CirrusMD is a physician-first virtual care company that connects over 13 million members with physicians in under 60 seconds, bridging the gap between patients, physicians, and plan sponsors. CirrusMD recognized several critical challenges in healthcare — plan sponsors are keen to ensure that patients maximize their benefits, physicians face time constraints and information gaps that hinder thorough treatment, and patients may suffer from lack of access to quality care.

CirrusMD sought a generative AI solution that could scale the quality of care and offer the best experience for members and physicians. Despite investing a year in developing in-house using OpenAI models, they faced challenges in quality and consistency in outputs. They switched to the Writer full-stack platform and top-ranked medical model, Palmyra-Med, which enabled them to swiftly implement new AI-powered workflows for benefits navigation and clinical documentation in less than six months.

Tell us about yourself and CirrusMD.

My name is Tanya Dillard. I am the SVP of Product Management and Innovation at CirrusMD, where I lead the product strategy, development, and design functions for the company.

CirrusMD is a physician-first virtual care company that provides healthcare to over 13 million members. We streamline the patient journey across clinical settings by offering not just acute healthcare but also longitudinal care, such as behavioral health, care for chronic conditions, and primary care.

CirrusMD’s high-quality chat-based care delivery model connects patients with a physician in less than 60 seconds.

What prompted you to consider using generative AI to support your clinical team?

CirrusMD sits at the intersection of patients, physicians, and plan sponsors, including employers, commercial and government-sponsored plans. Each of these three parties faces unique challenges:

• Plan sponsors want to ensure patients are getting the maximum benefit from their coverage.
• Physicians often lack time and information to diagnose and treat as thoroughly as they would like.
• Patients can suffer from lack of access due to cost, distance, and time constraints.

At CirrusMD, we want to deliver a great experience for patients and physicians, while maximizing health outcomes, so plan providers deliver the best, most effective, and most efficient care journey for their members. And we really believe that generative AI has the potential to make a tremendous impact in these areas.

Let me give you an example. We know that patients are more likely to act on benefits recommendations when they come directly from their physician. But it’s incredibly time-consuming and cumbersome to identify the appropriate benefits for each patient, tailored to their unique health situation, benefits plan, age, and location. This is exactly where generative AI can step in and make a difference. We used the technology developed by Writer to offer patient-specific benefits recommendations to doctors in real-time at the point of care. This is just one example of how Writer’s generative AI solution helps us improve quality of care. And our doctors love it because they’re delivering tremendous value to patients.

What has been your experience building with generative AI?

It took us about a year to figure out that we’d be much wiser to work with Writer than to try to do it ourselves. We started by building our own in-house AI model, which showed promising early results. We knew we were heading in the right direction and wanted to scale up our machine learning efforts.

That’s when we started experimenting with OpenAI GPTs, first with GPT-3, then GPT3.5 Turbo, and then GPT-4. Each model update was unpredictable and required rebuilding, making consistency a challenge. It just didn’t work for us. Plus, GPTs are trained on all of the internet, and the outputs didn’t quite meet our standards for medical use cases. As we looked ahead to workflows that were more clinical in nature, we needed something with a deep understanding of medical language and context. We were not likely to get that working with a GPT.

On top of that, we had just one ML engineer and were not resourced to build an in-house AI stack. The variance in model performance, combined with our limited staffing, led to months of delays in launching our first AI-powered use case. We were learning as we went, but AI isn’t our core business. I found myself wondering, “Why are we trying to do this? We are not an AI company. We don’t sell AI products. Why don’t we find an AI partner who could help us move faster and give us a competitive edge in what we do best?”

For us, that partner was Writer. Within just six months of working with them, we launched our first use case and completed the technical build for a second. This is a huge leap forward compared to the year we spent trying to build an in-house solution.

What makes Writer the right partner for CirrusMD?

Given the medical nature of our work, we need generative AI that truly understands medical context. Writer’s Palmyra-Med model is trained specifically for medical use cases, which means it delivers results we can rely on. In addition, the Writer full-stack platform includes all the technical components we need to build and deploy custom solutions quickly. This means we no longer try to stitch together AI tools or rely on our internal team to develop them from scratch, which saves us an incredible amount of time and resources. Lastly, Writer’s exceptional services team has been able to meet our tight development deadlines and high quality standards for outputs. With Writer, we’ve found a trusted partner who can help us evolve more quickly across more care categories and personalize in a way we had never been able to before.

Tell us about a bit more about your benefits navigation use case.

When choosing our initial use case, we opted for benefits navigation because it was technically challenging but less clinical in nature. This allowed us to build confidence before tackling more medically complex and higher-risk scenarios.

So, what is benefits navigation? Essentially, plan sponsors want to ensure their members maximize their healthcare benefits, and our physicians want to provide holistic care. At the end of a chat, a provider would love to say, “Hey, did you know you have access to digital physical therapy? I think that could really help with your back pain.” But manually matching benefits to each patient’s health situation and coverage is incredibly tedious and time-consuming.

With Writer, we built and deployed our AI-powered benefits navigation workflow in under four months. It automatically reviews chat history and provides personalized health benefit recommendations covered by the patient’s plan. We integrated this directly into the CirrusMD chat experience via API, making it easily accessible for our physicians.

The results have been outstanding. Automating this workflow led to a 234% increase in physicians sharing benefits recommendations with patients. Our analysis shows that 30% of patients engage with these recommendations when provided by a physician, compared to a mere 2-5% baseline engagement rate from methods like robocalls from health plans. With hundreds of thousands of patient interactions per year, the impact of this single workflow is immense.

As Dr. Sara Herstad, our VP of Clinical Operations and Primary Care, puts it, “Physicians want to provide holistic care but time constraints often make it challenging. Writer enables our providers to scale quality care while reducing administrative workload.”

And tell us about some of the other use cases you’re working on.

Another use case we’re excited about is our clinical documentations workflow. After each patient chat, physicians must complete SOAP notes—that’s Subjective, Objective, Assessment, and Plan—in a specific format so the entire clinical team knows the patient’s status and treatment plan. Our providers are extremely busy and may not get to their notes until the end of the day, which can affect the quality of the documentation. Clinical documentation consumes multiple hours each week for each physician, adding a significant administrative burden.

With Writer, we built an AI-powered workflow with Palmyra-Med to automatically generate high-quality clinical summaries in medical language. At the end of each patient encounter, it automatically reviews the chat history and drafts the SOAP notes for physicians to review and approve. This not only improves the quality of the summaries but also reduces the huge burden on our physicians, allowing them to focus on more critical tasks. Thanks to Writer’s support, we completed this workflow in under two months.

And we’re just getting started. Up next on our roadmap is using generative AI for quality assurance by conducting chart reviews to ensure care quality and adherence to best-practice guidelines. Following that, we will tackle clinical best practices and use generative AI to help with clinical decision support at the point of care — all with physician oversight, of course. We’re building from the outside in by starting with lower-risk use cases as we get started with generative AI and getting progressively more clinically sophisticated as we learn.

What’s been the overall impact of Writer on CirrusMD?

Writer is truly revolutionizing how we scale care. For patients, we can now offer more holistic, high-quality care in real-time, help them make the most of the benefits their plan sponsors provide, and achieve even better health outcomes than they would without us. For physicians, we’re boosting satisfaction. One of the things that we know is admin work kills the joy of practicing medicine. With Writer, we can lessen this burden, increase productivity, and give doctors more time to do what they love—practice medicine.

At CirrusMD, our goal is to deliver immediate, affordable, high-quality virtual care at scale, connecting members with doctors in under a minute. Writer aligns perfectly with this strategy, enabling scale while maintaining personalized care for our members and creating a more streamlined, enjoyable experience for our physicians. As CirrusMD continues to grow rapidly, Writer supports our ability to offer our services and maintain affordability without compromising care.

What is your advice for other healthcare leaders considering generative AI?

While generative AI can transform many workflows, it’s wise to start with a few specific use cases where you can measure the tangible benefits for patients, providers, and the system. This helps you focus and get started quickly. And I recommend a progressive approach: begin with non-clinical use cases like benefits navigation to experiment and learn before moving on to higher-risk scenarios.

My second piece of advice is do not try to build your own AI tool internally — it’s often slow, challenging, and resource-intensive. Generative AI is a specialized field, and the best strategy is to partner with the right expert. Make sure your AI vendor can deliver clinically-reliable outputs in medical language and is flexible enough to meet the unique needs of your workflows. By partnering with Writer, we’ve significantly accelerated our progress with generative AI, and I’m excited to see what more we can achieve together.