UnityAI on AI for Smaller Practices: Data and Process Discipline, Coordinated Agents, and Top-Line ROI

This interview was sponsored by UnityAI and developed by Health Tech Nerds with full editorial control.

Much of the hype in the healthcare technology world right now is centered around AI’s ability to enhance, augment, or even replace clinical judgment. The extent to which AI is ready for tasks like refilling prescriptions, making diagnoses, or conducting curbside consults is often discussed and hotly debated, with little in the way of satisfying conclusions. It simply seems too early to tell.

Where AI appears to hold promise for return on investment is in the day-to-day operations of a healthcare practice. Independent and small-group specialists, dental practices, and primary care clinics deliver a significant proportion of the care in this country. Many deliver this care in strip malls and nondescript office buildings, running their front- and back-office operations with clipboards, fax machines, and landlines. For mid-size specialty groups and federally qualified health centers (FQHCs), the challenges are exacerbated by scale that increases complexity but is not large enough to achieve the economies of scale their health system and medical group competitors enjoy.

There’s a bipartisan consensus in Washington, D.C., and state capitols that these independent doctors and their clinics are a social good worth preserving. They make markets more competitive, and patients have a strong revealed preference for care that feels personal and local. But it’s hard to compete with larger, well-financed health systems and provider groups that can afford to build or customize solutions in-house.

UnityAI, which recently announced an $8.5 million Series A, is imagining something different for these doctors and their practices. Rather than building AI agents for discrete tasks, UnityAI is care teams put agents to work in a coordinated way across patient, team, and practice operations to reduce no-shows, increase new appointments, and free up staff to focus on caring for patients across a variety of clinical contexts. 

I’m excited to welcome UnityAI founding team members Edmund Jackson, Cody Hall, and Jason Parker.

01:17

Edmund Jackson: Healthcare is incredibly complicated. That's not news to anybody. Your large operators in the space solve all these issues of regulatory, compliance, legal, HR, logistics, supply chain with armies and armies of humans. There are these huge manual processes that dissolve out the complexity, but they're very expensive. Large health systems, large platforms can bring that to bear.

Your smaller clinics, your rural clinics, they don't have that platform and infrastructure, and so they really struggle. Everybody ends up wearing four hats and never sleeping.

The opportunity for technology is to identify that quotidian work and make it technological. That's really what we're trying to bring: replacing manual processes with tech-enabled processes and tech-enabled services so that we can level out the playing field and provide healthcare at a lower cost with lower overhead.

02:41

Edmund Jackson: All three of us are HCA alums. An outstanding company. Truly the standard of excellence in healthcare, without question.

The key lesson that we bring forward from HCA is that consistency is the precursor for excellence: in quality, financial earnings, and everything else. If you can drive out variation in the processes, the operational processes as well as the clinical processes, you can deliver excellence. As I said earlier, that's often done through big, manual, human-driven processes today across healthcare.

Our opportunity, our vision, our goal is to create an agentic alternative to that, one that can be offered at scale, at a lower price, and in different settings of care. Things like veterinary clinics or dental clinics, which couldn't otherwise do this, with agents, they now can. And if you can create consistency, repeatability, and scalability, all these technological virtues, in those settings, you can reduce the cost and increase the performance.

Jason Parker: I think you're absolutely right to point to that HCA background. Very much the focus on consistency, standard playbooks, just how we're going to do things. As we go out to the market, we see a lot of variation in that area. Part of what we're finding is that people are actually interested in getting help from tech enablement to drive that kind of consistency.

One of the real challenges is that what's maybe not obvious about HCA is that they've done a lot of the data work under the hood to get things into a central place, governed by standards for that. That's uneven across healthcare today. Often, we're coming in, talking to people about the very cool thing that's possible, but there are some foundational prerequisites that need to happen first.

I'd say that's something I'd want people to know: govern the data, govern the processes. Have some sense of what you want that consistent representation of the work to be. That's what makes it understandable to these agentic flows. The really big, very competent players have invested a lot of time in this for a long time. Not everyone can go back in time and do it for 15 years, but putting that effort in up front today will pay dividends as we see AI continue to advance.

05:53

Edmund Jackson: FQHC is a really, really challenging market. We all know the financial headwinds those guys face. And a lot of the services that Peregrine offers, they're providing telebehavioral health across multiple different provider groups, across multiple different EMRs, across multiple different states. It's just very diffuse and distributed. Creating that standardized engagement is the difficulty they face. How do you make it the same everywhere? How do you provide consistent quality? We engaged with them to create Emma, which provides that.

Cody Hall: UnityAI has the FHIR model of interoperable data at our core. Everything we do internally to the company is centered around that. Where we initially had some agents defined in code, they're now defined as FHIR resources. What we were able to do and prove out with the Peregrine work is define a FHIR bundle specification, along with structure definitions for FHIR resources, which they can submit to us via an API. Our internal FHIR store generates an event that becomes a task, and AI picks up on it. The communication back to the Peregrine team about what was communicated, what the outcomes were — all of that is accessible to them as well.

We see that as an interesting way to start leaning into that standard. More and more EMR platforms are starting to speak a little bit more FHIR these days. It gets the ball rolling for how agents in healthcare start to reason over FHIR resources. You get the REST semantics, you get the data modeling and schema, and you have the ability to define some structure and standards there. It becomes a nice ecosystem, and we're really interested in pursuing that more broadly.

Edmund Jackson: Data, semantics, and interop are the precursors for AI or data science. FHIR was conceived as a data interchange for exchanging data, but it gives you that output of data governance, or you require data governance to even be able to speak in FHIR. So that's been done. And although FHIR was intended for data interop, what it actually unlocks is all of the actions in the agentic AI world that we're entering into, which weren't conceived of at the time, but is super exciting.

Jason Parker: They've modeled the domain super well, and so that means an agent can understand the domain super well. It becomes like your knowledge graph for the current state of the world. So the agent then knows: here's how I want to change that state of the world.

Jason Parker: Getting back to the question about Peregrine… I think their being an FQHC has been less of a thing I've thought about explicitly and more about them being in the behavioral health domain. There's a lot of sensitivity in that area. Any patient entering the healthcare system feels vulnerable at some level. I think behavioral health is one of the more sensitive ones.

So, thinking about how empathy exists in the AI world, computers don't really feel empathy, and yet we want to be able to express that to patients as they're going through a hard thing, so they feel comfortable engaging with us to take the steps we're hoping they'll take. That's been an interesting design consideration: how do we have it express empathy in a way that's good for a patient without being insincere? Because it's not feeling. These are interesting questions that we find ourselves debating with the Peregrine team and some of our other customers about what the right level is to dial in there.

10:52

Jason Parker: We don't want a world where agents are off over here doing a set of things, and we're not quite sure how we interact and have awkward ergonomics with that. We need to think about how it fits into what the team does today. My goal is that you can deploy these into healthcare settings to help with these operational tasks, and they behave a lot like your coworkers behave. Maybe they're faster, maybe they've got all those benefits, but you want those ergonomics to be good. Part of what that implies is that you've got multiple agentic entities out there doing their thing, and they should coordinate with each other as well.

I'll throw out a book recommendation. I talk about this book often: it's called Prediction Machines. It's about data science and AI from an economic standpoint. One of the great abstractions it proposes is this distinction between prediction and decision, or judgment. Often, we combine those things in our mind: we're implicitly making a prediction and then a decision about it. Their point is that you can separate these concepts at some level. When we think about that, we start to go: how do we atomize the world? How does that work break down into more discrete pieces? And that's how we think about agents handling those discrete pieces and informing each other.

To give you a concrete example: when we think about managing a clinic, we think of two schedules that are actually intertwined and need to be balanced. There's the patient schedule of people needing to come in for services, and there's the staff schedule of people who are going to provide those services. Kind of supply and demand.

As one of those changes, the other probably needs to adapt. In the classic case, a staff member calls out for whatever reason and can't make it in. What are we doing to respond? Some clinics can smooth and load balance and handle it; others cannot. It becomes a huge time drain and energy drain for the manager of that clinic. They're having to figure out the implications, calling and texting a bunch of people at 7 a.m., trying to get them in by 8:30 or whatever the time's going to be. Huge amount of toil and effort. We might not find anyone, so now I'm tasking someone to call all these patients and ask them to move. They're going to be unhappy about it.

Those same kinds of things happen on longer time horizons — they're not as dramatic, but someone's going to take PTO at some point, I don't have any coverage, I've already got things scheduled. If I can see PTOs coming and I'm also running the scheduling for this clinic, I'm going to avoid that week for scheduling patients. I'm going to try to overload a little bit the week before, push a few a little bit further beyond. Maybe I can find some part-time labor to cover a portion of the week.

If we could all sit down and think about it very carefully and calmly, we'd make lots of good decisions, but nobody's real job works that way. They've got just a ton of urgent things coming up all the time. That's the goal we're aiming at: can we have agents defined to handle these narrow pieces, with orchestration-level things saying, here's what we're perceiving is happening in the world, here's options for how we might proceed.

I think there's going to be a world where we're in a human-in-the-loop mode for a while, presenting those options, letting the human say "yes, I like that," and we're executing. I think we could get to a place where they kind of beforehand say "in these scenarios, do these things" and we're able to execute. That's what's going to give you not just linear value but something nonlinear over time as you start hanging those things together and they feed off of each other.

16:16

Jason Parker: On the patient side, that's probably where we've built up the most. There's what I call the full patient scheduling lifecycle. There's the appointment that needs to be scheduled because there's a referral or a routine visit. Let's get that thing scheduled. There are all the pieces about getting new patient insurance checked, all those different things that happen in the up-front process.

If you listen to calls done today by humans, you'll hear a lot of friction, "wait a minute, I need to get this record to pull up," and you're listening to them wait for their computer to load a screen. That experience can actually be streamlined because we're not waiting for the UI to render; we're just hitting a back-end system directly.

That lifecycle continues: schedule the appointment, confirm a couple days beforehand, and remind them. If they're not going to be able to make it, we can reschedule. That's great because it tells us there's a slot open that we might be able to fill with somebody else. There's the follow-up side — you came to your appointment, let's do a follow-up survey. You may have missed the appointment, let's call you back and get you back on the schedule. And then there's the front desk piece: can we take inbound calls and handle the routine things so that front desk staff can be focused on the in-office experience of the patient rather than on the phone all day?

For practice, the example I gave about balancing patient and staff scheduling is really how we think about that offering — that type of system-level optimization about everything about how my clinic is working.

For the team, it's the staff-facing version of scheduling. They have needs, life happens. "I can't make it to this one. I'd like to swap a shift." A lot of places, it's a dry-erase board. It's a Google spreadsheet. We just all text each other. I think SMS must be the main way scheduling happens for a lot of teams. That's probably good in a lot of ways, and then occasionally leads to disasters because I told this person and that person, but not the person who needed to know.

We see that going toward a staff concierge role… able to take your request, let you know the state of things, ensure the manager knows they need to take action, or maybe take it on their behalf and communicate back so everyone's on the same page. So much of where healthcare breaks down is coordination and communication amongst people. Can this play the role of ensuring we know what your intent is, communicating that back out, and confirming that it's been done for everybody?

Edmund Jackson: It really is about the coordination and the breadth. We have a few really deep examples — in oncology, for instance, matching the patient and the provider is an incredibly deep and complex problem. We have some experiments going right now where you're trying to match a sub-sub-specialist with a patient. Currently, this happens with somebody with a high school education, and everyone's trying to figure out what their glioblastoma is. Agents can really do this accurately and well. So that's an experiment, but really the breadth is the coordination across space and time.

21:50

Edmund Jackson: You've got it right. Some large health systems are like large boats, they take a while to turn, but when they do, one degree of turn is significant. But it takes a while.

Smaller, more nimble organizations are going to start to show these returns, and already are. If you look at things like revenue cycle, there's been a ton of return on investment on the provider side from being able to automate and agentify those processes. The return for clinical scribes may not be financial, but certainly in terms of quality and quality of life for providers, the story has written itself already. In healthcare, we're absolutely seeing returns on these things already.

The type of operational AI that we're talking about, paradoxically, might actually not reduce costs always. It actually improves performance. You see it on the top line rather than the bottom. Some of our customers, we'll work with one where there's a call center, and we enter into the call center. We take on the scheduling and confirmations. The people that were doing that are now providing higher-level services, and so the patient experience improves, the quality improves, it doesn't necessarily make a financial improvement directly.

This is Jevons' Paradox. If you make a thing cheaper, you consume more of it. If we make patient engagement cheaper, there's actually more consumption of it, and the upside is increased volumes.

Jason Parker: Part of that's also the efficiency play; it's not necessarily that your cost is reduced, but your volume might increase massively. Since we mentioned Peregrine earlier, part of what we were doing was calling patients to get their intent to schedule, and we were handing all of them off to their call center. Instead of their call center folks leaving 15 voicemails every hour, they're scheduling 10 patients every hour. The footprint of the call center stayed the same; the volume they're handling has gone way up. The top-line return has been high.

Jason Parker: I think the place you'll see it the most is the places that are focused on improving things about their operations, rather than solving the problem of "how do I use AI to improve my operations?" It's the same fundamentals it always was. People need to be focused on how their business works. That's why you're going to see quicker return on investment with some of these smaller operators. They're lean, they're trying to survive in the world, they're very focused on the actual problem right in front of them. If AI is the answer, that's great. If it's not, they're doing a different thing. They don't feel the same "I have to make it AI" pressure.

I think some of the big companies feel that pressure because that's the narrative out there, that it's got to be AI that drives the next step change. That becomes the goal, rather than step change in the actual operation itself. The places where you'll see that 1% that shoots them way up, it's going to be the places that do both.

Cody Hall: I was just going to add a quick perspective on what the scaled communications we're able to do with AI offer. We'll make a phone call and say, "hey, are you coming to your appointment?" That person will say, "no, I can't make it." So we transact back against the source system. This slot is now available; this appointment is canceled. Traditionally, someone might not be able to offer that slot to a new patient for days, or even never. But now, we can keep a list of who is requesting an earlier appointment and go make what I call micro-negotiations — the scaled communication where now the schedule's constantly repaired. That increases your schedule density, your patient access, and availability.

I went to get diagnostic imaging, which turned out to be completely benign, but it was a four-month wait. I got to the front desk that day and said, "what took so long? Why didn't you call me about this referral?" And the weird smile was like, “I don't know what the black box is behind the scenes, but there's a file folder somewhere of the stack of referrals, and nobody can call them all.” Just having a machine that can make contact with all those patients, get in touch with them, see their preferences for when to schedule, and then actually transact that and get it booked. It's a huge superpower that we have now in healthcare, and it improves people's access to care.

28:32

Edmund Jackson: Raising a round is the worst thing. It was super not fun.

No, we spoke to a lot of investors and learned a ton. The company, we're a Nashville company. We're 30 people in Nashville. Those investors you described are primarily Nashville. This is a Nashville phenomenon. And what was resonating was that we're not a bunch of tech bros who've fallen out of Silicon Valley, had a bad wait, and said, "I'm going to fix healthcare." We're deeply steeped in healthcare. We understand what's actually wrong. Problems that can actually be fixed with the technology that we have today. And we're not admiring the problem, we're actually set about solving it. That really humble, pragmatic approach is what resonated. There's a real problem, we can really fix it, and we're in the right place with the right background and the right team to do so. Moreover, the scalability of this is really exciting from an investment standpoint.

Edmund Jackson: As for putting the money to work, it's a growth story now. We're primarily, at this point, an engineering company. We are health tech nerds, physically, here in Nashville, writing code. We have an outstanding, second-to-none healthcare technology AI team here. What we don't have is a go-to-market machine, or a sales team, or a whole lot of customer support, or everything else that turns a product into a company. That's really the focus right now: how do we become a company?

Well, we ran our first ad last month. That was good. People clicked on it.

We're turning ourselves from just a product group into a real company. You should expect to see us in the news, expect to see us have a presence and be around. The product roadmap is well understood. We know what we're doing from that perspective. Unlike many companies, we have more than we've been telling anybody about. We're trying to catch up with telling people what it is that we're doing and really become present in the marketplace.

Visit unityai.co to learn more and get in touch.