- Lukac PJ, et al. “Ambient AI Scribes in Clinical Practice: A Randomized Trial.” NEJM AI, 2025. PMID 41497288.
- Sinsky C, et al. “Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties.” Annals of Internal Medicine, 2016. PMID 27595430.
- Arndt BG, et al. “Tethered to the EHR: Primary Care Physician Workload Assessment Using EHR Event Log Data and Time-Motion Observations.” Annals of Family Medicine, 2017. PMID 28893811.
