Latest / CareTalk: Healthcare. Unfiltered. / How AI Is Automating Provider Credentialing w/ Rahul Shivkumar, Co-Founder & President, Assured

Transcript

How AI Is Automating Provider Credentialing w/ Rahul Shivkumar, Co-Founder & President, Assured · CareTalk: Healthcare. Unfiltered.

• 0:00
  Before a provider can see a patient, someone has to verify their licensing, training, and work history, and then enroll them with every health plan that they're gonna bill.
• 0:10
  It can be a months-long process, and it depends on yesterday's tools like spreadsheets and faxes.
• 0:16
  There just has to be a better way.
• 0:25
  Welcome to CareTalk Executive Series, where we host leaders of innovative healthcare organizations.
• 0:31
  I'm David Williams, president of Health Business Group and host of CareTalk.
• 0:35
  Today's guest, Rahul Shivkumar, is co-founder and president of Assured Health, which uses AI to automate provider credentialing, payer enrollment, and network monitoring.
• 0:46
  Assured manages more than 30,000 active payer enrollments across all 50 states for digital health companies, health systems, and provider groups.
• 0:55
  Rahul, welcome to CareTalk.
• 0:57
  Thanks for having me Listen, I love the name Assured, but before you started that, you had, um, a role in co-founding Dawn Health, which I understand was a virtual sleep clinic, which you scaled across 15 states.
• 1:11
  So I'm curious about, what did this whole process look like from that side of the, of the desk?
• 1:17
  Yeah.
• 1:17
  It's a great question.
• 1:18
  Um, one of the biggest problems we had at Dawn was activating our providers.
• 1:24
  Um, when we started in one state with one provider, uh, the rules were pretty easy, but as we kind of scaled that organization across states to work with many different payers, uh, the rule set got very complicated.
• 1:38
  Um, I think one of my primary priorities was figuring out start dates, right?
• 1:43
  Once an offer letter was given out, once a provider was interviewed, and once we had confirmed that there was a mutual fit, that they wanted to work with Assured, it was kind of like a, like a race to figure out, hey, how soon can they start?
• 1:56
  And the big blocker there was how soon would they get paneled with insurance where we could actually get paid out for claims.
• 2:02
  Got it.
• 2:02
  So you've been talking lately about, from a COO or CFO standpoint, that they should track something called, like, provider activation time, which I think refers to this time between when you decided, "Okay, we're ready for you to go," until they're credentialed, they're enrolled, you know, and they're, and they're billable.
• 2:18
  So what are the things that are actually happening during that window, and how do you measure the financial impact that that must have on an organization that's trying to bring them in?
• 2:27
  There's a lot of things that hap- happen under the hood.
• 2:30
  Um, so maybe we could walk through an example of, uh, a provider that wants to see a patient that has an Aetna insurance, uh, a- as an example.
• 2:40
  Um, what, what happens on, like, the group side, or in our case, uh, like Dawn Health's side, was we would submit that provider under our roster, um, telling the health plan, "Hey, you know, this provider has recently joined us. Um, this is where, you know, they studied. This is their license." We'd give them a set of information.
• 2:59
  And what we would wait in return was them to give us an effective date.
• 3:03
  "So this is the date from which you can actually see Aetna members and, and get paid out for it." Now, that process can take anywhere from 30 days all the way up to in, in some cases like a year.
• 3:15
  And, and the big blocker there is the health plan is doing their own checks on the provider, right?
• 3:20
  They wanna make sure that a provider that joins their network is who they say they are and is not gonna be a liability for their patients.
• 3:27
  Um, that check is called credentialing, where they're verifying their credentials.
• 3:32
  Once they've confirmed that the provider is valid and, um, there's a requirement for them to have, uh, a provider of that type in their network, um, what they do then is roster the provider.
• 3:42
  So they upload them to their internal directories, they upload them to their claims system, and only after that process is complete do they actually give you an effective date and, um, you can actually get paid out for your claims.
• 3:54
  So there's a lot of things going on under the hood there.
• 3:57
  And if you don't time the start date of the provider accurately, you're missing out on quite a bit of revenue, right?
• 4:04
  Um, if a provider, say, could bill $1,000 a day and this process is taking 60 days, like basic math is kind of 60 times 1,000 is about $60,000 of, of, um, uh, top-line revenue that you're leaving on the table because of this processing time.
• 4:22
  Okay, so it's obviously significant.
• 4:24
  Now, I mentioned in the introduction that, you know, we gotta get all these different payers.
• 4:28
  Aetna is just one payer and, uh, tell me if I'm going too, you know, into the weeds here too much, but, you know, there's a lot of different payers and presumably they're not all marching in lockstep.
• 4:38
  So what is the typical situation if I've got a provider, you know, coming on and I've go- I've got Aetna, I have Cigna, but I don't have United and I don't have Blue Cross of, you know, Minnesota or whatever.
• 4:49
  Do you wait, does an organization wait to bring them on or do they then bring them on and they can't see certain patients or they don't get, they don't get reimbursed from certain plans?
• 4:56
  I mean, what happens?
• 4:58
  That, that's, that's really what happens, right?
• 5:00
  I think in, in states where it depends on your care model and the type of patients that you're seeing, but for us, a majority of our patients were, you know, commercial insurance, um, less so on the managed Medicaid or Medicare Advantage side.
• 5:15
  Um, and what ended up happening there was we needed, you know, a certain level of like density within a state in order to see patients, right?
• 5:23
  So if we got a network with, say, a really small plan that was only 1% of covered lives in a state like that just didn't add a ton of value.
• 5:32
  Like we could open up time slots for our provider, but 99% of the patients that would come in wouldn't be able to see that provider.
• 5:39
  Um, I think you can also tie it to like your wait list or things like that.
• 5:43
  If you, if you have a wait list of patients and you've collected their insurance, like maybe there's some correlation there of like, hey, 10, 10 of my patients on the wait list have Aetna, so getting activated with Aetna means this provider can actually start seeing 10 patients right off the bat.
• 5:57
  But it's so dependent on the kind of, uh, type of patients that you wanna see and how much coverage each insurance plan has in that state.
• 6:06
  Now you've done some benchmarking work, I think mostly through surveys, and you determined that the median enrollment is, is close to, uh, 10 weeks, I guess 69 days, and also that there's a few specialties that are the big drivers.
• 6:20
  Um, is there any rhyme or reason, you know, behind the certain specialties being big drivers?
• 6:26
  Totally.
• 6:26
  I think, um, it, it, it's kind of in-- it falls in two buckets.
• 6:29
  I think the first one is like network adequacy for a lot of these payers, where they are looking for certain types of providers, and if there's a big pull on their side, um, for that provider type, they might prioritize it internally and get that provider credentialed quicker.
• 6:45
  Uh, the second piece is also, uh, the, the credentialing checks that actually go in for each provider varies a lot by provider type.
• 6:53
  Um, as an example, if you are a, uh, psychiatrist, you might have a DEA, you might have a controlled, uh, drug substance license, and all of those things have to be validated, versus if you were, say, a dietician where, you know, um, there's only licensure in a few states that have to be like validated and there's no DEAs or CDSs, like that process can go faster Got it.
• 7:16
  All right.
• 7:16
  So there's some way to say, "I don't need one more internist in my network," and so they may- Yes … you know, just take their time on that as they get whoever it is a-ahead of that, um, in the queue.
• 7:26
  Is that right?
• 7:27
  Totally.
• 7:28
  Totally.
• 7:28
  And I, and I think it also comes down to, um, uh, specific like zip codes even, right?
• 7:33
  Like- Okay … oh, this area we don't have a cardiologist where folks have to see them in person, we will speed them through credentialing because we need them in our network very quickly.
• 7:42
  Got it.
• 7:43
  Okay.
• 7:43
  So I mean, that's the sort of a sign of a, an unhealthy system where rather than just having something that everything flows through, you know, very quickly with a small cycle time and not a big, uh, queue, they have to prioritize because everyone is sort of in the, you know, just backing up otherwise.
• 7:58
  So let's talk about what's happening with credentialing teams.
• 8:01
  You know, I-- this is always the easy question in healthcare, you know, are things getting better or worse?
• 8:05
  So I'm just gonna skip the drama and say it's probably getting worse.
• 8:07
  I mean, I'm thinking about stuff like, you know, there's Medicaid redetermination, and then there's the one big beautiful bill act.
• 8:16
  They're all kind of reshaping the, the payer mix, and then you've got providers I see especially on the telehealth side that are going into a lot of different states, kinda like what you're doing with the sleep clinic, maybe even, even more so.
• 8:27
  Am I putting my finger on the, some of the, the drivers?
• 8:29
  And well, correct me if I'm wrong, is it getting better or worse for the credentialing side of things?
• 8:34
  Yeah.
• 8:34
  I think there has been-- there's been a lot of talk of centralizing credentialing, of making credentialing more streamlined across multiple health plans.
• 8:44
  But the reality is, uh, the fragmentation has, has never been, uh, m-more obvious, right?
• 8:51
  Like you mentioned.
• 8:51
  Um, every single health plan has its own unique requirements.
• 8:55
  Those break through for specialties.
• 8:57
  They even break through per state.
• 8:58
  Like what a health plan sometimes requires in one state, if it's a national payer, i-i-is different in another, and the application process looks slightly different.
• 9:06
  Um, so, so the, the, the high-level answer is, uh, I, I think there has been a lot of talk of making things easier.
• 9:14
  I don't think that's what's being seen in, in the market today.
• 9:18
  So I understand that a not-- a lot of your customers are some of these fast-growing behavioral health or telehealth companies that are expanding across multiple states, like what your experience was with, with Dawn, but maybe even more.
• 9:30
  Where, when they're, you know, whenever you're trying to scale up, you're gonna hit different sort of bottlenecks and different sort of strains.
• 9:36
  Where does this issue of credentialing and the other, uh, you know, related areas that you handle, where does that fit in?
• 9:43
  Does that tend to be the constraint, or is the constraint something else that they're dealing with?
• 9:48
  It's often the constraint, right?
• 9:49
  Because you have to get a network in order to get paid out, and a lot of the customers we work with, like their business model is so reliant on billing insurance, right?
• 9:59
  Like 90% plus are insurance pay versus like cash pay, where I think, um, uh, it, it's, it's a lot easier going state to state.
• 10:08
  Now, you are limited by state licensure, right?
• 10:11
  But I think there's been a pretty good push on that side with, uh, uh, compact licenses that, that kind of like got pushed out during COVID, where it's much easier to practice across multiple states.
• 10:22
  But, but on the payer side, I think, uh, what throws a lot of groups off when they go from one state to another is, um, even if they're working with the same national payer in another state, the processing time suddenly drastically changes, right?
• 10:37
  You've built your business equation on, I'm going to give an offer letter, this provider's gonna start 60 days from now.
• 10:44
  That's when they're gonna be billable.
• 10:45
  You go to a new state, and the same payers all of a sudden take 90 days, 120 days.
• 10:49
  And if you're not careful about that, and if you expand too quickly, say you gave an offer letter out to 50 providers and they, and, and they start in a rapid kind of push to expand, um, and you haven't gotten a network, that can be very dangerous for a practice.
• 11:05
  I wanna talk about what you're doing that may be different from other things I'm familiar with.
• 11:09
  So there has been, uh, software, like this, this is a recognized issue.
• 11:14
  I've been aware of it for, I think, 20 years or so.
• 11:17
  And so the… I think for at least the last 10 years, there's been some credentialing software, but somehow the timelines haven't really improved.
• 11:25
  And I think you've been talking about a distinction between software that tracks what's going on, which now I can officially see it in a nice chart that it's going slowly, versus actually, you know, doing some of the work.
• 11:37
  And e- explain that a little bit and if that's in fact what you're seeing.
• 11:42
  Yeah.
• 11:43
  Uh, I think what we saw at Dawn, right, as an example, was we used software, but the underlying work, collecting the information from a provider, making sure that information is accurate, sending it over to the health plan, following up with the health plan, all of that was done by people, right?
• 11:59
  There was a nice software where I could open it and see where along the process a provider was, but every single one of the inputs or stages that went into that was done by a person.
• 12:10
  This could be an external outsourced vendor, it could be a, uh, old school credentialing company, it could be in-house like we did at Dawn.
• 12:18
  Um, regardless, the function of scaling was tied to labor, right?
• 12:24
  I think what we're seeing-- what, what we saw at, at, at Dawn was that there were really two kind of constraints.
• 12:29
  One, um, as you scale your provider network, you had to constantly hire operations, right?
• 12:34
  'Cause the work kept kind of piling up, and you needed more people to do it.
• 12:38
  Um, and then the second piece was, uh, that labor didn't necessarily… Like, the quality of it didn't necessarily scale as we got a lot bigger.
• 12:47
  Um, as in new people would make mistakes.
• 12:49
  Uh, any mistake here is very punishing, where if a submission was done wrong, the feedback loop was four weeks, six weeks, where it would get kicked back, and then you're six weeks out from actually being able to see a patient.
• 13:00
  So one of the things we did after, after Dawn got acquired was, um, we were looking at a bunch of problems, like particularly problems we had faced before that we had a good understanding of.
• 13:11
  Yeah, and, and this one came up, right?
• 13:13
  It was like, hey look, every single time we had to expand, every single time we had to hire a provider, um, this was like a ticking clock that you had to solve for, and, um, you constantly had to kind of, uh, try to hire labor to, to, to do work that honestly isn't a lot of fun, right?
• 13:28
  Um, it's like copy-pasting the same fields, putting it over, following up.
• 13:32
  Um, and then kind of, uh, GPT-3 happened, and we were like, "Wait, maybe now there's the possibility of building like reason-- like technology that can reason and actually can, you know, navigate a website, understand what the website is asking for, fill in that information." So it's not just a piece of tracking software, it's software that can also validate this data by navigating somewhere, verifying that the data is accurate, then actually compile that data together, submit a form, follow up via email.
• 14:04
  Um, and, and that's really what we're building here at Assured.
• 14:06
  It's like these functions where I think in the past people were just doing a ton of manual work.
• 14:11
  Let's have like the agents do it because, one, these problems are not extremely creative or open-ended.
• 14:16
  They're reasonably closed box.
• 14:18
  Um, and the second is, uh, an agent can repetitively do something and not make a mistake versus a, versus a human being So if agents can take on a lot of this work, then what role, if any, uh, should humans still have in these operations?
• 14:35
  And, and assuming there are some things for a human to do, how do you decide which ones?
• 14:41
  Yes.
• 14:41
  Uh, I, I think there's two categories we put, uh, kind of the work that humans have to do in.
• 14:48
  The first is, um, anything regulatory, right?
• 14:51
  Um, as an example, these credentialing checks that run, um, the governing bodies like the NCQA require that a human sign off on any work that even an automation does, right?
• 15:02
  So when we're doing that as a CVO or when our customers are using our credentialing platform, they always have a human in the loop signing off on these credentialing, like, background files.
• 15:12
  Um, and the final decision-making, even if the agent is able to detect someone was in a-- on a, on an office of inspector general list or, you know, they were found on some other sanction list, the decision-making of should we hire them, should we not hire them, should we terminate them, will always be in the hands of human beings, right?
• 15:30
  So where we have kind of used the agents is find the facts, but the final decision is always kind of made by a human being.
• 15:38
  And I think the second place where humans are extremely valuable are, uh, building human-to-human relationships, right?
• 15:45
  Uh, an AI can fill a form, it can submit it, it can follow up over email.
• 15:50
  But at the end of the day, like, building a relationship with a payer, right, like getting a deeper relationship with them where maybe, uh, you're negotiating contract rates, you're maybe getting delegated credentialing, you're, you're working towards these milestones, like, those touch points are all significantly better handled by a human being.
• 16:08
  And, and what we see from a lot of the organizations we work with is actually they don't, like, lay off people.
• 16:15
  They, like, reassign them.
• 16:16
  They're like, "Hey, we actually have a backlog of things where people can talk to other people and solve them, and now they finally have the time to do it." So there've been a lot of advancements in AI since the ChatGPT-3 that you were mentioning.
• 16:29
  So how many of these things that you're talking about, like when you say a human will always be in the loop, is that always for the next three months or, you know, the next two versions of, of Claude?
• 16:39
  Or what does that mean?
• 16:41
  What's gonna change?
• 16:42
  It's, it's a great question.
• 16:43
  I, I, I think, um, some of it comes down to, like, regulation, right?
• 16:47
  If the NCQA comes out and says, "Hey, if you can show a sample set of data that is, that, that, that goes through without human intervention that's very accurate, and you pass a bar and we don't need a human being in the loop for signing off on a file." I think still the decision hiring yes or no should, should be a human being that decides it.
• 17:05
  Um, but there are portions of that that I think are regulatory that could change.
• 17:09
  Um, whenever we deploy our agents, we always tell our customers to treat it like a new employee, right?
• 17:15
  The first few times, like- Watch over its shoulder, right?
• 17:20
  Like as they're doing the work, like make sure that, that it's doing it correctly.
• 17:23
  And then once you feel comfortable with, with, with, with your new employee, you're, you're gonna delegate the work to them and maybe quality check them, uh, do a performance review, things like that every once in a while.
• 17:33
  I would think of the agents as very similar.
• 17:34
  Today in our enrollment workflows, we have agents that run fully autonomously, and then their quality is spot checked, you know, every month, every quarter, depending on our customer's, like, policy.
• 17:44
  But, um, uh, I think it's less on the models getting better because some of these functions can be fully automated, and it's more on, uh, uh, re-regulatory stuff.
• 17:58
  Yeah.
• 17:59
  So of course, when people hear about AI agents these days, you know, in the past people said, "I don't wanna talk to that. Let me talk to a human." Now there's these stories about, you know, the agents going rogue or, you know, if you want to assign, uh, agency to the, to the agents and say, well, they're in charge of it as opposed to whoever, you know, set them loose.
• 18:15
  So let me just not push back, but, uh, just discuss a little bit more of this, this idea of, okay, there's treat the agent like a new employee, kind of supervise them.
• 18:25
  Well, the new employee, you know, they eventually learn the rules of the road, and if they're not good, you fire them.
• 18:30
  W-whereas some of these agents are learning how not to be supervised or they're doing something different when they're not, not watching.
• 18:35
  How-- W-whether or not these are an issue on the, you know, the credentialing side, how do you, how do you deal with just getting people to become comfortable with that?
• 18:44
  And als- also partly it's just things are, are changing.
• 18:47
  So how, how do you kind of keep up with that?
• 18:50
  Totally.
• 18:50
  So we, uh- Do a lot of things around access control and harnesses.
• 18:58
  So we scope down the agents to only ever have access to, you know, the particular things it needs to do a function, right?
• 19:06
  I think when you start increasing that, it starts to get, uh, like, like the risks obviously go up, right?
• 19:11
  As an example, on the credentialing side, like, uh, there's no need for us to ever, like, for those workflows integrate with an EHR, right?
• 19:19
  So we don't do that, like we're closed off from it.
• 19:21
  We don't touch any of the patient data.
• 19:22
  None of that ever flows into the system.
• 19:24
  And even a particular agent is scoped, uh, i-in its work to just maybe touch some data fields for a provider, only have the ability to use a browser to go to maybe like one website, right?
• 19:34
  So it kinda like narrows down the scope of what they can do.
• 19:37
  Um, the second thing we do is like we're constantly like, uh, testing like evals, right?
• 19:43
  So we're testing like, "Hey, um, what-- which of these reasoning models are really good at doing what type of function?" And maybe the super high like reasoning models that hallucinate a lot more, you don't, you don't really have to use them for, for filling a form, right?
• 20:00
  Like that could be done by like a significantly like dumber model that, that really doesn't hallucinate.
• 20:04
  It really just does a task.
• 20:06
  So we have like this pretty like big like kind of like, uh, benchmarking exercise we do every time a new model comes out, being like, "Hey, one, let's test it on our like sub-use cases, make sure that accuracy and things like that stay high." But also, um, if there's a model that maybe isn't as sophisticated but is really good at filling out like a form on a website, like maybe we use that just for, for those workflows, right?
• 20:28
  It doesn't have to be the smartest model in order to, uh, get the job done.
• 20:33
  Yeah.
• 20:33
  That, that makes good sense.
• 20:34
  And maybe just to emphasize a couple of those points.
• 20:37
  I think a lot of these kinda, you know, rogue agent breakouts and all that, the, the models are very sophisticated and on the frontier, but the, the level of controls and cybersecurity, they're just not doing the basic things that you would do.
• 20:49
  It'd be like putting an animal, you know, out in the, in the barnyard and not closing the door on the- you know, the fence and doing the, doing the basic stuff.
• 20:55
  So I think that's part of it.
• 20:56
  And then you're making another good point.
• 20:58
  You know, people are using, uh, you know, the frontier model to proofread their email and tell them whether a capitalization is needed or not.
• 21:04
  And not only is that just a waste of tokens, it's also just the same reason you're not gonna hire like a theoretical physicist to come and fill out a form.
• 21:13
  Uh, you know, you don't want s- you don't want someone that's got a lot of ambitions to go, to go beyond that.
• 21:17
  So you may not need to really innovate in terms of the capabilities of the form fillers to get into, like onto some frontier, if I'm, if I'm understanding right.
• 21:28
  Absolutely.
• 21:28
  I think you, you explained that way better than I did.
• 21:30
  I'm gonna, I'm gonna actually steal that physicist bit.
• 21:33
  It was really good.
• 21:34
  Sound, sounds good.
• 21:35
  So, you know, I'm sure some of the compliance teams are starting to raise questions about, you know, AI credentialing, and I know you mentioned NCQA before, but you have things like NCQA certification, SOC 2 as well.
• 21:49
  How do those things tie into kind of getting the market both excited to deal with this big problem and also not to say,"Well, I don't know AI, I'm scared of it"?
• 21:58
  Yeah.
• 21:58
  Um, I think we, we play within the guidelines, right?
• 22:01
  Like, I think, um- Healthcare, for maybe the right reasons, and again, we could maybe spend ano- an entire hour talking about this, but for the right reasons is, uh, like risk-averse when it comes to things like this because again, the, the downside of, of skirting compliance is really bad, right?
• 22:21
  Um, so what we do is like we, we think of like all the super manual work that a person is doing that again, at the end of the day needs like a human being to, to, to… Like a human is making the decision on it.
• 22:34
  Like how do we replace those things with AI?
• 22:36
  And again, we are NCQA certified, so our CVO and our, and our platform and the files that have been generated have obviously been through that.
• 22:42
  So, um, we know, we know, we know how difficult those audits are to, to pass and, and the kind of quality bar required there.
• 22:50
  I think what we're, we're seeing across the board from, from a lot of these organizations is, um, they are spending more and more time looking at things like role access, right?
• 23:00
  Like I think that's becoming super, super, super important because, uh, the amount of damage a system can do is, is only tied to like how much data it gets access to or how much, uh, read/write capability it has, right?
• 23:14
  So one of the big things in the NCQA audit, they look a lot at like, at like role control, like audit logs, right?
• 23:19
  Like who did what?
• 23:20
  Like after something was done, was it altered again, right?
• 23:24
  Like, like having like very strong I think like audit logs and like access control is something a lot of them are pushing for, and I think it's actually good.
• 23:31
  It's like going in the right direction.
• 23:33
  It makes sense.
• 23:34
  You know, this, uh, this, this question of, you know, role-based access control is a big one in healthcare and just in, in general.
• 23:40
  Um, it's easy to go onto a system and just assign everybody to be like a super admin- Uh-huh … which have all sorts of things that they don't need.
• 23:47
  So the kind of principle of, you know, of, of minimum required access is what you wanna give an individual and an agent, I think are, you know, are very real and then they, they, you know, stay in the realm, uh, of where they, of where they belong and don't have to go out and set new-- Like if you don't want an agent to go and say, "Well, I can't get past this, so I'm gonna go and like be the administrator and change the rule." Totally.
• 24:09
  Totally.
• 24:10
  And I think you also wanna log everything the agent does, right?
• 24:13
  We are, uh, like we almost log too much.
• 24:16
  It's like, oh, this is why I decided this.
• 24:18
  This is what it… 'Cause if, if there's ever an incident where it goes off script, you need to be able to go back and diagnose exactly where it went off script.
• 24:26
  How can you prevent it from doing that again in the future?
• 24:28
  And that transparency is something our customers want, right?
• 24:31
  Like if a, if an employee did something wrong, they're gonna wanna know the root cause of why it went wrong.
• 24:35
  Uh, it's no different.
• 24:36
  Yeah.
• 24:37
  It, it leads a little bit into, I think, also a question about incentives and how things work in healthcare, and, and certainly this is an issue with thir-- I'm not talking about credentialing, but with like third-party payment.
• 24:49
  If I have an AI that can say, "Hey, a person came into the ER for a particular reason," but then, you know, it was noted that they were complaining a little bit about their breathing, well, that, that may actually be a higher code, and the AI may pick up on that in a way that the, you know, somebody wouldn't have otherwise.
• 25:05
  And so, and then you've got like the AI to fight back against that.
• 25:09
  So let's talk about how that may relate in terms of having new providers being credentialed.
• 25:14
  I know here in Massachusetts at one point there was actually a limit on how many people Partners Healthcare could add to its primary care network because they got higher reimbursement, and just having them move over to that network was just gonna raise costs overall.
• 25:29
  Are there incentives around this that if once you try to speed things up, there may be some sort of a counter push?
• 25:36
  Yeah.
• 25:36
  It's, it's a really good question.
• 25:38
  I think on the, on the network sizing thing, we- we've definitely seen instances of that, right?
• 25:44
  During, um, COVID, there was a crazy need for telehealth, like behavioral health, like providers.
• 25:51
  There were a bunch of MSOs that came out, um, and maybe the, the reimbursement rates for a large organization like that, um, was higher than just an individual provider.
• 26:00
  So you had a bunch of providers kind of working off of that network, and a lot of those networks had very similar providers, right?
• 26:06
  If you really looked at them, like, maybe they had, uh, 30% delta, but 70% of them were like overlap between all of them.
• 26:13
  And now you're seeing the payers kind of correct, right?
• 26:15
  They're like, "Hey, we're-- we see this in our data, uh, we're gonna like compress your like reimbursement rates." Right?
• 26:21
  So the, the, the supply and demand dynamics are very interesting because if there's a sudden event that happens like COVID, there's like a crazy correction on one side, and then it starts to slowly correct on the other side, and there's a lot of people kind of like reacting to it.
• 26:36
  On, on the credentialing, like, uh, kind of like velocity piece, uh, there's a little bit of, uh, I think like, uh, incentives there for the payer to have control of that credentialing timeline, right?
• 26:49
  It's a way of controlling supply and demand on their side.
• 26:53
  Um, what we've generally found is, uh, the best organizations, um, tend to actually use their outcome data to get things like delegated credentialing with the payer, right?
• 27:05
  Where they're telling them, "Hey, we are so invaluable to your network, uh, not because we have this provider type, not because of this, but because the outcomes that your members are seeing are really, really, really good." Right?
• 27:17
  So, so, so there's some degree of like you almost wanna tie like the, the, the patient like benefits to it in order to get faster like credentialing timelines.
• 27:26
  Because the pa- like, like, I think your, your question was like the payers are not that incentivized sometimes to do it, and I, and I think that is true.
• 27:34
  Yeah.
• 27:36
  G- ex-- you mentioned a couple times this delegated credentialing.
• 27:39
  Does that mean that the provider organization can do more itself?
• 27:42
  W-what does that… Yeah.
• 27:43
  Correct.
• 27:44
  Yeah.
• 27:44
  So sort of… There's things like that in other industries, you know.
• 27:46
  So if you're getting your car repaired, you know, some shops are, are trusted by the insurers more than others, and they get less of a, less of a hassle and less of a, less of a runaround.
• 27:56
  That also encourages the, um, you know, the insurers to, to go there.
• 28:01
  Is there something similar with credentialing, is one question, and then how does that affect your business if there's delegated credentialing?
• 28:07
  Is that good or bad or indifferent for you?
• 28:10
  Yeah.
• 28:10
  Uh, the first question is yes, right?
• 28:12
  When you reach a certain scale, um, and a certain level of reputation with the payer, and again, this is why I think, uh, humans on the, on the provider group side should be spending a lot of time building relationships with the, with the payer side and not really doing a lot of the manual work, because, um, what the payers will do then is tell you, "Look, you guys handle credentialing yourself, and you just give me credentialed providers, and I will credential them again." Right?
• 28:38
  Like, it's delegated out to you.
• 28:40
  Um, and the benefit of that is you control the timeline then, right?
• 28:44
  And for the payer, the benefit is they don't have to do that work'cause it's a, it's a cost for them, and they're gonna audit you.
• 28:48
  They're gonna trust but verify, right?
• 28:50
  They're gonna check, uh, uh, every time and, and that can really, really speed up timelines with a lot of the payers.
• 28:56
  For us, um, it actually is beneficial for our business because as a NCQA-certified CVO, we are one of those trusted parties that can do credentialing at that high level.
• 29:08
  So if a group used us to generate those files, and again, we do it, uh, I think our, our average time is about, like, same day to two days now versus, like, you know, 45 to 60 for the payer.
• 29:18
  Like, you can actually get a network a lot faster, and we are handling all those audits, right?
• 29:23
  The payer's coming to us.
• 29:24
  They're checking our files.
• 29:25
  They're making sure things are operating at, at, at, at the standard it should be, and then the NCQA audits us every, every two years.
• 29:31
  Yeah.
• 29:32
  Got it.
• 29:32
  So let's look ahead a little bit.
• 29:34
  So if I understand correctly, you've got, you know, tens of thousands of active payer enrollments that you're managing across all 50 states.
• 29:42
  But you recently raised some money, I think, uh, close to$20 million in a Series A, which is, which is pretty robust.
• 29:48
  So where does that take you?
• 29:50
  I mean, both in terms of you're in all the states, um, but sort of beyond credentialing and r- and enrollment.
• 29:55
  I've heard about things like network monitoring and privileging.
• 29:59
  Give me, give me a sense, give us, uh, all a sense of where, where you're looking ahead.
• 30:03
  Yeah.
• 30:04
  We, we use like two things to really dictate, like what we're gonna build next.
• 30:08
  The first is, what are our existing customers asking for?
• 30:11
  You know, when we initially started with credentialing, it was like,"Hey, help us like get a, get a, get a network with payers.
• 30:17
  Help us with payer enrollment." So we, we, we built some of those workflows for them. And, and now like a big ask for a lot of our bigger customers, especially the ones that have an inpatient like presence, is privileging, right?
• 30:28
  Like, "Can you help us streamline some of the hospital privileges?
• 30:31
  Can you help us like streamline keeping those privileges up to date every two years when they come up for review?" Um, and then the second thing is like what we're seeing from our go-to-market team, right?
• 30:40
  Which is, hey, these are all like the opportunities that are coming to us that we're saying no to right now because we don't do them.
• 30:46
  So I think that the, the two big directions we've been spending a lot of product effort on are, one, uh, building some of these privileging workflows.
• 30:53
  We're actually gonna have those live Q1 of next year, so we're very excited for that.
• 30:57
  And then the second is, um, I think there's a lot of post, uh, credentialing work that goes into running a very, uh, uh, like a very efficient operation.
• 31:10
  Uh, I'll give you an example.
• 31:11
  One of them is if you're multi-location, if a provider ever moves from one location to another, if, if the payer's, uh, system isn't updated about that, um, your claims could get denied, right?
• 31:22
  So we're working on some level of like, uh, like using the existing agent infrastructure to build agents that can maybe even detect if something in the payer system is not yet up to date with what is happening in reality, and maybe protect like claims.
• 31:36
  Um, so that's a huge thing our customers have asked for because again, they're using Assured as like a source of truth.
• 31:41
  It's like can this source of truth also go reflect truth out in the rest of the world and keep everything else up to date?
• 31:47
  Yeah.
• 31:48
  It's a, it's a logical approach.
• 31:49
  If I think about it, you know, as you're first getting your foothold, you're doing that sort of initial transaction that's a big pain point, get the provider credentialed, get them to be able to, you know, come on board and actually start billing.
• 32:04
  But once all that work is being done, there's kind of-- There's the initial setup, and then there's the, the maintenance of that, and then there's the use of all the kind of data and information that you have for other aspects of, uh, optimizing and running the, the operation.
• 32:15
  So that makes, that makes a lot of sense as a growth path, uh, to me.
• 32:19
  All right.
• 32:19
  Well, thank you, Rahul Shivkumar, co-founder and president of Assured Health, for a look at one of healthcare's unglamorous but more consequential bottlenecks and what you're doing about it.
• 32:31
  That's it for this executive feature of Care Talk.
• 32:34
  I'm your host, David Williams.
• 32:35
  Thank you for listening, and thank you, Rahul.
• 32:38
  Thank you so much for having me.