In the late 1990s, two graduate students in Stanford’s computer science department set out to organize the world’s information. Shortly thereafter, a visiting scholar named Mario Schlosser arrived on campus, set on figuring out how trust could be built into peer-to-peer networks. The original server used by the graduate students, who were now running a little outfit named Google, had formerly been crammed under a desk in the office Schlosser now used.
Now, 20 years later, the graduate students have done OK, and they no longer need to borrow server space. But they’ve continually been stifled by one kind of information that’s very hard to organize: health care data. The industry in America is a mess: yoked together with confusing regulations, perverse incentives, and computers running Windows XP. Meanwhile, Schlosser has moved on from academia and created a company, called Oscar, with Joshua Kushner (brother of Jared) to try to solve those problems. The goal of Oscar is to do to health care what Uber did to the taxi industry: use smart digital technology to make everything faster and easier for customers, and then use the data gathered to build radically new services, which can collect more data that leads to new services. (Ideally, Oscar would like to accomplish this without cracking as many eggs on its own head as Uber did.)
Alphabet, Google’s parent company, invested early in Oscar through its venture capital fund Capital G and its health services spinoff, Verily. But today they’re announcing a much larger, and more strategic, investment of roughly $375 million. Neither company will give exact figures, but it seems that Alphabet will now own roughly 10 percent of Oscar. One of Google’s earliest employees, Salar Kamangar, former CEO of YouTube, will also join Oscar’s board. I spoke with Schlosser for an hour on Monday about the deal, privacy, data, and whether, one day, we’ll actually treat our gastroenteritis through an app.
Nicholas Thompson: Good morning, Mario. It’s a pleasure to talk to you. Thanks for speaking with WIRED.
Mario Schlosser: Good morning, Nick. It’s a great pleasure talking to you.
NT: Thank you. So you just raised a lot of money. You’ve already raised a lot from Capital G and from Verily, so this sounds more like a strategic partnership. What do you get out of it?
MS: We’re humbled that Alphabet’s coming in. We raised our first round of funding from Capital G three years ago and since then they’ve been following us closely, seeing what we’ve been building, and Alphabet has come to the conclusion that they want to put more behind the company. It’s fantastic for us because it will really allow us to focus fully on the core model we’ve been building for the past six years, which is: use technology, use data, use design, use a human approach to build a very different health care experience. And that’s what this allows us to do.
NT: And why are you doing it now? You just raised a lot of money at a pretty similar valuation in March.
MS: We weren’t out there trying to raise additional money. We raised a round a couple of months ago. But Alphabet has just been talking to us for the past three years, and it took them awhile to get to the point where they really said, “This is something we believe in and want to put more money behind.”
NT: So it’s kind of like they’ve been chatting with you, you guys have been going out to dinner, and now they’re asking you maybe to move in with them? Is that the status of the relationship?
MS: Well, they’re a financial investor, and they’re still a minority shareholder, so it’s a good trajectory for an ongoing relationship.
NT: Alright, so you’re dating but you’re not moving in together yet. Why are you interested in a deep strategic relationship with Alphabet? It obviously has financial benefits but also has certain costs.
MS: So, again, first of all it’s a financial transaction, right? They’re making an investment and, of course, Salar [Kamangar] is joining the board. It’s not that we have anything else to announce in terms of deeper ties at this point. Now, I do think we’ve benefited a tremendous amount over the past few years from being in touch with them in a good way. From my point of view, they’ve always been among the technology companies I’ve admired the most. They are the smartest, in my view, when it comes to data; they’re the smartest when it comes to the building of pure technology. I’m thrilled about that as a vote of confidence that I think is hard to parallel in today’s world.
NT: What are you going to do with all the money you’re getting?
MS: First and foremost, really make sure we invest and continue to invest in our differentiation. We were the first greenfield health insurance company in New York State in about 25 years or so. Trying to change how the health care system in the US operates by starting an insurance company wasn’t an idea that really anybody else had at that point. The other thing we did over the past six years is really very methodically rebuild the entire infrastructure we have from the ground up. We have our own claims system we’ve been building, we have our own clinical management system, we built our networks ourselves—pretty much everything that we do internally to manage people’s health care was reinvented and rebuilt from a technology perspective. And so that’s what we can now go faster toward. We can hire more engineers, we can hire more data scientists, more product designers, more smart clinicians who can think about health care a different way. It’s the acceleration of that product roadmap that fascinates us the most. The second, more tangible piece, is that we’re launching new product lines. And the most notable one will be Medicare Advantage for 2020. So we’re coming into an additional business segment. We’re in the individual market, we’re in the small employer market, and we’re going into Medicare Advantage in 2020.
NT: And so how big a part of your business do you expect that to be? Right now, you have three parts: You have individual people who get insurance, you have companies, and now you’re going to Medicare? Is that correct?
MS: Yes, exactly right, yes.
NT: And what will be the revenue split from those three categories?
MS: You know, that’s a good question. I would expect them to be equally successful. What all these markets fit into is a high-level view that US health care will individualize over the next however many years. The biggest reason why the fax machine is still more valuable than the smartphone in health care is because you as an individual generally aren’t buying your health care and your health insurance by yourself. There are all kinds of middlemen in that process. And what the middlemen do for the most part is remove the competition in the health care value chain that would go towards building something that’s a compelling, seamless user experience. They remove the cost-containment pressures as well. That’s one big reason why health costs in the US have risen so much. In the end, you as an individual can’t really vote with your own feet and oftentimes don’t even realize how much health care costs behind the scenes.
NT: One of the most interesting decisions that you’ve made is to use what you call a narrow network. So every health insurance company comes out there, says we want better care, we want lower costs, and all the other ones also say we want a bigger network. You actually are saying we want a smaller network. How do you use data to prune doctors from your network, and how did you make the decision to try to entice people with a smaller network of doctors?
MS: First of all, the high-level motivation for this comes from the fact that we’re spending a fifth of our GDP on health care and every other rich country is at half that. And so the system we have is already way too costly. The other interesting phenomenon is that cost in US health care has never really been proven or shown in any way to be correlated to quality of outcomes. If you go into the market of, let’s say, cars or phones, you expect there to be a correlation between how much you pay—you pay more money, you get a faster car with higher gas mileage. That isn’t the case in health care. You have an equal chance of going to a high-cost doctor or a low-cost doctor and getting the same outcome, the same satisfaction, the same sort of readmission rate.
NT: Well, is that true because the system’s screwed up? Or because we don’t know how to measure quality? If we had better measurements of quality do you think the data would suggest that it correlates more with price?
MS: So, you should ask yourself why we don’t have better measurements of quality. And in my view that is a secondary function of the root cause of having a screwed-up system. In a system where every network is broad, where insurance largely gets sold through big employers, where there’s a big incentive to keep everybody in the network, where there isn’t the kind of competitive pressure on the value chain, there isn’t that much of a reason to develop a better correlation between cost and outcomes, or even visibility into how certain parts of the system perform. Who is going to measure it at that level if most providers get paid on a fee-for-service basis for anything that they do? It’s not in anybody’s immediate business model that they actually start paying attention to this.
I’m not trying to be cynical about this, but to me, that’s the reason why you haven’t seen bigger investments in data infrastructures in health care that would even allow you to start deriving these kinds of metrics. When we go into new cities, we find health systems that want to build a differentiated experience with us in a different network design. The way we get data about the doctors working with those systems is oftentimes literally Excel files from these systems, and they come on a quarterly basis, and they’re 30 to 40 percent wrong. Why is that? Because the biggest purpose for these data files is for claims to get paid. This data wasn’t built for driving better clinical outcomes, attaching better clinical payloads to the data transmissions, for having more real-time insights and things like that. It’s a payments system first and foremost, and that’s it.
The only way to unlock that and to actually build a product that can get lower unit costs by having fewer doctors and hospitals at the same or even better quality with better user satisfaction—that would be the model in a nutshell—is if you have user engagement. If members of Oscar realize from the beginning that if they need a doctor, if they need help, they can come to Oscar first—they can go and search in the app, they can talk to their Concierge team, they can use us to make appointments, and we will take care of all of that. We will have the right data flows, the right tools, the right metrics, to make sure we hold the doctors accountable and make sure that whatever happens to you, whether it’s a small thing or a really big thing in your life, you’ll get the best possible care at the right point in time.
NT: Before we dig into that though, back to the narrow network for a second. If data is so screwed up in health care, because good data would change and perhaps undermine all of health care, how can you be completely confident in your decisions about what doctors to let into your network, who to partner with, and how to measure patient quality of care? It’s like a chicken or egg problem.
MS: Because as the insurance company, we already have a lot more data than anybody else in the system. That’s the key reason to me why I felt we had to be an insurance company from the very beginning. If you try to go to employers and get their data, if you try to go to health systems to get their data, you’ll have a limited view. We see in real time what’s happening, and that’s a huge difference.
The other interesting thing here is that you have to really rebuild that data infrastructure. Most partners and vendors in health care—the big drug vendors, the big imaging vendors, and so on—are not set up for anything close to real-time data transmission. And that’s been one of the most interesting conversations for us from the beginning. We sort of say hey, we want to have this in real time, and then we get looked at as if we have three heads, because it’s not clear to people the importance for anybody else in the system to have more real-time visibility. But we’ve consistently seen that seeing what is happening with somebody in real-time has one of the biggest possible impacts on whether that person would be willing to take advice and think differently about their care and so on.
NT: So that brings us to the Concierge system. You’ve built this sort of incredible system where you have, as I understand it, nurses who talk to patient groups and you have far more communication than any other health insurance company. Was that set up because you wanted to transform how health care works, or was it also set up because it makes for really great data collection which then leads to your ability to improve all of your systems and all of your care?
MS: So it’s really about the fact that we would like for members to see Oscar as the entry point to health care. The only way to build trust with an Oscar member is to be in any conversation from the very beginning, and that’s what the Concierge team does. It’s a six-person team. One of the six people is a nurse, they’ve got several people sitting behind them that can do things like, discover the right physician for the right case, escalate issues, and things like that. And whenever you talk to us, you’ll talk to one of these six people. That builds an amount of trust that you otherwise wouldn’t get. There are very simple, tactical things like, when we show you the pictures of those six people in the app, you have a 25 percent increase in probability of sending that team a message when you see the pictures versus when you don’t see the pictures.
Once we build that trust, and when it goes the other direction, when we see something in your data that prompts the team to reach out to you, you will have a much higher chance of picking up the phone, responding to the chat, and so on. To give you some data behind this, in a given week, 25 percent of members will be in some shape or form engaged with Oscar—they’re using the app, they’re in the website, they’re talking to the Concierge team. If you only look at those members who end up going to an urgent care clinic or an emergency room, which is sort of like a high acute utilization of health care that often isn’t very efficient, then 80 percent of those will have used the product in some shape or form in the week leading up to the event. And then oftentimes we can reach out to make sure you can go see a doctor, in the same afternoon, without having to go to the emergency room. We can give you a different recommendation of where to go, we can help you directly with our own telemedicine physicians. We can connect the dots in a different kind of way.
NT: Before we talk about the telemedicine thing, I just want to make sure I understand that statistic correctly. Eighty percent of Oscar customers who go to the ER have talked to the Concierge team or been in touch with Oscar at some point in the previous week? Don’t you want that number to go down? Wouldn’t you want the number of people who chat with your app to be less likely to go to the ER?
MS: Yes, that’s exactly right. And that’s what we have to do. People just need to go to the right place of care.
NT: The ideal scenario would be to model out what are the interactions they have with you, what are the odds they go to the ER, what are the odds that ER trip was necessary. Once you have that data, then to be able to interact with them differently in certain situations so that some of them go to the ER sooner and some of them don’t go to the ER.
MS: Yes, that’s exactly right. And to give you an example of how that works in our systems, we have a clinical segmentation model that categorizes members into different levels of complexity of their ongoing health care that is driven by all kinds of real-time data points—the moment a new drug comes in, the moment a new lab test comes in, this gets updated in real-time, so we have a very nice real-time view into the complexity of somebody’s health care history. When a nurse calls you back or the Concierge team comes back to you, we visualize your health care history through our own tool, Grouper, that’s sort of like a beautiful, easily accessible tool or way for a clinician to understand your health care history. So we can tell in real-time what kind of care you’ve had in the past and then based on that make sure you talk to the right person inside of Oscar or on the telemedicine team to then get you to the right provider afterwards.
NT: That makes sense. Tell me more about telemedicine. Where is it heading? When does it get to the point where a significant percentage of care can just be done digitally through the app by doctors?
MS: Yeah, we have a staff of doctors that takes telemedicine calls and secure messages. We were the first insurance company that made telemedicine free in 2014. So in all Oscar plans from the beginning, you were able to click a button, talk to a doctor, and get free health care that way. And these doctors were able to prescribe medication for you. What we’ve built over the years is a much tighter integration between the Concierge teams and those telemedicine physicians. So both directions, the Concierge teams to the doctors, doctors to the Concierge teams, can hand off cases more directly and more tightly, and that’s evolved powerfully to the point where now about two-thirds of all routine conditions of Oscar members, things like pink-eye and smaller injuries, have a telemedicine encounter as part of the episode of care. And that’s about 10X what any other insurance company has. What it does is reduce the cost of an episode of care for a member from something like $200 for pink eye to something like $40 or $50.
The metric we’re tracking pretty closely is the percent of all cases that we have that are going to telemedicine and can be solved through telemedicine. And there I think we’re at the beginning. As the insurance company, we’re in a unique situation to essentially say that the more cases our telemedicine doctors can resolve, the more we can lower total cost of care and increase the satisfaction for the member.
NT: Alright, I want to talk about the way we use AI, but of course part of the way you use AI is dependent upon the kind of data you’re getting in. And my understanding is that one of the most interesting projects you’re working on is how to reimagine a claim form and what to put in on claim forms and how to structure those data sets. Explain that project to me and your ambitions there.
_MS_The overarching idea behind this counterintuitive notion that we have to build our own claims system is that every incentive in the US health care system is really just a configuration in some insurance company’s claim service. And so if we say that the incentives in the US health care system are off, which you hear a lot of people talk about, it oftentimes translates directly back to the fact that claims systems can be more creative and more versatile in configuring smarter incentives.
To give you a very simple example, if we tried to give you a discount for where you get an MRI or for going to a doctor in off-peak hours, which I’m sure is something that’s familiar to you as a concept from any part of the recipe of your consumer world, that literally wouldn’t work right now because the most common claims format, by which your provider submits claims to the insurance companies, does not have a time of day field on the claim. And so the claims adjudication literally wouldn’t work. And to get to the point where we have the utmost flexibility in configuring smart incentives, in getting rid of weird authorization rules, in configuring more risk sharing and risk taking providers into the system, we thought we needed our own claim system. And that has, immediately, an impact on how much we do manually, and how much we can do in an automated way.
We pay claims, for example, generally in three to four days; most insurers pay upwards of 14, 16 days or so, and that’s because about 91 percent of our claims get paid through auto-adjudication. Meaning they get paid without a human being looking at them, a machine just kind of runs the whole thing through. Most insurers are somewhere in the 80s on that, so that right there makes a difference. That, of course, is less time spent on administrative overhead, it gives us more transparency into our data, and it lets us spend more time on configuring smarter incentives in way I think we actually should be doing it.
NT: So let’s go back to the time of day thing. So that’s super interesting. Once you have that you can implement the equivalent of surge pricing or anti-surge pricing, discounts at off hours, or more expensive at peak hours, right?
MS: I would rather do it as anti-surge pricing, but yes.
NT: Well, as Uber of health care, certainly you want to avoid the phrase surge pricing!
MS: And honestly I think [discounts] is where much more value lies. There is a capacity that’s lying dormant because people don’t want to go get an MRI done at 8 at night. We’ve literally had those conversations with contracting staff at imaging centers. And they told us “other health insurance companies can’t do it but we’d love to do it with you guys,” and that’s now finally going to become possible.
NT: Well, of course, right, because hospitals have to be staffed 24/7, because you have to be able to deal with any emergency, but that means there’s lots and lots of downtime. So how can you more efficiently maximize that downtime?
MS: Yes, for example, exactly.
NT: And are there other similar things you’ll be building into the claim form, like time, that haven’t been built into claim forms before due to inefficiencies that have similar impacts?
MS: There’s a lot around payments at the point of care that will get much easier, and we’ve been meaning to build this for the past couple of years. And this is one of those things where we can finally get to it with more staff and more money with Alphabet backing us. The way you pay bills in health care is still kind of odd. You pay the insurance company something, you pay the doctor something, the two have to reconcile behind the scenes in strange ways. It is oftentimes difficult to predict what something will cost, even if you have full control over incentive configurations or benefit configuration. So we’re trying to get to a world where, before you go to a doctor, here’s a button you push, here’s what this will cost, and we’ll settle everything up before the utilization even occurs. I think this is, again, a very unique way in which only an insurance company can really rewire the system and make it more consumer-oriented in a way that the rest of the world already is but that health care just isn’t.
NT: Got it. Now tell me how exactly beyond the things we’ve mentioned are you using AI? Obviously you have a background in AI. And I mean that sort of in the specific machine-learning AI sense, since basically any company that does anything with electricity now calls it AI.
MS: I haven’t heard the definition get extended to electricity, but I wouldn’t disagree with you there. So we use analytics, I would just say, in the broadest possible sense, in a number of different ways that span from very pragmatic, bottom-up construction of ontologies—in my days in the early 2000s, building ontologies was still what people thought of as AI; those were days of the rule-based systems and the semantic web and things like that—all the way towards training machine-learning models. With ontology construction, most drop-down fields for finding physicians distinguish something like a hundred specialties or so. Internally, we knew we had to redo all that and distinguish about 350 specialties to ensure members could get the right care with the right doctor. And that was essentially somewhat of an analytics-driven exercise to look at the clusters that form between what specialists treats, what members go to the doctor for, and things like that. So there’s a bunch of like clustering in there.
But also some very bottom-up clinical manual classification of issues that we had to redo. We did the same for CPT codes, diagnosis codes, all of these formats and ontologies you have in health care were developed for payment systems and not necessarily for clinical categorization of issues. On the machine-learning side we have very predictive models running. For example, the model that tells us whom to reach out to because the member might be about to go to the ER or urgent care is a machine-learning model that uses lab data, drug data, categorization data, and things like that. We’ve got tools internally that tell us in any particular geography how much utilization will there be for specialty care, for primary care, by these 350 different specialties and do we have the network in the area to be able to manage that in a good way.
NT: And do you use this system as you figure out what cities to move into next? Are you taking giant data sets of cities, demographics, et cetera, and figuring out where you can move?
MS: I’m glad you asked. We indeed do that. We have an internal process called market scorecards that takes all of the health care markets, about 350 or so in the US, and maps them into a categorization: How do the provider networks look in the city? What does the population look like in the city? Which health systems could we work with, and how good are they at what they do already on the population health side? And it distills for us a prioritized list of markets. So that’s how we go into new markets and how we roll into new cities every year.
NT: Wow, that’s really interesting. And then on these different problems, the different problems you just described, you’re using AI. Going back to the news, how much will your now deeper strategic partnership with Alphabet help? Obviously it’s the company in the world most focused on AI right now.
MS: So again they’re financial investors; there’s no access that they would have. We are extremely careful about that. We run an insurance company with the goal of making sure we look at your data only so you get better health care, and that’s the only reason why. So that’s the first important point to make. I think it will help us indirectly by the fact that, as I mentioned, I personally am inspired to work in a company that Alphabet has confidence in. I just think as a technologist, it’s a vote of confidence that I find pretty amazing. I admire what they’ve been doing in machine learning and data analytics and so on. I think indirectly there will be plenty of opportunities for us to learn from them as to how they look at data, how they analyze their own health care data sets, and questions they would like to ask of how the health care system operates. Whether in academic settings or other settings, I think there will be ways to work with others on answering smart questions. Salar is joining the board, and he’s been a friend and adviser for the past three years. He is a person who has been able to look at consumer industries from a data-driven perspective in a way that few other people have been able to, and so just having him give more of a look at this will be powerful as well.
NT: And tell me how you protect patient data, since the big story of the day today is the AP’s report that Google has been tracking location, even when people have asked it not to. You have the most sensitive data in the world, and you just started a partnership with a company that is in trouble over data privacy. How do you ensure people that you will protect them?
MS: Yes, to be sure to remind people, they’ve been an investor for the past three years and there’s obviously been no sharing of data before. That’s going to continue to be the case.
We have been an insurance company from the very beginning. The first thing we ever did is go through a licensing process. And so the attitude towards the regulatory bodies that maybe the other Silicon Valley companies have had in the past would have never worked. Not as an insurance company. One of our internal values is “Respect the rules. But fight for better ones.” You know, we can have arguments with regulators, we can say they ought to think about changing certain things, but we are fundamentally extremely at the behest of our regulators and work very closely with them. And so that extends to data protection and data privacy. From the very beginning, we have had to be extremely careful about building our systems in a way that they are very secure and uphold the highest privacy standards and HIPAA compliance and the various other standards that play a role there in a very important way. That will never change, it is very important, and every day is a new challenge, there’s no question about that, but it’s something we’ve been paying lots of attention to and have been very, very tight on.
NT: And have you changed any of your privacy policies since the Cambridge Analytica stuff?
MS: Not to my knowledge, because it’s not really something that applies to us. If you’re a provider under Oscar, your doctor can log in to a system of Oscar and can request your clinical data, but none of those things are related, from our legal counsel’s point of view, to any of the things that I think the social networks had to go through.
NT: But presumably you have much richer data sets. If you know the time that I talk to somebody about a health care condition, that’s hugely effective in giving more care, but it also means wherever it’s stored in your system it somehow also makes it easier to de-anonymize me, right, because every little bit of information is attached to every little person, and every little file in a large data set in some future time is possibly shared out of network, or possibly a hack could put more people at risk. So, as you get these incredible data sets, how do you add extra layers of protection at each step of the process?
MS: One, by having a team of dozens of people who spend their time on security every single day and making sure that our internal systems are tight. That’s the engineering team, the security team, and so on. Two, by having dozens of people on the compliance side that watch what we do. And as an insurance company and a technology company, our organization is an interesting sort of amalgamation of what you would expect in a typical insurance company and what you would expect at a typical technology company. Therefore we have a very big compliance staff, a very big legal staff, and they have layers and layers of safety and security in there in ways that I feel comfortable with. But we also say, every single year, we’ve got to be compliant, got to make sure we push for better protection and more security.
NT: Great. Another question I’m sure you get often: Your cofounder is the brother of the son-in-law of the president of the United States. How have you navigated that in the year and a half that President Trump’s been in office? How has your view on how to navigate that issue changed?
MS: It hasn’t changed from the beginning, which is: This does not affect what we do. The business hasn’t gotten any easier in the past couple of years, but it’s never been easy, and I think if we don’t obsess over trying to create a better health care experience and making sure we deliver that promise to our members, then we wouldn’t be spending our time right. And that’s what we’ve been totally focused on. And whatever happened in the press, whatever happens on the regulatory side, I have personally always thought, and I think the company shares in this, that if we have something that leads to lower costs and happier members in some shape or form we’ll be able to turn this into a successful company. However, the regulatory environment changes over time, it’s something we have to be mindful of, of course, as this happened as we watched it, but it never affected what we do day to day. And so from that point of view, any kind of disconnectivity, however it looks, never played a role inside the company. When we get out there and make a point about how we think health care should work, we do it publicly and we answer questions to the Congress and whatever else, but that’s it.
NT: Got it. OK, last question since I know we’re running out of time. Tell me one problem in health care that you have not been able to solve yet that you hope to be able to solve in the next couple of years.
MS: I would say—and this will take many years—but curing a complex issue from afar. I think there will be lots of ways in which telemedicine will become even more powerful, and that’s one we haven’t been able to solve. I think that’s one that will keep us occupied for years to come.
NT: Excellent. Alright, well thank you very much. Thank you for talking with WIRED.
MS: Absolutely, it was nice to talk.