"Technology in healthcare needs a purpose-built solution to solve the problems"

Q. Can you tell us about Conversa Health and your involvement with the company?

Murray: At Conversa, we are pioneering a new care delivery model – automated virtual care. It sits at the intersection of what is happening in automation and what is happening in virtualization. We think about it as a complement to current care delivery models. It complements in-person, digital, telehealth, remote patient monitoring, etc. It adds a piece of the puzzle that we think has been missing and will be standard of care in the future. I have known the founders of the company for years. We have worked together in different healthcare ventures, and this is my fourth digital health venture. I was involved in companies focused on consumers and patients, providers, and the payer market. I was super excited to join the company to help form the strategy and then take over as the CEO about 18 months ago.

Q. You are a private, VC-funded company. Can you tell us who are your major investors and how much money the company raised to date?

Murray: We’ve raised a little over 30 million dollars at C round, A round, and B round. Our investors are a great mix of supportive investors. We have financial investors, builders, VC, and Northwest Ventures, all the big health system in New York. And then we had other folks who were a combination of strategic investors like university hospitals in Cleveland, Allscripts, Pfive, and other healthcare-focused venture firms in Connecticut. We know the space as well as strategic investors with who we have nice operating strategic relationships.

PP: What is the current state of AI in healthcare?

Murray: The way we think about technology in healthcare, especially digital health, you must build a purpose-built solution to solve the problems. Then in that solution set, see if the applications of AI or machine learning or deep learning make sense to improve what you are trying to achieve. Those tend to be the most successful applications. In image recognition, in radiology a lot of good work is being done with AI. But even there, there is a need to recognize where AI needs to complement actual intelligence from people. There are a lot of studies that show this notion of co-bot. For example, the person working with AI in radiology can identify breast cancer tumors and characterize them. But on the corner cases, we need to have an actual trained professional distinguish and then make a judgment call. So, Garry Kasparov, the chess master, distinguishes between what humans are good at and what AI is good at. Humans are good at judgment, and AI is good at decision-making. Decision-making is computational all the way down, and judgment is knowing what matters and why it matters. If you can get those to work together, I think you have the best solutions. There’s a lot of conversational AI being thrown at natural language processing. If you are modeling physician language well, you can get high accuracy because physicians might use big words, but it’s a very prescribed and precise vocabulary. When you start to step into a patient world, they can say anything, and it can mean anything. You must infer if you are trying to rely on that platform for accuracy to determine whether you need to intervene with the patient, probably not the best approach. So, I think there is a lot of technology in search of solutions and the successful ones have understood the problem deeply.

Q. Conversational AI tools, especially chatbots, are having a moment considering the pandemic. What COVID has meant for your company?

Murray: When you think of conversational AI, there are many applications, but they are mainly administrative ones. For instance, I call a call center, and I am trying to make an appointment and understand my explanation of the benefits. There is some good AI because there’s minimal language, a lot of ability to get you to the right place, and its pure cost savings. It is reducing the number of customer service reps on the phone—a lot of proof points outside of healthcare are being brought in healthcare. There is a category that I would call virtual urgent care where an anonymous patient is walking to the door with symptoms. You need a big database of symptoms correlated with outcomes, and you dynamically update that so that you can decide whether I’ve got COVID or I’ve got a rhinovirus or a cold. There is a bunch of companies doing that. Doctors are primarily skeptical, so they are having success selling the payer market to some employers. The category that we live in is more care management, care coordination, transitions of care, pop health. For instance, you are enrolled in a heart failure program, the AI reaches out to you, talks to you, collects information, and checks how you are doing. It is using an evidence-based pathway to determine whether we can automate the next step, where we need to ask you what the right next step for you is, it’s very difficult to do with AI. The state of the technology is still not there, so we have taken a structured approach. Permutations are enormous, and there could be billions of permutations. Our intelligence is how do I stitch together structured conversations so that it’s personalized for you, and you’ll engage. We can collect the information we need and then use the right nudge, escalate you to the right next level of care. We use the AI/ ML in the prediction piece, it’s not just a chatbot, a chatbot is the user experience, but everything we collect from you could be biometrics, could be Piros, could be informal answers to questions, structured information. We then, in real-time assess that and check what we should do next. We use a lot of AI and ML there to predict if you’re going to decompensate because we’ve seen lung function like this with this characterization from your FEV1 scores and you’re likely not to do well in the next month.

Q. Do you think we are further along when it comes to applying conversation AI in the context of administrative use cases like the ones you describe or do think we are further along with clinical use cases?

Murray: Many things moved forward because of COVID. We’ve got about a hundred and fifty automated virtual care programs running at various large health systems around the country. I would say the clinical has caught up, and the stakes are higher. We strive for one hundred percent accuracy in determining whether a patient can be automated or isolated on the next step. You can’t do that with natural language understanding technologies. It must be a very deliberate and structured approach. But then you have the smarts to understand the status of the patient to make the right decision. Conversational AI in the context of administrative use cases was ahead. But clinical is probably a priority right now, and the opportunity for clinical is enormous. They come together in the mid to long-term future because you would want to have the administrative use cases attached to clinical all via one platform.

Q. Based on what you are seeing and the work your firm is doing, where is the low-hanging fruit today? Is that in certain types of clinical conditions, for instance, behavioral health?

Murray: We’ve conceived this to be a platform, meaning that it needs to work across all the meaningful use cases of a large health system or health plans. So, we have decided to build a platform that can accommodate programs or automated virtual care pathways from chronic care management to post-acute care to perioperative to women’s health, cancer, pediatrics, and in the ED. We have programs in those areas, and we are continually building them. Patients do not necessarily fit easily into one use case; you might have diabetes, hypertension, and suddenly you need a hip replacement. We want to accommodate it, be an extension of the health systems care virtually for patients in a seamless way that has a great user experience and leverages the full 360 views of the patient. So, that is where we are heading. We tend to start post-acute, 30 day, 90-day post-acute programs, and monitor people when they leave the hospital, and focusing on helping them recover and reducing unnecessary readmissions. We want to focus on patients who are walking out of the ED, understanding discharge instructions, picking up their prescriptions, going to their follow-up appointments, really focused on lowering recidivism back to the ED, where it’s not necessarily chronic care management.

During COVID, we worked with UCSF Health in San Francisco and shifted our focus to the vulnerable population. We helped reduce the risk of getting infected from COVID and provide a better experience for patients who could calibrate all the parameters remotely. There are many examples where we have identified decompensating patients, whereas otherwise, they would not have to escalate. And then, like ED, we are also now seeing a lot of interest in targeting both pregnancy and early pediatrics. Behavioral health is another area, it was a pandemic before the COVID pandemic. It is amplified because of that and so that’s another area that we’re getting into.

Q. Who do you mainly serve – payers, providers, employers? What is your ideal client profile?

Murray: We primarily work with a lot of midsize and community hospitals. We are also provider-focused because we want to make sure we understand how to extend a trusted relationship. And we have very high enrollment, activation, and the ability to change behaviors and drive measurable outcomes. So, the way that a patient thinks about it as a health companion. It is a twenty-four by seven extension of my doctor and nurse.

From the provider side, they think of it as an automated care team member who is helping to reach out to all these patients on their behalf and can practice at the top of their license. So, within that model, we have expanded to work with health plans. Our focus with health plans is where they are acting as a provider. We work with them to create programs used by patients and health systems and then for employers, schools, and the community and this got accelerated during COVID. We have many employers and universities using our COVID programs to screen for COVID to manage people who are positive, monitor people who have been vaccinated, and now deal with mental health from COVID. All of it is delivered through our healthcare partners. Our focus is that the health system in your community should be responsible for caring for the community. We are giving them a platform to amplify that help that they are already providing.

Q. What does conversational AI compete with in the context of a healthcare provider?

Murray: As a company that is positioned itself as an enterprise-wide platform. So, we want to be your automated virtual care partner if you are a health system. If you are using automation for administrative purposes, that is complementary to what we do. If you are using it for a digital front door, virtual urgent care, it’s complementary to what we do. So, you are now managing your patients; you’re enrolling heart failure and diabetes patients into programs. Our platform will compete with point solutions. If someone says I have an app that can help manage diabetics, I can come in with an entire stack device and coaches. So, somebody might want to choose that to work with their diabetic population. We say, hey, you can use the same platform to treat your diabetes and cancer patients. That is pretty compelling because health systems increasingly want to consolidate. Working with one partner is easier, and you start to understand patient IDs across the continuum. We aspire to manage patients across a lifetime, which does put us in competition with point solutions in certain areas in the future. Our challenge is to figure out if there is a perfect point solution. How do we integrate it into our platform, and how do we start to allow other solutions to plug into our platform?

Q. Does your tool sit on top of an Omada or a Livongo kind of platform? You also mentioned about clients wanting to consolidate into a one-stop-shop. We see this in our work with health systems, where they are trying to reduce the footprint of vendors, they must deal with because of all the complexities involved. How do you fit in that context, and how do you help your clients work through the tradeoffs involved here?

Murray: Companies like Livongo have chronic care management for diabetes, hypertension, weight loss, and behavioral health. They have devices and coaches wherever applicable. Because it is a service-based company, it tends to be with payers and employers very successful. Most companies with full-stack solutions are doing it because payers and employers do not have clinical resources and devices. Health systems already have clinical resources caring for patients; we want them to be more efficient and effective. It is purely a software platform where devices are involved. If they have an RPM partner, we are complimentary. In the health system world, they do not need the provider networks of any of those other companies. They are looking for technology solutions, and we excel there. When we go into the world of payer and employer outside of the health systems, those companies become partners. So, like Livongo, we can say that if you want to add the conversational AI and decision-making we bring, we can help leverage the platforms that they have built in the same way we do for the system. So, if you aggregate what they do, they have clinical resources like the health system does. They have the device as the RPM’s do. We can bring the piece to the puzzle that we get that table in that world.

Q. Where does the voice fit in all of this? Do you compete with voice-based solution providers like Nuance for instance?

Murray: We do not offer voice today. However, I demo our platform using voice a lot, but I am just using the native voice on the phone to do text or voice. We have not gone there yet because we are very driven by where the market need is, and the impact that we are having is enormous. When we see that people interact with voice, we realize it is not a big thing to add. You want to make sure that you are designing the voice interface as per the requirement. You are not translating a text to voice because understanding what someone is saying with 100 per cent accuracy in voice is a different design requirement than doing it through a chatbot.

People like Nuance probably have the best-known value out there. But it is not an accident that they’ve chosen to do transcription for wires because when you’re looking at what a provider says and being able to transcribe accurately, you can do that. You get into the patient world where a patient can say or respond to anything in that world. The way you would measure it is precision and recall precision, where the recall rates will be 80 per cent at best, which means the error rates are 20 per cent plus. No hospital system will use that error rate to decide whether they can automate the next step for a patient.

Q. When you talk to your clients, what do you ask them to prepare for in terms of the most challenging aspects of rolling out a conversation AI tool?

Murray: We have a very rigorous process that has four different swim lanes. There’s integration, configuring the pathway or the program. Everybody delivers their care delivery model for diabetes to slightly different guidelines. So, there is an integration pathway, there are best practices on how you enroll the patients, and then there is how we’re going to measure success.

Some of our clients are very sophisticated and want to collaborate. We have taken off the table things like NLU and liability. So, the real focus is on making a program that we have designed and figuring out how to deliver care that fits in your model, works in your workflow, and integrates into your data flows.

Q. What do you see the next 12 to 18 months looking like from the point of view of VC money flooding the market? Also, what is your advice to a digital startup that is on the receiving end of this money?

Murray: Markets, in my experience, always overshoot. They are trying to get an equilibrium, but there is a massive bait on both sides. There are unprecedented amounts of money in digital health, and it is concentrated in certain areas like behavioral health, which is a big problem. Every time that happens, we are starting to see a massive consolidation. To your point, companies go public and use that capital very quickly to acquire. Teladoc, Livongo kicked off a big part of that, and there are others like Grand Rounds and Doctor On Demand. There are tens of hundreds of these deals happening, and many companies are getting funded. I think what we will see is continued consolidation, and there will be a whole bunch of companies that don’t make it above the threshold to be viable or to be attractive to be purchased, and they’ll go out of business, or will be acquired. It will happen quickly, and that’ll make the companies that are above there much stronger.

So, advice to somebody coming in the space is it is always better to start in a cycle where things look horrible because that is how you develop your product. If you are starting a company now and get funding, spend this time developing your product and getting product-market fit. Pick a problem because I think there is a lot of technologies out there in search of solutions. The market will give you an opportunity if you can solve it better than someone else or if it is an unsolved problem. Once the product is available in 18 to 24 months, that is probably an excellent timeframe to come out with a product. I am in a position right now where I am not worrying about generating revenue but worried about just building the product, and I have the funds to do it.

You must be doing lots of things, but it all comes down to the patient if they feel it is important in their care and have better outcomes. The providers can care for more patients and spend the time doing what humans can do; then, you have a winner. Those are the only two things I look at to see if we are successful. I look at what patients are saying and doing with the products, and I look at whether providers embrace it. If we have those two things in place, everything else will go well and ultimately; it will be successful.