AI in Your Healthcare Practice: What I’m Actually Seeing and Doing Right Now

Q: What’s your read on the AI scribe space right now?

The AI scribe space is overcrowded, and a lot of companies that dove into it are now struggling to find takers. Here’s why: the low-hanging fruit was ambient documentation — record a session, turn it into a note. That made sense two or three years ago. But platforms and EHRs have since built that capability in-house, and now most practice owners aren’t shopping for a separate AI scribe tool because their EMR already does it. If you’re a startup entering the market right now with AI-scribe-first positioning, you’re fighting for a shrinking piece of a pie that the major platforms are eating up themselves.

I’m also seeing what I’d call acqui-hires — where a larger EMR company buys a smaller AI company, not really for the tech, but for the team. The marketing makes it sound like a big win, but behind the scenes it’s more like: we’ll take the engineers and fold them in. It’s not failure, but it’s not exactly a success story either. Expect to see more of that.

Q: Where do you actually see AI adding value in a PT practice?

I think about it in two buckets: patient-facing and back-office.

On the patient-facing side, there’s real opportunity in making the intake and onboarding experience smoother — automated messaging, digital check-in, guided triage before a patient even sees a clinician. Done right, this doesn’t feel robotic; it just removes friction. I actually just worked with a company that developed an agentic solution to obtain consent, onboard, and provide initial clinical education to patients enrolled in chronic condition management programs. The system gets the patients onboarded, tracks the data, and then documents it for the clinician ahead of the patient’s in-person clinic time. It save the practice admin time, provides more touch points for the patients, and allows the provider to get more informed on the patient before walking into the exam room.

On the back-office side, credentialing, billing, and authorization workflows are where I think AI has the most legs right now. If you can tighten up your eligibility verification and pre-auth processes so patients aren’t getting surprise bills and you’re not drowning in denials — that’s where the ROI is. And with payers actively using AI to screen claims and find reasons to deny, you almost have no choice but to respond in kind. It’s becoming a digital arms race, and practice owners who aren’t paying attention are going to fall behind.

Q: Where are you drawing the line — areas where you won’t use AI?

Billing decisions. Full stop.

The idea of an AI system whispering in my clinician’s ear — ‘hey, you should probably bill this code instead of that code’ — makes me uncomfortable, even if the suggestion is technically correct. That’s a conversation I need to be having with my team directly. It touches compliance; it touches clinical judgment; and it touches the trust relationship between me and my clinicians. I’m not willing to outsource that to an algorithm, at least not right now.

What I am okay with is surfacing information and letting the clinician decide. If our system sees that someone entered a code that doesn’t match the documentation and flags it as a ‘hey, double-check this’ — that’s supervision, not replacement. There’s a big difference between a tool that nudges and a tool that decides. I’m building toward the former.

Without getting too technical, it’s the idea of “supervised” vs “unsupervised” AI implementation. For things involving care decisions, billing, and clinical practice, I’m squarely in the camp of using supervised AI only.

Q: What’s your take on clinicians using AI tools on their own — without formal approval?

It’s happening. Especially in private practice where there’s less infrastructure and oversight than at a hospital system. I’ve heard from practices where clinicians are recording sessions on their personal iPhones, transcribing them, running them through ChatGPT for a summary, and copy-pasting into their EMR at the end of the day. And honestly? I get it. The impulse to save time is real.

But here’s the thing: as the practice owner, you are on the hook for that. Even if you didn’t know it was happening. So at minimum, you need a clear policy in your employee handbook that spells out how patient data can and can’t be used. We incorporated it into our HIPAA and PHI policies — patient data stays in the EHR, full stop. It doesn’t go into ChatGPT, it doesn’t go into any non-approved platform. That policy exists partly to protect patients, and partly to protect us if something ever goes sideways.

The other thing I’d say: don’t just clamp down without offering something better. If your clinicians are gravitating toward these tools, that’s a signal worth paying attention to. Figure out how to meet that need in a compliant way — because if you just lock everything down and don’t provide a solution, you’ll lose good people over it.

Q: How did you introduce AI scribing to patients at your clinic?

Very deliberately. We have a small practice, with a loyal patient base, and the last thing I wanted was for people to feel like we were experimenting on them or treating the visit like a data collection exercise.

The framing we used — and I really believe this — was about human connection. When a clinician says, ‘I’m going to hit record so I don’t have to be nose-deep in a computer while we’re talking,’ that’s not a tech pitch. That’s a promise of a better conversation. We made sure every patient had signed a consent form before we used any ambient recording. But beyond the paperwork, we thought carefully about how to communicate the why. Not ‘this makes our life easier’ — that’s true but it’s not the point. The point is: ‘This means I can actually look at you while we talk. I can be present with you.’

That message landed well. If you’re rolling out something similar, I’d encourage you to lead with the patient benefit, not the operational efficiency.

Q: How do you stay plugged into what your team is actually doing with AI day to day?

Two things that have worked for me: showing up and asking.

I’m in the building most days, even though I’m not treating patients anymore. I do a walk-through, I watch interactions, I pay attention to what’s on screens. People will tell you one thing in a meeting; their behavior will tell you something else. You have to be present enough to see both.

We also run a bi-weekly all-hands meeting, and the tech stack is a standing agenda item. Not in a formal way — it’s not ‘AI update #14.’ It’s just part of the regular conversation: ‘Have you guys tried the new feature? What do you think? Is there something you’ve been using on your own that you want to explore together?’ Keeping it casual lowers the barrier for people to actually tell you what’s going on.

For practice owners who feel out of the loop: the answer isn’t a big town hall about AI. It’s smaller, more frequent conversations embedded in how you already operate.

Q: What does your device and data security setup look like?

Company-issued computers only. No personal devices for anything that touches patient data.

I know that sounds rigid, but my background is federal government and management consulting — my SOP brain doesn’t mess around with HIPAA exposure. The standing rule is simple: if you’re doing clinic work, you’re on a clinic computer. If you’re working from home, you’re taking a company machine and connecting through the VPN. We don’t have a formal refresh cycle — when a machine starts to go, we retire it properly (factory reset, physical destruction of the drive, the whole thing) and replace it.

I’ve seen what happens when someone accesses PHI on an unencrypted network at a coffee shop, even with the best intentions. It only takes one complaint to trigger a HIPAA enforcement action, and those are not fun. Don’t give anyone a reason to look at you.

Q: What does the next year look like for your practice?

I think what we’re really moving toward — and I’ve been saying this for a while — is a true hybrid model of care. The ability to meet a patient where they are and give them the right kind of support at the right time. That might be a traditional PT session. It might be a digital triage tool that helps someone understand their symptoms before they ever book an appointment. It might be a remote monitoring check-in between visits.

The bigger shift I’m watching is the move away from fee-for-service and toward direct-to-employer and value-based models. Some of the platforms being built right now will eventually push patients directly toward physical therapy as a first line of care — which is great for us. But some of them will let large health systems self-refer and keep patients internal, which cuts us out. How that plays out over the next 12 to 24 months is going to reshape referral pipelines across the industry.

I’m not waiting to see which way it goes. I’m thinking now about how my practice fits into that broader landscape, and positioning accordingly.