Ep 73: How Candid Health Is Fixing the $280B Healthcare Billing Fiasco, With Lessons from Palantir

0:00Unsexy businesses are the ones that can scale really big. Again, I, for better or worse, get excited about unsexy businesses. I think Stripe at one point was an unsexy business, like payments, and they made it cool. I also think it is a very worthy problem. At the end of the day, I'm sure we've all heard about patients who have these crazy bills by insurance and they owe this thing. And we end up saying like, oh, insurance are jerks for not wanting to cover that. And in some cases, that is true. In most of those cases, the provider submitted the claim wrong. At the end of the day, what we could be doing if we get help all of our providers do this is we're dramatically improving patient experience, which I think is a very important problem.

0:42My friends, Nick and Doug, we're superstars at Palantir. They're building a company called Candid Health. Thanks to new possibilities in AI, we're going into areas of the economy that sound boring, such as healthcare billing. These guys are working on healthcare billing, a$280 billion area of the economy. It's a giant mess. There's hundreds of thousands of people tearing their hair out trying to deal with healthcare billing. This creates all sorts of problems for healthcare systems, creates all sorts of problems for building and fixing things in healthcare. And it turns out that thanks to new possibilities with tech and AI, you can actually solve this, make it a lot cheaper, a lot better.

1:13This is going to add a couple of points to our GDP. It's going to make healthcare work better. And I'm excited for you to get to know them and hear how this is possible. I'm Joe Lonsdale. Welcome to American Optimist. We have Nick Perry with us today. Thanks, Joe. And Doug Proctor. Thanks for being here. Thanks for having us. We both were at Palantir for about five years each. Nick, you were there in 2012, you're saying, for five years? Yep, 2017. And Doug, you got out in 2020. Where'd you grow up, Nick? Yeah, I grew up outside Seattle. Stayed there my whole life until college time. Came down to the Bay Area to go to Stanford.

1:40And then I was at Palantir for five years. I did mostly healthcare work there. And we'll probably get into this, but some of that ended up being the genesis for Candid. And Doug, how about you? Grew up in the D.C. area. Youngest of four boys. Played a lot of aggressive sports with my brothers, but also went to Quaker school. knew nothing about the military, but felt a really strong sense of wanting to do something in government, which is what took me to Palantir. I ended up spending most of my time there working in the defense space and yeah, joined Candid when we were about six or seven people.

2:08Nick had started it. I had done some other stuff before that. I remember we were curious what you're going to do next year. People are looking at watching you after you left Palantir. You're one of the top talents as was Nick. And so very excited to see you guys teaming up, you know, just starting on Palantir briefly. What are, what are some of the most important lessons you guys learned from Palantir? Yeah. So I think the three things that really kind of resonate and continue to resonate. One is, I guess I should just caveat to start, like unbelievably grateful for the opportunity and experience work at Palantir.

2:32I think it's a incredibly formative experience for me in my life. I think a couple of the lasting impressions, one is take on hard problems, just like seeing firsthand how we can actually have a huge dent in really big ambitious problems that are, I think for a lot of people's imaginations, just kind of like unsolvable. I think giving us that sort of space to work on those things was one big one. I think another big one was just like really heavy, heavy, heavy focus on the quality of the people. Just seeing firsthand the amount of energy, tradecraft, care that goes into the hiring decisions that were made at Palantir, and then the quality of the people that they were able to bring in, and then how to deploy those people against the problems.

3:07I think that was number two. And then I think three was really just like really heavy focus on going to first principles, thinking about what is the actual problem we're trying to solve, and then getting really tactical. You said some good things about Palantir. What are things you'd want to change and do differently? Yeah. I mean, I think the big thing that I reflect on having now left Palantir is obviously there's a lot of stuff we wanted to kind of bring from the Palantir experience while we're doing it candid. I think if there were one thing that we're probably doing somewhat differently, it's a much narrower and tighter focus.

3:35I think Palantir was often kind of solving a wide range of problems across most industry sectors. Try to build somebody to go everywhere at once. Everything for everyone, always, everyone. Very much my personality as a 21 year old. Yeah. And at Candid, it's quite the opposite. At Candid, we're making medical insurance billing better for healthcare providers in the US. And that's a pretty tight focus relative to everything for everyone everywhere. And it sounds like medical insurance billing is a pretty small problem, but I guess it's about$280 billion a year. It's a massive industry. It's a huge industry.

4:04So one of the reasons I want to talk to you guys and show people what you're doing, aside from the fact that you're a top talent from Palantir and I'm excited about your company, is there's this theme in the economy right now, which is that there's all these areas we spend tons of money on. And it looks like the new advances in AI, it looks like really top tech cultures combined with AI can make these areas more efficient, could actually meaningfully impact the productivity of the economy. I mean,$280 billion spent a year, that is like a meaningful part of our GDP. And it looks like you guys are going to be able to make it more efficient.

4:31So first of all, before we go into this, what is medical billing? Why is this costing$280 billion a year? Yeah. Medical billing at the highest level is the process of that a healthcare provider goes through to get paid by a health insurance company or by patients and sometimes employers in the US. And so there's a bunch of steps in that process, starting from going in network with insurance to ultimately submitting claims and to the insurance company and then getting this claim adjudicated. And that process in particular, which is where we focus, submitting claims, submitting the correct claim to the correct payer is extraordinarily difficult.

5:04How many payers are there? Over a thousand. And then how many types of claims are there? There are many types of claims. I think probably thousands of different procedures or codes that you can actually submit to insurance. And all of all those thousand, over a thousand insurance companies literally have different rules for how to submit the claim. Why didn't they normalize? That is a great question. They didn't. They standardized the form, but they did not standardize the field, what you put in each field on the form. And so they might be like, one parent might be in this box. I want a taxonomy code.

5:33And if you do not put the taxonomy code, I will deny your claim. And another parent will actually just they don't care about a taxonomy code so you need to you need to know all that stuff and the process what we're focusing on is how do you and why this is so expensive so manual is because there's no good software today people have to do this by memory or they're like creating their own rules or they're like doing things they're hand jamming there's a lot of manual work to actually try to do all of these things correctly the first time and there just hasn't been any good software built to to do this stuff for them it just seems like an infinitely complicated a problem for software.

6:04You'd have to have people who become experts on this. I mean, are these people paid a lot of money because there's like, or do they have this like a large book that they're reading to do it? Like, how do you figure this out? Like, how do you know what to put where? Yeah, there are some pairs. It's a mix. Some pairs will publish stuff. So there might be some PDFs. Sometimes pairs won't tell you and they haven't published anything and you need to submit a claim and then they do something and you call, they literally will call the pair and they'll tell you why they adjudicated and what they want you to do.

6:31Do you have something to tribal knowledge? People call them and take notes then because sometimes you have to call even if you're perfect. We're focused on, we will call if we have to to learn these rules. But what we're focused on is we want to take that feedback loop of submit. We want to submit as many claims correctly the first time. And so we're trying to emphasize, call fewer times. Cluster your claims. Find which ones were denied for the same reason. Make one call. Don't call for all thousand of the ones were denied. That's smart. Take that rule. Put it in the rules engine and don't make that mistake anymore.

6:58Just move on to the next thing. Another way to think about the dimensions of the problem is think of 10 or 15 different disparate data sets that don't really have purpose-built systems. And the task of medical insurance billing is perfectly aligning to the center of those overlapping Venn diagrams every time you submit a claim. The challenge is that there are many, many, many different specifications of the rules that need to be followed to submit a claim properly. Moreover, the payers are also at the mercy of their legacy system. So as much as a pair might codify a policy in a PDF that they release quarterly, they have a 30-year-old system that they're instrumenting to adjudicate these claims.

7:32So it's buggy, as you can imagine. So sometimes they might accidentally reject something and you have to tell them that they messed up? In fact, there's a great story of times when... Yeah, actually, I know there's a large pair that I know manually processes one in eight claims because their adjudication system doesn't do it. It's a large pair. We'd all know the name of it. Do you think eventually you could help them too with what you've built? I 100 % think we could because if you think about what we're doing, we're pre-adjudicating the claim. We know what the payer wants on their claims. We know what their rules are for how to submit a claim.

8:01And there's a bunch of noise of every time they deny these claims that could have been avoided, the provider's going to have to call the payer and now the payer needs to pay for support people to answer those questions. They don't want to pay for that, I don't think. And so there's very much a world where I think we could just sell what we're doing, our rules engine, back to the payers and be like, look, we're pre-adjudicating these things. You would have adjudicated it anyway, and I can save you the support staff money. So Nick, you're the co-founder and CEO, and Doug, you were looking at a bunch of options, and you ended up joining here early.

8:27What made you guys passionate to do this? Why is this the thing you want to spend your lives on? You're extraordinarily talented guys. What about this is interesting to you? A couple of things. One, I was looking for a good business, and I think I subscribed to the Polygram. I think it was Polygram, saying like unsexy businesses are the ones that can scale really big. And I, for better or worse, get excited about unsexy businesses. I think Stripe at one point was an unsexy business, like payments, and they made it cool. I also think it is a very worthy problem. At the end of the day, most of the time, I'm sure we've all heard about patients who have these crazy bills by insurance and they owe this thing.

9:02And we end up saying like, oh, insurance are jerks for not wanting to cover that. And in some cases, that is true. In most of those cases, the provider submitted the claim wrong. And so at the end of the day, what we could be doing, if we get help all of our providers do this right is we're dramatically improving patient experience, which I think is a very important problem. It's a good mission. I mean, do you actually think about like, if right now it costs$280 billion a year, should it cost half that much or something if someone's doing this better? Easily. At the end of the day, there's no software.

9:28They're throwing bodies at the problem. The last innovative thing we did in this industry was offshore it. I guess you want to make it cost like a much, much smaller fraction of that, but then capture most of that yourself. Exactly. That's right. That's awesome. So let's go to the tech. I'm very interested in what you guys are doing here. You know, Palantir early on was like specifically not doing AI. It obviously is these days, but it was all about augmenting people. I think even when you guys were there, but obviously in this case, you want to augment people, but then you also want to do a lot of AI, I'd imagine.

9:55So similar, I guess, more to Palantir today. That's what's possible. And so like how have the recent AI breakthroughs the past few years impacted your strategy and what are you doing with AI? Yeah. So I think like, first off, I would say from the pretty outset, I think we, because we've had the luxury of building Candid on modern software tools, we have the ability to really be thoughtful and forward thinking about how do we deploy machine learning from early on. So some of the initial prototypes that we've built and deployed and sort of run against production data essentially allow us to do things like predict if a claim is going to be denied pre-submission.

10:23So you can actually codify the rules for the payer, determine based on the metadata on the claim before it's submitted to the insurance, like what's the likelihood this claim is going to be denied and then just not submit it just because you've seen a thousand claims just like it. And you're like, okay, cool. This little thing needs to go in this value. And if you make that change. We have historical data that suggests that will get paid. And in fact, that's how you really derive the rules. And that's the second piece. How do you get the data to start training these things? It seems like a hard chicken and egg problem.

10:47We submit claims. We have around a hundred different customers in production today. You just started doing the job as best you can, and there's adding AI to it. Yeah. And what we really have is an opinionated data model with really good fine-grained detailed data reporting, where we can say, here's the very precise problem that we're seeing. And that is one thing that our legacy incumbents don't have. They have this like very messy data model, very messy systems, and they're just throwing stuff into things like a JiraQ for offshore to fix, whereas we're actually going through with a fine-tooth comb, root-causing claims, and then codifying the rules engine, which is really our secret sauce.

11:16It does sound very Palantir in a way, obviously, in terms of the logic and the frameworks. It requires you to go and sort through claims data, which is very gritty, very unsexy. And it's categories that organize, they have the ontologies, have the labeling. And it's interesting. So when you first start submitting claims for the business, that was obviously a very negative margin business because you didn't know what the hell you were doing, I assume. So you just start going. What was that like when you were your first month or two? It was painful. I was our first biller for a very long time. And I did.

11:42I think it's important not to get on a soapbox. I think it's important for founders to truly understand their product and to actually get in the weeds and do this stuff. And so I thought it was very important for me to actually be a biller for a little bit to truly understand what is the problem? What is the state of the art of the technology available today to figure out what it needed to be. And did you use other software at first? Like what do you, what does a biller even do? Do you have PDFs you're searching through and trying to figure out what to do? There are various ways to do it. There's software that might be through an EMR.

12:10Most of the, um, many pairs will let you submit claims on their front end. So we might've done it that way. We, we started, we just built a hacky version of a very manual process of just to make claims over the clearinghouse, go to the payer. And I was in charge of submitting all that stuff. So a lot of the late nights of manually submitting claims until we figured out how to do it correctly. What is really powerful about where we are today is we have this incredibly rich data set that we essentially have the end-to-end operational loop that we control. So we control the preparation, we control the submission, we control the entire cycle of how you work this claim to completion, and then instrumenting the upstream rules to essentially never solve the same problem again.

12:49It's a very rich area to apply AI. Yeah. There should be ways to basically get the margins way, way up more. With like the configurability and the tooling, because I think one of the things that our income, our sort of legacy incumbent competitors suffer from is they don't have tools that they can configure to actually close the feedback loop. So once they have identified the problem, they don't have a place to go back and say, OK, cool, mutate this claim the next time we have that high precision plus the closed feedback loop, which lets us essentially operationalize the insight in a much more streamlined way.

13:21that is i think going to be a lot of the value prop for how we're going to deploy some of these like new machine learning and ai techniques i think as we learned from palantir automation and the actually useful application of machine learning is the tip of a very large iceberg and we're building the infrastructure under the ocean right now of like the infrastructure for how do you do all that automation correctly and the machine learning correctly in a way that actually works and that's not something our competitors have done this is what people don't realize about ai you don't just like magically cast like ai you don't sprinkle you actually have to like do all the really hard work to get the infrastructure, get the ontology, present this in the right place.

13:53And now you know how to use it. And now it's not as hard to use it. And owning the operational feedback loop. Like you're actually at the point of the operational decision. You're not just building some like AI analytics engine. You're actually like changing the destiny of the claim. Because you're in the loop on that. You own it. Exactly. And then you go after the$280 billion industry and then you make it a lot cheaper and make everything work better for everyone. The thing just to throw out is like, I don't know if this is interesting to weave in as well but in terms of like probably 20 if you think about sort of 2024 product themes i think a big part of what we've built is this end-to-end platform for managing the preparation submission of claims parsing the metadata back from the payer modern apis sdks for developers integrate with but we also have built in a way where we plan to modularize this thing so we can express just the rules engine via an api to a hospital system that wants to fix their claim routing problem, right?

14:45So there's pieces of this puzzle that I think are deeply intrinsically valuable, even if you peel them off and provide them sort of as a one-off sort of API. There's just so many different pieces of healthcare where you can go in and fix something. So rather than have one key thing, you want to be able to touch everything else separately. As an option. Awesome. Makes sense. Tell us more about the business. How many people are in the company now? Just over 50. How many of those are engineers building things? 20, oh, just over 22. So it's a big, it's a big engineering team. Big engineering team.

15:11And what have they built? What are they building? What's going to change things? Yeah. Right now we have the end-to-end platform for submitting claims to insurance, getting the results that ERAs back to tell you how the claim is adjudicated. What's an ERA? Electronic remittance advice. It's the, in normal terms, it's like the receipt. It's like the pair telling you how they've adjudicated your claim. I thought ERA was like, you know, how many. Earn run average is different. I wish. Um, uh, so we have the whole platform for, I think phase one of our product is just get the, all of the features.

15:45There's a lot of steps to RCM. We needed all the features to actually submit claims, get the results and figure out what's going on in the build the rules engine. That's building, getting all the infrastructure in place. We needed a place to put all of the important data elements. Doug was talking about your provider contracts, your credentialing information, all that stuff we need to know before we split the claim. I think phase two, the next things that are coming out for us are doubling down on our rules engine and letting it do increasingly sophisticated things. The rules engine, I think, is probably the brain of what we're doing.

16:13It's probably the most valuable part of what we're building, and it's how we can submit, in some cases, 99 % of claims correctly the first time. So you have a pretty complicated rules engine, depending on which insurance company, which type of claim, all sorts of scenarios that make it more complicated. Exactly. And I think, as Doug was talking about, continuing to layer on, I think the recent AI stuff has been pretty cool. I think - So how does the rules engine, because the rules engine is not AI. That's just like you having rules, although the AI can help teach you how to make the rules engine better, I guess, right?

16:39Suggested rules. That's the other big one. Interesting. So actually, you actually create new rules for yourself. Exactly. Yeah. So that's another area where we're experimenting essentially is like, if you think about what our business is, we're basically helping to manage the complexity with this insane process. The way we manage that complexity is through storing the data in interesting ways, operationalizing the logic in the rules engine, and then closing the feedback loop between how the pair adjudicate the claim and what should the next claim go out with in terms of mutations or validations.

17:02And so step one of how we've sort of prototyped and experimented with some of this new AI stuff so far is like predicted denial, don't submit the claim. Phase two is essentially instead of having to have a really like analytically oriented person digging through claims data and finding out the rules, it's actually like, how do you suggest the rules? Okay, these are the patterns of issues we're seeing. Consider this as a recommendation engine. So you really turn the user from more of an author into more of an editor who's then going to go in and say, this looks good. This looks bad. Let's test this.

17:30And that's kind of where we're headed with the product. And every new client, every new, every week, you probably get better. Exactly. Are there lots of times now where you've submitted something that you knew was right and then you had to teach them that they were rejected it incorrectly? Has that happened very much? I think a useful thing to talk about a bit is how we onboard customers maybe. So like when we have a new customer joining the platform, essentially the first thing we do is we spend a couple of weeks going through, if they've previously submitted claims, we go through all their historic claims data.

17:54We basically go through it and run it through our system and say, that's really valuable to you. Here's a bunch of stuff where we can create value for you going forward. Here's a bunch of things where you're doing it wrong today. And this is why you've seen denials. And in many cases, these are customers who have like 20, 30, 40 % of their claims were being denied for reasons they didn't have visibility into. Now it's like, here's the problem and here's how we fix it. And that's the value problem. There's literally customers out there with 20, 30 % of claims being denied and they don't know why at all.

18:20And you could fix the majority. It's complicated. And it's like, you have to become an expert in this very arcane legacy administrative task. And people are just trying to deliver healthcare services to patients. Of course, if they're delivering healthcare, they're not thinking about arcane rules for insurance companies. We didn't talk about this at all, but another layer of complexity is, in the Palantir sense, these are adaptive adversaries. The rules change. And so that's another layer of complexity. I mean, how often do the rules change? Is this like every year or all the time? And do they tell you when the rules change?

18:52Sometimes they'll publish stuff, like I said, but other times they'll just change sorry we just changed our mind about how we do this this might blow your mind but a couple years ago one of our customers all of their claims started being denied it was just their claim we could not figure out why we ended up getting in touch with the payment the department the claimant department of that payer and it turns out we were the first company I think we figured it out we got in touch with them within a week and a half of this happening we were the first company in the country to identify that this very large payer had put a bug into their rule system wow and they fixed the bug based on like us pointing out, hey, you guys, something seems to have happened here.

19:27Wow, because you just like, you just figure it out faster than anyone else because everyone else thinks they're doing it. It's about that feedback loop. It's about submitting claims correctly the first time, reviewing the results of reporting and building rules and rules. Seems like you need to get the business on the other payer side pretty soon too. Because that's going to make this a lot more powerful for everyone. Exactly. So tell us about the business. You raised your Series B earlier this year. We've been involved with the company, obviously at AVC since early on. Tell us about your growth.