Here is a little background on Yu before joining Innovaccer: She trained as a primary care internist at the Brigham & Women’s Hospital, and her first job was as chief innovation engineer for Boston-based Atrius Health, which is now part of Optum. During her residency, she founded Twiage (acquired by TigerConnect), an early response platform for first responders and emergency departments. After Twiage, she led Verata Health, an AI prior authorization startup, as chief medical officer. That company was acquired by Olive AI. After the acquisition, Yu served as general manager and chief medical officer at Olive. She was later named chief product officer and spearheaded acquisitions by Availity, Waystar and Humata Health. Most recently, she helped found and launch Care Lumen, which is focused on the payer technology space.

Healthcare Innovation: Our publication typically focuses on the health IT and analytics deployments that impact population health and value-based care. But this Innovaccer report focuses on an area that's more directly focused on patient access, revenue leakage, and referral conversion. Is that typically the purview of the CFOs or chief patient experience officers, rather than the CIOs or CMIOs we typically write for?

Yu: I think it's probably something that every C-suite executive at a health system is battling, because it's a CEO-level imperative. CIOs are being pressured to describe how they are enabling digital front doors. How can we get these patients in? The same thing with CMIOs. 

When I work on the provider side, the constant struggle is that our capacity from a clinical standpoint feels incredibly limited, and our schedules are always booked out. Dermatology takes three months to get to see a doctor. You want an MRI — well, that's going to be another two to three weeks to get in. Yet if you look at actual utilization — the number of clinical visits completed — most organizations would probably come in around anywhere from 85 to 95% of utilized capacity. There’s a gap there. If patients are not going to be seen at your facility, they're going to go somewhere else, and that's the revenue leakage. So, at the end of the day, a CMIO or a CIO is still trying to think through how to solve this. The health system can’t just hire a bunch of clinicians and expand capacity…What am I going to do in terms of my technology investments to actually solve for this?

Sometimes it’s viewed in a piecemeal way, looking at hold times and scheduling capacity, or prior authorizations. The truth is that you have to solve for the full stack. You can have the shortest wait times ever, but if you don't solve for your other capacity issues, you'll have leakage anyway.

HCI: The report mentions five systemic failure modes of referral leakage, but I think it also mentions that these can cascade on each other, and that they're connected, so it recommends taking a holistic approach...

Yu: Exactly. I'll give you one example that I see. You have a patient who has new back pain. A PCP in your network evaluates them and they want this patient to get an MRI. That MRI needs to undergo a prior authorization. Many health systems will say, OK, thank you for this referral; let me do all the insurance checks and try to get this authorization, and I'm going to call that patient back, and we'll schedule when we're available. In that situation, the patient is told at checkout that somebody will call you. A few weeks goes by, and they are anxious. They might call back. They are told, "Well, I don't know what the status is. Let me check up on it.” It is an incredibly frustrating experience. With some of my patients, their voice mailbox is full, so my clinic tries to call them once, and it goes to voice mailbox full. Then I get the notice that says we've tried to call them twice. Now we’re sending a letter to say we were unable to schedule you for your MRI. Does that MRI ever happen? Does the patient go to an ED instead? So that's what a lot of health systems do today. 

Another approach is to schedule that patient, but schedule them far out, so that gives the health system enough time to get that prior auth done. Maybe you have availability later this week, but instead you schedule it several weeks out, so that your teams have time to get the prior auth. Now the best systems — not only do they solve their prior auth through AI automation, but they schedule that patient at the appropriate time —  let's say in two or three days. They know that there's perhaps a 5% chance that the prior auth does not come through, but they're willing to reschedule you, rather than that patient leaving the office without a scheduled appointment in hand. That is best practice. Actually, it's really difficult to do, and most organizations don't have the infrastructure and the technology system set up to do that, so that's why they default to the other two options. When you give a patient an appointment in hand before they leave their doctor's office, that patient is so much happier and so much more likely to actually get that MRI. They’ll show up. Organizations that do that have a much higher referral conversion, and much higher revenue.

HCI: The report offers a few case studies of health systems launching AI-powered patient access platforms that had this positive impact, but what does that look like from the perspective of the staff in a patient access center or from the patient perspective? Some of the AI features mentioned are scheduling optimization or virtual callback. Could you talk about that a little bit, and how that impacts the staff trying to do that work?

Yu: What happens is that most of these organizations feel like they don't have enough staff and they can't find enough hires to scale these centers. In healthcare, we need every one of those employees to be focusing in on the complicated cases, the ones where you really do need to sort through a bunch of doctor schedules to get a complicated patient in. We’re using AI for applications where it is an easy solve and there's no wait time. The call is immediately answered after they select the right department. We ask them and triage the questions. If they need to schedule a physical, and it's not an urgent physical, then it will automatically run through it for them. In some cases, it's just duplicating what you might do in a patient portal by yourself, but just over the telephone. A lot of these phone calls are about questions like: What are your hours? Where can I find parking?  In knowledge spaces like that, an AI agent can absolutely help answer those. 

In the call center, I think folks generally feel like it helps them and relieves a lot of stress from their staff, but it becomes even more powerful when you connect the entire enterprise. Again, one of the things that we focus on at Innovaccer is how do we solve for all of it instead of just one piece. If we just did the call center agents and don’t solve for prior auth, it doesn’t solve for your availability, so part of what we're doing at Innovaccer is linking those next steps, Not only are we able to schedule, but we can help actually relieve those downstream tasks from your team. Now we're solving a full problem end to end.

HCI: Does the size and complexity of the organization impact how quickly they might implement these changes? I think the report has three case studies, and in one of them the health system focused on process improvements first, and then they were going to use some savings they saw from that to make the AI investment. Do people have to proceed at their own pace with how quickly they make this transformation?

Yu: I think so. You have to find the right place to start for your organization, and it might be that your organization, whether it's because of the EHR or the other technology, you aren't ready for full enterprise transformation, so you want to do a couple of processes first, and then go further. 

We have some tables where we have benchmarking data, and you can try to find yourself in that, in the same way you should find yourself in tech readiness or AI readiness. Some organizations that are on the leading edge of this have invested in AI platforms and AI orchestration layers. They know that AI startups and agents are proliferating. I think the last data I saw was that there was nearly double the number of AI vendors from 2025 to 2026 and it's expected to be like 20x that by 2030. It’s just going to be crazy, so the most forward-thinking organizations are already thinking about AI governance and AI orchestration layers. But we understand that you have to go at the pace that you can imagine. What we hope is that when you see the proof of concept and how quickly we can deploy these solutions compared to even five years ago with RPA and bots — they were so brittle, they were slow to deploy. They were the best that we could do at the time, but now with LLMs for everything, AI is leaps and bounds better. So I think you can start by implementing in a couple of areas, and then show the art of the possible.

HCI: Is there anything else from the report that you'd want to stress?

Yu: What I loved about the report is just the fact that health systems are measuring these impacts now. I think that shows a maturity in their organizations of measuring how much of this drives revenue and not just focusing on bottom-line costs or staff savings.

One fear is that we see some of the top-quartile health systems really stretching ahead, so our hope is that we don’t leave a bunch of health systems behind on this. I think every organization should be thinking about how not to be left behind, because the really well-funded organizations that have been investing in technology and have been thinking about this, they're actually leaps and bounds ahead. I worry that without adequate investment or help in some of these areas, it's actually just going to widen the chasm. So, I would just say for every CIO and CMIO out there, the time is now to really think about investments in AI, and who you want to partner with on that journey.