‘LLM-Native’ Startup Parakeet Bringing AI to Physician Practice Call Centers
After securing $3 million in seed funding led by Canvas Ventures, startup Parakeet Health has launched a generative AI voice platform focused on patient engagement. Two of the San Francisco-based company’s executives recently spoke with Healthcare Innovation about what's involved in bringing generative AI to physician practice call centers.
CEO Jung Park previously held executive positions at One Medical and Epocrates, and co-founder Eric Mao led product and engineering teams at Microsoft and Primer AI.
Healthcare Innovation: Why did you decide to focus your generative AI efforts on the patient contact aspect of healthcare?
Park: People may think focusing on using AI to solve administrative challenges is not that sexy, right? But when you think about it, deploying AI on the clinical side has challenges of its own. One of the things that I learned when I worked at Epocrates was don't change the workflow. So if you actually try to do something that takes over what a physician does, you may not necessarily get the adoption that you expect.
HCI: What are some of the challenges physician groups face with managing patient contacts and calls?
Park: It's something that I think all of us can relate to. It's not uncommon to wait five minutes before a call is answered. No one wants to be put on hold. Everyone's time is valuable, and if you don't take care of these issues, patients feel like they're not valued. Also, for many medical practices, staffing is the biggest challenge. There's turnover. Labor costs keep rising, and for the people who are actually addressing a lot of the support issues, the concept of empathy fatigue is real. It's only so often that you can actually have a smile on your face, and do this constantly, day in, day out. At some point people burn out. That's an issue that we all face as healthcare administrators.
Another challenge, I would say, is revenue leakage. Running a medical organization, you have lots of opportunities to generate revenue, but also there are a lot of opportunities to lose revenue. Let me give you an example: When you have a late cancelation, that means the patient who is supposed to come in and take that provider's time slot doesn't show up. So it creates these gaps in the schedule, which results in lower revenue.
When a specialty practice receives a referral from a primary care practice, you need to convert this referral, because they're like a golden ticket, right? If a doctor says to you, I'd like for you to go see this dermatologist about this rash, then you go. But it turns out people don't always go. So when you're actually able to convert all these referrals, that's really the way that you're going to generate revenue.
HCI: So how does Parakeet work to address these challenges?
Park: First of all, it improves the patient experience by answering every call 24/7, with zero wait time. Parakeet also automates repetitive tasks like scheduling and answering frequently asked questions, so that the staff, instead of actually doing these repetitive tasks, can focus on r more valuable things that patients need, like checking in, checking out and really attending to patients' need when they're actually in the physical office.
From a revenue leakage perspective, Parakeet actually makes outbound calls. So if someone cancels, Parakeet can actually call the next person who's in the wait list, or maybe there's a patient two weeks out that might be available to come in and fill in that gap in that schedule. That's a way to actually bolster the provider utilization and make sure the revenue is captured for the healthcare organization.
HCI: Over the years, most people probably have experience with a chatbot on a retail or airline website, and they often get frustrated with it because it's not able to answer their question. But has the technology improved exponentially now?
Mao: I've been in AI for a very long time. I don't think it's ready to replace doctors, but we see firsthand that it now can take on these challenging scheduling questions, and I think it's ready for prime time for this set of use cases. The transcription quality now is so much better now with this wave of large language models. And the speech sounds like a human, which is a very recent thing. It's in the last two years that we've had this.
HCI: Does your system have to integrate with the practice's EHR, billing system, customer relationship management system, in order for this to work?
Mao: We spent the past year building out all this infrastructure to do that. We integrate with your EHR, with your telephony system, with your CRM ticketing system — all those things to enable that very natural experience. So when you ask for an appointment, it looks it up in real time, it offers you real-time availabilities, and then we book it directly into the system so there's no human interaction needed throughout that whole process.
HCI: Is that process challenging or do all these systems now have APIs that make it go pretty smoothly?
Mao: You would hope so. But what we've realized is a lot of these EHRs do not have very good APIs, probably on purpose sometimes, because they want you to do everything on their platform. But we spent a lot of engineering effort to make this reusable. We’ve made it easy to onboard the next customer, because we've done all the heavy legwork already.
HCI: It looks like a few of the customers that you have announced are virtual themselves. Have you targeted those kinds of businesses initially?
Mao: We’re targeting medical practices that typically have a call center. So we're not targeting mom and pops. When you have a call center with 10 or more people, you're going to see more cost savings and more use out of what our platform does. We have dermatology; we're in talks with OB/GYN and primary care.
HCI: So would it make sense for a large health system like Sutter Health or Kaiser Permanente?
Mao: We're going to target the Sutters later on. We’re targeting outpatient first, but that's absolutely in our roadmap tackle the Sutters, the UCSFs, the university health systems.
Park: We're already seeing that our solution has hit a nerve. Whether it's those folks who are actually working more on the traditional side or the digital companies, at the end of the day, the opportunity to interact with patients is a real thing. Ninety percent of all of us still use the phone to make an appointment, so that means that there's an opportunity across the board to really engage with patients.
HCI: Are traditional vendors in the patient engagement space trying to bring AI solutions to market as well?
Mao: You have incumbents in the patient engagement space, but they're using old technology. They're trying to steer this giant cruise ship to use LLMs, while we are LLM-native. We started on day one with this technology, so our technology works better. Eventually, the incumbents are going to try to adopt this new technology, but we think we can move faster, and our technology is already here using those things.
Also the competitor really is offshore call centers. When we talk to customers, when they have challenges of too many missed calls, they're looking at expanding more people via offshoring to solve that challenge. They're not even looking at technology. So when we show them what we have, they're like, ‘Wow, I didn't even know this was possible today.’
