Artificial intelligence (AI) is rapidly moving from a future-facing concept to a practical necessity in ophthalmology practices. As providers face growing patient demand, increasing administrative complexity, and ongoing workforce challenges, AI-powered tools are emerging as valuable solutions for improving efficiency across clinical, operational, and financial workflows. In this Q&A with Ophthalmology Management, Nitin Rai, chairman of EVAA.AI and MaximEyesAI, discusses where practices are seeing return on investment from AI today, how to distinguish meaningful innovation from added complexity, and what role AI operating systems may play in addressing the specialty’s long-term capacity challenges. He also explores common misconceptions about AI adoption and outlines the safeguards practices should prioritize to maintain patient trust, data security, and physician oversight.

Ophthalmology Management: Which practice management functions are currently seeing the strongest ROI from AI?

Nitin Rai: Revenue cycle, documentation, and patient communication—those are the 3 areas where AI is paying for itself fastest. EVAA's 3-product suite—Billing Assistant, Scribe, and Virtual Assistant—maps directly to those workflows. We measure success by staying close to our clients after implementation and tracking what's actually changing in their practices.

Hermann & Henry Eyecare, a 25-year MaximEyes customer running a 3-doctor, 18-person practice, is a good real-world reference point. With our Billing Assistant, they saved more than 5 hours of administrative time daily, cut manual workload by over 50%, pushed accuracy past 99%, reduced claim errors by 40%, and went from claims processing in days to claims processing in hours. Eligibility verification—historically an 8-to-9-hour-a-day job—runs in real time in the background. Billing Assistant also auto-posts payments via ERA and EOB integration and generates a prioritized denial worklist, which means staff are no longer buried in posting and chasing, they're focused on the denials that actually move the needle. The downstream effect is that accounts receivable decreases over time because the entire revenue cycle is operating as a holistic, connected system rather than a series of manual handoffs. Even end-of-day reconciliation, which used to take hours, now takes 5 to 10 minutes and balances to the penny.

Documentation is the second major win. We’ve learned that AI scribes and ambient clinical intelligence are saving physicians up to 2 hours of charting per day, and practices are finishing notes before leaving the exam room and seeing 3 to 4 more patients per doctor per day without adding staff.

Patient communication is the third, and this is where a virtual assistant comes in. Practices using our platform have a patient messenger embedded on their website that auto-responds to inbound SMS, transcribes and replies to voicemails, books, reschedules and cancels appointments, captures and OCRs insurance card images into the right MaximEyes queue, provides order status updates, and handles balance lookup and payments through WorldPay with everything posting back into MaximEyes. When it can't answer or a patient asks for a person, the thread is flagged “action required” and routed to staff.

Prior authorization is the next domain we're tackling—automation support for prior authorization is coming to Billing Assistant in a future release, which will reclaim significant time for both staff and providers stuck filling out tedious payer forms.

OM: How do you tell meaningful AI from AI that just adds complexity?

NR: The simplest test is to ask, “Does the tool remove friction, or just relocate it?” A lot of AI in eye care is highly accurate but isolated—it optimizes one step while creating new handoff problems with everything around it. Real value shows up when AI operates as a connected operating system: a patient texts the office, the virtual assistant books the appointment and captures an updated insurance card, eligibility runs automatically, prior authorization runs in the background, documentation flows out of the exam via the scribe, claims get scrubbed and submitted, payments post automatically, and denials route to a prioritized worklist—all without staff stitching it together, all in 1 record in the electronic health record (EHR).

The opposite pattern—and this is what most practices are accidentally buying—is a chatbot that takes appointment requests but doesn't write to the schedule, a scribe that produces notes you still have to reformat, a billing tool that flags problems but doesn't post the payment. Each of those tools may be individually impressive, but if their output still has to be moved, reconciled, or babysat by a human, you're buying complexity, not efficiency.

The other thing I'd add: don't measure AI by the vendor's slide deck—measure it by what your peers actually report after they implement. We make a deliberate practice of going back to our clients after deployment and asking what's changed: hours saved, errors avoided, phone volume reduced, AR trends, staff sentiment. That feedback loop is how we separate the features that genuinely move the needle from the ones that just look good in a demo, and it's how we keep refining the product. The honest question for any vendor is: "Does this make the next step in my workflow better, or just make this step faster?"—and the honest follow-up is: "Who in my specialty has been using it for 6 months, and what do they say?"

OM: Where will AI have the greatest impact on workflow optimization over the next 3 to 5 years?

NR: The supply-demand gap is the forcing function, and the numbers are stark. HRSA projects a 12% decline in full-time-equivalent ophthalmologists by 2035, while demand climbs 24% over the same period—a net loss of roughly 126 physicians a year.¹ Rural access is collapsing faster than the headline numbers suggest: workforce adequacy in rural areas is projected to fall to 29% by 2035, compared to 77% in metro areas. Meanwhile, an aging population is driving annual increases in the conditions that consume the most chair time—cataracts +2.5%, glaucoma +2.3%, macular degeneration +2.9%, diabetic retinopathy +2.7%. By 2030, more than 1,200 fewer doctors will be managing an estimated 6 million cataract surgeries, up from 5 million in 2025.² You cannot hire your way out of that math.

AI is the only realistic force multiplier, and a meaningful part of its value is that it doesn't keep office hours. An AI operating system works 24/7—a patient can book an appointment at 11 PM, get an order status at 6 AM, leave a voicemail at midnight, and have it transcribed and replied to before staff arrive in the morning. That's not a luxury feature; it's a capacity expansion you can't buy by hiring.

The biggest impact over the next 3 to 5 years will come from AI operating systems that own the full visit lifecycle: a virtual assistant handling inbound calls, texts, voicemails, web chat, and appointment booking on the front end; autonomous screening for diabetic retinopathy and glaucoma at the primary-care or pharmacy level so specialists see only high-risk referrals; ambient documentation in the exam room; AI-driven intraocular lens calculations reducing surgical revisions; automated prior authorization eliminating one of the most-hated administrative tasks in the practice; scrubbed claims; automated payment posting; intelligent denial worklists; and home-based monitoring that triggers visits only when patients actually need them.

EVAA is built to be EHR agnostic—it integrates with any EHR or practice management system, which matters because most practices aren't going to rip-and-replace to get AI. Practices that move from point solutions to an orchestrated platform will see real throughput gains, decreasing AR, and revenue capture that used to slip through the cracks.

OM: What are the biggest misconceptions physicians have about implementing AI?

NR: Four misconceptions keep coming up. First is the assumption that more AI tools equals better outcomes—practices buy a scribe, then a billing tool, then a chatbot, and end up with more systems to manage, not fewer. The value isn't in the tools, it's in how they connect.

Second is the fear that AI is going to replace clinical or front-desk staff. What we actually see is the opposite—as Dr. Jay Henry at Hermann & Henry put it, "AI hasn't replaced our staff—it's empowered them. It lets everyone focus on what really matters: the patient experience." The virtual assistant doesn't eliminate the front desk; it eliminates the queue of routine calls, voicemails, and "what's my balance" texts that prevent the front desk from doing higher-value work. The billing assistant doesn't eliminate billers; it gets them off payment posting and onto denial follow-up, where their judgment actually matters. The scribe doesn't replace physicians; it gives them back 2 hours of their day.

Third is that AI will route emergencies the wrong way. Every conversation in the virtual assistant is monitored for medically urgent language—if a patient describes something concerning when booking an appointment, the virtual assistant immediately redirects them to call the office or seek urgent care, and tags the appointment for staff triage. AI for routine workflows does not mean AI making medical decisions.

Fourth, and most important clinically, is the belief that large language models (LLMs) alone are enough for health care. They're not. Clinical-grade AI requires a hybrid stack—rules engines for guardrails and protocols, machine learning for pattern recognition, LLMs for natural language, and proprietary clinical data to ground all of it. Anyone selling pure-LLM clinical AI is selling risk.

OM: What safeguards should practices prioritize for patient trust, data security, and physician oversight?

NR: There are 4 priorities. First, insist on a clinical rules engine underneath any AI that touches care decisions or patient interactions. Rules give you deterministic guardrails, enforce established protocols, and minimize variability—repeatability is non-negotiable in clinical workflows, and it's equally important for patient-facing channels where consistency builds trust, and for revenue-cycle workflows where coding and payer rules are unforgiving.

Second, build human handoff into the design, not as an afterthought. EVAA's Virtual Assistant is built so that any conversation it can't confidently handle—or any patient who asks for a person—gets flagged “action required” and surfaced in the staff work queue, and AI is paused on that thread until a human takes over. Medically urgent language triggers the same escalation automatically. AI shouldn't be a wall between the patient and the practice; it should be a triage layer that protects clinical judgment.

Third, demand auditability. Every AI-generated note, claim, payment posting, or patient interaction should be traceable: what data went in, which model produced it, and which rules validated it. That's how you maintain physician oversight and defend the chart—or the claim—later. This matters especially as autonomous diagnostic AI for diabetic retinopathy and glaucoma reports sensitivity above 98%—accuracy that highly still requires a clear human-in-the-loop for cases that fall outside the algorithm's confidence band.

Fourth, treat data security as table stakes. EVAA is HIPAA-compliant with enterprise-grade security—end-to-end encryption, per-customer data isolation, authenticated access. These aren't differentiators, they're the floor. Practices should be asking every AI vendor about PHI handling and what data, if any, is shared with third-party model providers.

Fifth—and this one gets overlooked—pick a vendor who stays in the room after the sale. We measure the value of our AI services by sitting down with our clients after implementation and asking what's actually working, what isn't, and where the workflow is still rough. Adoption data, accuracy logs, and dashboards are useful, but the most important signal is whether the doctor and the front desk would buy it again. That feedback shapes the roadmap, and it's how trust gets built—not at signing, but in the months that follow.

Underlying all of it: Treat patient trust as a workflow concern, not a marketing one. Transparency about where AI is being used, human review on clinical decisions, clear escalation paths when the system isn't confident, and an easy path to a live person on any inbound channel. Innovation is moving faster than safety frameworks, so the practices that succeed will be the ones that pair speed with discipline. OM

References

1. Health Resources and Services Administration. Physician workforce: projections, 2020-2035. November 2022. Accessed June 12, 2026. https://bhw.hrsa.gov/sites/default/files/bureau-health-workforce/Physicians-Projections-Factsheet.pdf

2. American Optometric Association. Optometry increasingly shoulders medical eye care as ophthalmology shrinks. February 17, 2026. Accessed June 12, 2026. https://www.aoa.org/news/clinical-eye-care/public-health/optometry-increasingly-shoulders-medical-eye-care-as-ophthalmology-shrinks