The ShiftShapers Podcast
Change either paralyzes or energizes - the choice is yours. Hear from businesses and entrepreneurs who have become energized and who have profited by shaping the shifts in their markets and practices. Become a SHIFTSHAPERS INSIDER and get our latest download, advance notice of all podcasts, podcast summaries, and special INSIDER-ONLY content. INSIDER SIGN UP
The ShiftShapers Podcast
Ep #313: Using Data To Decide When To Go Partially Self Funded - with Hari Sundram
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
This episode of The ShiftShapers Podcast is called "Using Data To Decide When To Go Partially Self Funded" with Hari Sundram, CEO and Founder at Verikai. Today, David and Hari discuss the kinds of data that are used as a surrogate for risk.
Hari explains how brokers can inform their clients on predictive modeling and how having large data sets helps cut through anecdotal biases to provide new premiums and new outcomes.
What You’ll Learn From this Episode:
- 01:27 The right time to go self-funded
- 04:32 Data as a surrogate for risk
- 09:36 Data analysis: How brokers should inform clients
- 16:14 Predictive modeling: Creating new outcomes with repetitive data
Quotes:
07:45 “Ultimately, the outcome we try to look at is loss ratio applied at an individual level. That’s our goal because ultimately that’s the lynchpin of understanding whether the group has the risk necessary.”
11:19 “When you get back that score, a scenario analysis is laid out for you that says this is a score and this is how it translates into the placement of the business.”
18:24 “The only thing more dangerous than no information is a little bit of information because with a little bit of information, you’re able to apply your bias even more than with no information at all.”
21:30 “The value of predictive modeling is its ability to cut through those things that are anecdotal and find pattern that have occurred historically but have occurred historically repetitively.”
24:41 “If you want to write new premium, you need new information. What you want to do though is make sure that that new information does actually index to the outcome because the outcome is actually consistent.”
How can data help you decide if a partially self-funded plan is right for a group and save money on stop loss and predict claims? We'll find out on this episode of Shift Shapers.
AnnouncerChange either paralyzes or energizes. The choice is yours. You're listening to the Shift Shapers podcast. You're about to learn firsthand from businesses and entrepreneurs who have successfully shaped the shifts in their industries. Get ready to become the change you want to see. This episode is brought to you by Shift Shaper Strategies. In sales, if you confuse, you lose. Clarify your message so you win more clients, crush your sales goals, and build your practice. Learn more at ShiftShaper Strategies.com. And now, here's your host, Story Brand Certified Guide and Chief Transformation Strategist. At Shift Shaper Strategies, David Saltzman.
DavidOn this episode of Shift Shapers, we're speaking with Hari Sundrum. Hari is the CEO and founder at VeriCai. They use an awful lot of very deep, very clean data, and they do it in two key areas. And we're going to talk about both a little bit today. One is in capturing new business and something that's of interest to an awful lot of listeners, which is how do you take a small, non-credible group into a self-funded environment? When do you know it's the right time to do it? How do you figure out when the wrong time to do it is? And also, we're going to look forward and talk a little bit about predictive modeling and how you manage looking out the windshield rather than looking out the rear view mirror. And with that, welcome, Harry. Thanks, David. Great to be here. It's my pleasure. So as we said in the open, the problem for a lot of advisors these days is there's this renewed interest in partial self-funding. And yet it's very difficult for an awful lot of those advisors to either A, get data, so they end up relying only on health statements, or the data that they do get is either imperfect, not clean, partial, or just completely, sometimes I think intentionally goobered up. What does an advisor do?
SPEAKER_02I think that's actually a great question because it almost appears as if the system acts in a way to limit your ability to do the right thing for the employer, right? And I think that that's the overarching observation for me is how there's an obvious methodology to be able to leverage to get your groups into the rate that's commensurate with their risk, and an almost active process in market to literally limit that from happening. So today, your optionality is doing IHQs, which, from my observation, don't actually inform you of anything around the rate, your ability to place the business, your ability to get a quote. It almost appears like a subjective filter that gives even more reason not to do something, right? Or to be able to go ahead and find pockets and/or areas of the country where discounting the manual, even at small group sizes, allows you to still be competitive against a fully insured rate. As you would imagine, neither of those two things is ideal, right? Because neither of them advance the ultimate objective, which is I want to be able to go ahead and get my small group into a tolerable premium that will cover their risk, right? That's really the goal. And the existing systems don't allow you to do that. I would argue that the best way around that is one of two things. Either A, you find information that serves as a surrogate for risk, that doesn't require you to have to force people to divulge information, or B, you're going to have to ask people to divulge information, but then extract that information on their behalf. A is a lot easier than B, right? And if you can find that data that acts as the bridge of credibility, then you can start moving information and moving groups into rates that are commensurate with their risk and are credible to quote into market. And then you get change, right?
DavidThen you get change. So what would be an example of information that serves as a surrogate for risk?
SPEAKER_02That's a great question. So just an example that we see in or have seen in our business repeatedly is that when you look at the actual manual and the partially self-funded arena, women between the ages of 18 and 34 are always priced from a loading perspective and a premium perspective at roughly around $300 per member per month, right? That's obviously factoring in all of the other components of the load. But the reason that it's set that way is actually the biggest risk of for women between the ages of 18 and 34 is pregnancy and all of the possible complications of pregnancy, from just having it, requiring it to be surgical all the way to NICU babies and neonatal risk. But what if you had tractable information that's highly correlated with the probability that that woman will not get pregnant within the next 12 to 18 months? So we've discovered segmentation from a data perspective that indexes to women that are that drive two-door cars, that are rapidly increasing their savings rate and dine out at an extremely high amount a year, anywhere from $10,000 to $20,000 a year spent on dining out. And of course, they also have purchasing behavior that is correlated with purchasing birth control pills. As you would imagine, this segment obviously describes women that are highly unlikely to get pregnant within the calendar year, right? They're not family planning, as evidenced by all of their behaviors. And the clinical superimposition of them actually being on birth control tells you that even clinically, they're obviously trying to actually mitigate childbirth. Why then should that person be fit into a pregnancy weighted premium or a pregnancy-weighted load, right? If you knew that information and you knew that not only at the individual level, but you could roll that up to a group understanding, that group should be discounted. And that discounting would be advantageous to anybody that's sponsoring that group to be into a plan, right? And it's that kind of information where you're able to get meaningful delta from something that's actually around consumer behavior, very logical to understand.
DavidSo you do this by essentially dumping a census into a system that looks at all of these different behaviors?
SPEAKER_02Yeah. And so to peel it back a little further, the way it actually works is the entirety of the system, right? All 5,000 variables we have on 260 million people is partially based on the things that people probably commonly understand, things like purchasing behavior, econometrics, credit, social, demographic, financial, things like net worth, you know, net income. And then superimposed on that is a massive amount of claims information, primarily claims, historical claiming, total assigned premium, loss ratio. Ultimately, the outcome we try to look at is loss ratio applied at an individual level. That's our goal, right? Because ultimately that's the linchpin of understanding whether the group has the risk necessary. What we're very, very, very, very, very good at on our platform is our ability to predict that aggregate attachment point. And the ability to predict that aggregate attachment point provides you a lot of flexibility in your ability to inform what the true risk of the group is. So it's not just the behavioral data that we incorporate in, it's the behavioral data trained and hardened on loss ratio outcomes from our carrier customers that spans 20, 25, 30 million claimants over multiple years, right? That actually allows us to say, okay, we now understand that because we not only have seen certain behaviors before, we've also indexed them to an actual total claiming and an understanding of the assignment of premium to that total claiming. We know where the manual is fundamentally mispricing the risk of an individual, and in particular, is not accounting for the fact that human beings have one, a probability of getting a disease that honestly is fairly predictable, actually. But the other is their desire capability or the means to actually manage that disease. And that is not factored into anything that goes into pricing risk. And that's what we're solving, but we're solving it by making sure it's always grounded on an understanding of the true outcome, which is the loss ratio.
DavidSo if I'm an advisor and I have, let's say, a 70 life case that's been in the fully insured universe forever, and I feel is maybe being taken advantage of, or they they ought to at least look at some other options, and I bring you that case, what's that process like? And what do I get back at the end of the process that that informs my decision and my client's decision?
SPEAKER_02So the way the system works today is when you submit that 70 person census, what you'll get back is a historically what customers will do is they will also input in the fully insured premium and the commiserate rate hike, because let's not kid ourselves, everything's got a rate hike these days, no matter how good the group is, apparently. What we will, what the system does is it processes it, it runs it against all of these behavioral elements that we have, and then superimposes that against the known loss ratios of of similar people, and is then able to take that and project it under an understanding of where the manual would rate every one of those individuals and can roll that up into a group level score. Now, that group level score is then directly correlated to the fully insured premium on a weighted basis because we know we the system already knows what the network discounts and everything else that's layered into the fully insured premium is. So what the system knows is exactly what the fully insured premium should be, and in fact, what the discounting authority of it maximally could be as well. And so when you get back that score, the a scenario analysis is laid out for you that says, this is a score, and this is how it translates into the placement of the business, right? Should you stay, it'll inform you, should you stay self-insured? Should you ask for rate normalization, i.e., that the rate hike is not commensurate with what the group's true risk is, or should you accept the rate hike because there is impending and persistent risk of the group? And there's also the capability for you to download a report that will walk through what the scoring is for each individual and the data elements and segments used to determine that risk, right? So you'll have a complete understanding of why we feel that way. In the future, not too long into the future, we're also going to give the ability for agents and brokers to be able to place that rate through us as well, because we have a multitude of carriers and MGUs who have assigned us credibility against that rate through a validation process. Our goal is ultimately that not just to tell you information about your group that tells you what that true risk is, our goal is ultimately to get a find a home for that risk that gives your employers optionality. If at the end of the day they choose to stay being fellow self-fully insured, that's their choice, right? I would argue that the vast majority of groups in the US shouldn't be, but that's you know, one man's argument. Having said that, being empowered with information, especially credible information, that could that does ultimately lead to a quoting value is my goal. I'll be completely transparent and as frank as possible. One of the things that I have always admired, even though I think people could say Elon Musk is crazy, one of the things about Elon Musk I've always admired is that he built Tesla ostensibly to have the build the greatest electrical car that has ever been seen. But a secondary goal that he hasn't been shy about saying is that he wants electrical cars to be the standard, the automobile standard in the world, right? And he's willing to do that, but he realized the only way to do it is to make it a capitalist constraint. Meaning you have to believe that making an electric car is a profitable endeavor and more profitable than what you currently do, or else there's no reason to adopt it. So building an economic ecosystem allows for the right thing to happen. I believe by telling people what the real risk is of the group and where the premium really should lie, my hope is I really hope that fully insured carriers do bring down their prices, that they limit their rate hikes. Like that would be just as great as being able to get more groups into being partially self-insured. So I should be, I should be truthful. It would be even better if more were partially self-insured. But if the ultimate value proposition is a true social good and the rates stabilize, have not come down, then we've done our job. And that would actually ultimately make me unbelievably happy. We're all to blame for what this has become. Brokers, agents, doctors, insurance companies, we're all to blame. So we should all be the ones to help fix it, right?
DavidEmulating Elon Musk is probably a good thing to do as long as you don't let him name your babies.
SPEAKER_02Yes, I still don't know how to pronounce it. I've been given the syllabic pronunciation, but it's obvious that the state government of California can't pronounce it either. So I don't know. I think David is a totally solid name. So if you want to suggest that to him, you should. I uh David or Harry, or you know, there are loads of totally solid names. A lot of good names out there, a lot of good names.
AnnouncerAnd now, a word from our sponsor. It's a fact. Salespeople and organizations lose opportunities because they don't clearly communicate their value. In today's market, your story is your message. It should be crystal clear, perfectly arranged, and precisely targeted to attract the clients you want. As a certified story brand guide, we use the exclusive SB7 process to create that story and the websites and collateral that deliver it. If your message isn't cutting through the noise, we can help. Visit us at ShiftShaper Strategies.com to learn how we can help you find, clarify, and deliver a message that wins clients, crushes sales goals, and builds your practice. In sales, if you confuse, you lose. So learn more and schedule that call today at ShiftShaper Strategies.com. That's ShiftShaper Strategies.com. And now back to our discussion.
DavidOne other question before we move on to talking briefly about predictive modeling, which is how valuable would this kind of a process be for an enforced, partially self-funded case where maybe the broker thinks that the numbers, even in that sense, aren't working in their advantage or aren't where they should be?
SPEAKER_02So one of the things that we've observed by looking at the data is that spec risk is spec risk and it'll always sort of be spec risk. So whether you have 36 months of contiguous data or 48 months of contiguous data, the probability of a high-risk claimant showing up is inherently something that the existing methodology wouldn't predict anyway. The assumption is just that there's enough premium existing in the group that it can account for it, and that there's enough claims history that the accuracy of the aggregate attachment point is much, much higher, right? Where that breaks down is when you have 18 months or less of data. And certainly, certainly, if the plan has historically BUCA sponsored from an ASO perspective. And they nickname it that because unless you have like top secret Cigna access, the information that's sent out is just so laughable and so sparse that your ability to give it credibility in terms of saying this should be the aggregate attachment point is unbelievably risky. So we view groups that have nominal history, so less than two years, to be groups that work really well with our behavioral information. And as it turns out, in our experience with customers, even better. Because the only thing more dangerous than no information is a little bit of information. Because with a little bit of information, you're able to apply your bias even more than with no information at all. And so that window between this is a scientific fact and this is no information at all is usually where fiction lies, right? And so many people in that self-insured to self-insured market today, where there's a limited claims history, are basically re-rating on fiction. And, you know, just like it would be with no information, our goal is to make that fact right. And what we've actually found is that the mispricing in self-insured to self-insured year over year, especially with less than two years of inform of information, is actually worse than it is in fully insured to self-insured. So that's pretty amazing. That is pretty amazing. Yeah. And I think anybody who's been in the business would probably say that as amazing as that is, observationally, they probably have noticed that that's probably true.
DavidAmazing is not always surprising. It's just sometimes it's just amazing. Sometimes we've got just a couple of minutes left. I I know that you guys are working with incredibly rich, incredibly deep data sets. How does this factor into predictive modeling or trying to get ahead? I mean, all of underwriting since cavemen has been looking in the rearview mirror, even probably before rearview mirrors were invented. How do we start taking this data and using it to look out the windshield?
SPEAKER_02So one of the things that was interesting when we first started this business is we we, I mean, legitimately, quite honestly, a fairly smart agent literally went ahead and said to us at a huge PEO, by the way, literally said to us, I don't understand how you can predict the future on past information. And I really genuinely tried not to laugh because obviously I think it was a heartfelt question. But anybody who's been a parent knows that when their parents told them something and they didn't listen, their kids basically did the same thing, right? And everybody would sit there and recognize, like, well, yeah, you know, these patterns have been seen before. But the thing is, they haven't been experienced by you. It's very hard for people to wrap themselves around predictive modeling because predictive modeling kind of flies against the face of human inference, right? And the way human inference works and has always worked is I know it's true because I've seen it, right? It's the basis of underwriting, actually, which is anecdotal evidence of experience. Now, like, is it actually something you should drive your entire business around? No. Do insurance companies do it anyway? Sure, right? But uh, because you can't go wrong saying that something is really bad risk. If it's happened a couple of times, it's something you can point to. But it usually is not in and of itself actually predictive. So the value of predictive modeling is its ability to cut through those things that are anecdotal and find patterns that have occurred historically, but have occurred historically repetitively. Not just occurred historically in singular events, but have occurred historically, repetitively in a way where its repetitiveness can have an actual value assigned, and that That value assigned could be validated even if you were blind to all the elements of the data. And see, that's the key, right? So the key with predictive modeling is the system says I've seen this pattern before. B, it says this pattern is repetitive. C, it says that repetitive pattern has a tractable value. D, that tractable value is something I can operationalize as a business. And then E, which is really the most important thing, is that tractable value that I can operationalize as a business and that I've seen before occurs enough that it will occur again. Right? So it can't just be that you've seen it multiple times. You've got it's got to be that you've seen it multiple times at a high enough rate for it to be predictive and valuable. And that's what I would say. What a lot of people do is they put in filters and say that those filters are predictions, right? Like let's not talk, like we are starting a prostate health company. Let's not talk to women, because you know, women don't have prostates, for example, right? So historically that's true, until you find out that when you look at the predictive patterns, what you learn really quickly is 60 to 65% of initiated prostatic discussions come from wives, not husbands. So that's kind of an interesting thing to learn, but you won't learn it unless you see the data, right? Unless the data has repetitively shown that that's true. And you've shown that that's not only consistent, but at a large enough amount that it's not a spurious observation, right? That you're not seeing a pattern, but it's a pattern that happens so rarely that it could just as well disappear. And so it's not just what predictive modeling is. What I actually tell people is you should be able to distinguish different types of predictive modeling by value. And the five things I delineated will immediately tell you if what you're being told is a predictive model is actually valuable. What most people do, however, is when somebody tells them something is predictive, they fall back on what they know anecdotally to determine if that's true or not. And that's the and therein lies the danger, right? And the reason that that's dangerous is because nobody in this world has ever gone ahead and created a new outcome doing the same shit over and over and over again. Nobody. Right? Nobody. So if you want to write new premium, you need new information. What you want to do though is make sure that that new information does actually index to the outcome because the outcome is actually consistent, right? That rarely changes.
DavidSo there you go. That's where we're at on predictive modeling. But it's it is indicative of the fact that you really need what you guys have, which is a very deep, very wide data set, because then you can actually determine if something is predictive if you see enough repetitions of the same thing.
SPEAKER_02Yeah. And I think that the other thing that you see when you have a deep, wider data set is you're able to index meaningful amounts of information even into small amounts of data. The thing that I've always been amazed by is that there's a meaningful amount of underwriting authority in our business handed to third parties who many times do not retain information to be informed. And it goes to show you the mentality of the business, which is I don't actively market to a broker or agent. You got to show up to me. Hello, I'm glad you showed up to me. Here's all that I can do. I don't actually have a maximum amount of information. So I'm gonna fall back on somebody in this company going, we have a bad experience in California, or that one time I wrote business in Missouri, that didn't go well. Or my sister lives in Wisconsin and she says stuff there is bad. Those are not predictive elements. That's not predictive modeling. That's just anecdotal bias. And it is the cornerstone of the business today. So people shouldn't be afraid of predictive modeling, they should embrace it, right? If it falls within the five components that I went ahead and delineated, it's it's well worth leveraging. I think that the thing that's really fascinating to me and will really transform, I believe, this business is as we apply these, it's not like I said, it's not just that the data elements themselves are rich. And because of their richness, even in small outcome sets, lift can be validated or shown. It's that that lift that is shown and validated can actually be transformed into the business because that lift is consistent, right? And so it's not just the richness of our data, it's that the way we model the outcome fits the way the market does business. And in fact, that's actually been the hardest component of our business. It's not getting the data, and it's not even the machine learning, and it's not the AI-assisted modeling, none of that. What's really been the hardest part is now we've got information distilled to an outcome. How do you, the customer, use that outcome credibly? And that's when predictive modeling becomes fact, right?
DavidWell, and the hardest thing in the universe, as you know, is to change a human being. Oh, yeah. To change the way they go about doing something. And that's a great place to end our conversation for today. Harry Sundrum, CEO and founder at Varakai. Harry, we we certainly hope you'll come back and chat with us again.
SPEAKER_02If it's as fun as this, you better believe it, man. This is great.
DavidIt's always fun on Chip Shapers. That's what we do. We do fun.
SPEAKER_02We do fun. We do fun. This is Wacky Friday. This is wacky Friday. It is indeed. Thank you for the time. Thank you for the time. Greatly appreciate it.
DavidThank you.
AnnouncerThe Chip Shapers podcast is a production of Chip Shaper Strategy and may not be reproduced or quoted in whole or in part without our express written permission. Copyright 2020. All rights reserved.