Christie Callahan of Arrive Health joins host Ayush Jain to take apart one of healthcare’s most stubborn administrative problems: prior authorization for medications. An engineer who moved through manufacturing, Bain & Company and CVS Health’s Medicare Part D and appeals operations, she explains the three reasons plans put a prior auth in place (safety, off-label use and cost) and why the process still frustrates providers: in six months, Arrive Health saw 416,000 unique drug and plan combinations, and small procedural misses push patients into denial and appeal loops.
She then walks through arrive ACCELERATE’s three-part approach: help prescribers avoid the authorization through better real-time benefit options, use large language models to put the payer’s criteria in front of the provider before the question set starts, and resolve the authorization before the script ever reaches the pharmacy. On AI, she keeps PHI out of the models, uses validation loops with other LLMs to catch hallucinations, and sees AI as an assistant rather than a decision maker. Adoption, she argues, depends on being native to the EMR and building on prior authorization workflows inside the EHR rather than adding another portal.
Christie puts on her “plan hat” first. Plans add a medication prior authorization for safety (dosing ramps, or a sleep medication suddenly used every day), for off-label use that lacks validated outcomes (her current example is GLP-1 drugs prescribed for weight loss when the plan is paying for diabetes), and, less often, for cost, where a patient must show lower-cost therapies were tried first. From the plan’s side, the goals are better outcomes and lower premiums.
The trouble is scale and opacity. In six months Arrive Health saw 416,000 unique combinations of drug and plan, each in effect its own authorization. To providers it feels inconsistent, and the criteria are hard to read in advance. Her example: a lab result that is nine months old when this plan wants one from the last three months. The request is denied, the lab is redone, the appeal is approved, and nothing about the patient’s eligibility has changed. Payer rules buried in documents and portals are the same gap that eligibility and prior auth workflows in Epic run into.
arrive ACCELERATE works in three steps. First, help the prescriber avoid the authorization: out-of-the-box real-time benefit checks fall short because pharmacists do so much work to make claims pay, so Arrive built models and an intelligence layer that surface real alternatives at the moment of prescribing. When that is done well, she says, providers avoid 48% of prior authorizations. Kyle Kiser, Arrive Health’s CEO when he joined the show in 2024, described the groundwork for this step: real-time, patient-specific pricing inside e-prescribing workflows.
Second, show the requirements up front. Arrive uses a series of large language models, including Claude Sonnet, to extract the criteria from coverage requirement documents and summarize them before the provider starts the question set, so a provider can stop, send the patient for an updated lab and come back when approval is likely. Third, there should never be a denied claim at the pharmacy: the script is written, the authorization resolved and the script sent within hours. With two health system partners, she says, resolution now takes about four hours.
Christie separates the technology’s limits from organizational readiness to let AI act. Where it is proven is as an assistant, “a sophisticated augmented intelligence.” Arrive has so far kept PHI out of its models: the models read plan documents that are the same for every member, and the summary is then inserted into the normal, PHI-rich authorization flow.
On quality, Arrive runs three follow-up loops with other LLMs to validate each output and catch hallucinations, and she expects better prompt engineering and better loops to reduce the clinical review needed over time. The guardrails that keep AI output reliable are where she is most optimistic. What she does not expect soon is AI making healthcare decisions on its own: for now it is a human-assist tool, not an autonomous one.
On the real-time benefit side, Arrive runs about 22 million transactions a month across hundreds of thousands of providers, and the lesson Christie takes from it is that providers do not need more apps or portals. That makes for a long growth journey and less control over the interface, so Arrive keeps tight feedback loops with health system partners and fixes input data rather than adding screens. Holding the script until the authorization is resolved, with rules that protect the patient, created the urgency that drove adoption. She also builds on standards where they exist: imperfect, but a starting point that cuts custom work.
That is the same principle Mindbowser works from. For prior authorization, we build the provider-side flow: CRD, DTR and PAS clients against payer FHIR APIs, decisions written back into the EHR, and X12 or portal fallbacks where a payer isn’t ready.
Christie sees her job in two parts: help people perform at their best (the right roles, the right tools, cohesive cross-functional teams), and act as a thought partner who gets into the weeds to ask questions and bring an outside view. Underneath both is a habit of sharing context widely, because, as she puts it, information is the only way people up and down the organization make the right decisions.
[The first 15 seconds are music.]
Hi everyone, welcome to another episode of the HealthTech with Purpose podcast, where we deep dive into the world of healthcare technology and the people building innovation. I’m your host Ayush, and today we have a truly exciting guest: Christie Callahan, CEO of Arrive Health. Christie is a health tech leader with expertise in strategy, innovation and operations, and she has spent over a decade tackling one of healthcare’s biggest pain points, which is prior authorization. She’s here to help us unpack the challenges of prior authorization, the role of AI and automation, and what the future holds for patient access and transparency. Plus, we’ll get to know a little bit about her journey and also her leadership philosophy. So let’s get started. Christie, welcome to the show.
Thanks for having me. I appreciate it.
So Christie, we always start with understanding a bit of your own background and your journey into healthcare. How did you get into healthcare? Did you always envision being here, or how did your career start? Please tell us more.
Okay, sounds good. The short answer is no. I didn’t think I would end up in healthcare. In school I was an engineer, and I was actually working in manufacturing. I was working on the Puffs tissue line in Green Bay, Wisconsin. But as a young twenty-something I wanted more variety, and I had been interested in consulting throughout college. So I ended up joining Bain & Company and consulted with them for a while, but that wasn’t a good fit either.
And so I got recruited out of Bain by CVS. At the time they were building almost a SWAT team, Ayush, to come in alongside the operations and help solve problems when there was an issue. When I joined, the SilverScript plan, which is a Medicare Part D plan, was in sanctions with CMS (the Centers for Medicare & Medicaid Services), and we needed to come in, fix the tech, fix the processes and help the operations team get back on track while they were underwater processing enrollments and premium billing and all that kind of stuff. From there I jumped into coverage determinations and appeals, which is the prior authorization version of Medicare Part D, for a similar thing: we were being audited by CMS, we had client audits, and the team really needed support in terms of automating their workflows and using better integrated technology to improve their pace, reduce the load and support the team.
Was it pre-COVID, or which years are we talking about?
Pre-COVID. Yes, far before COVID. I was with CVS for a while, and in my time I got to see the specialty and mail-order pharmacies, the call center and the prior auth operation, obviously. Along the way I met Arrive Health, and what I realized in working for the PBM (pharmacy benefit manager) was that for the problems I was trying to solve for patients, I could make it better from a time-to-resolution perspective within the PBM, but I couldn’t actually fix the root cause, which was providers having better information and a better workflow to clear that path for patients. So when I met Arrive, it was a really good fit with the way I wanted to tackle this problem from a different perspective. And at this point we were mid-COVID, we were all remote, and there was an opportunity to join Arrive and not have to move to Denver. So it was a lot of good things coming together at the same time.
Right. Great. So you found your calling.
Yes, exactly. And what’s been fun is truly getting closer to that decision point, closer to providers and patients. Why did I stay in healthcare? The problems are big, but they do feel solvable, and that’s what gets me out of bed in the morning. It’s super complicated, there are a lot of competing priorities, but it’s not unsolvable. So it’s fun to be tackling this problem from a different perspective: how can we share information better, how can we use technology to make it simpler, and how can we get patients the medications they need?
I like the way you look at it: problems are big, but solvable. There’s that silver lining that exists in health, which I think keeps all of us moving and working towards one solution at a time.
That’s right. And I tell the team, all you need is a glimmer. There’s a glimmer of a solution out there. We can make it happen.
Right, right, right. And then you are in a boardroom selling to a CXO, and they have their own ideas.
That’s true. That’s true.
So, great. Christie, let’s talk prior authorization, because that has been your deep expertise. Why don’t you start by giving us a breakdown of why prior authorization exists and why it’s such a persistent challenge?
Big question. Let’s start with why it exists. Specifically around medications, which is what we’re focused on, there are usually three reasons a prior auth gets put in place.
One is safety. For some reason there’s something about this medication where there needs to be safety monitoring, whether that’s dosing related, maybe you ramp up on the med, or it could be addiction related. I think about sleep medications: when the plan sees that they’re being used every day, they put a prior authorization on basically to alert the doctor and say, “Hey, it looks like this is being used every day. Are you sure you want to continue?”
The second is off-label use, and I think this is the one that probably has the most swirl around it. These are drugs that are really popular for off-label use and a lot of times have good results, but just aren’t validated and don’t have the research at this point. The plan typically doesn’t want to pay for that off-label use. They’re saying these aren’t proven outcomes, and they don’t know this is going to be beneficial to your overall health. So from that perspective, unless you can prove a few things, they’re not going to pay for the medication for off-label use. A great example of this right now is our GLP-1s, Ozempic and Wegovy. People want to use them for weight loss, fine, but that’s not the reason the plan is paying for it. They’re paying for diabetes.
And the third is cost. That one is infrequently used, because plans use plan design in other ways to manage cost, but sometimes there is a prior auth driven by cost. Typically it’s saying you have to meet some threshold: you have to have tried other meds before you step up to this more expensive therapy, you have to show you’ve tried the lower-cost options and they’re not working, and therefore you really do need this more expensive med.
If I put on my plan hat, they would say two things. One, they’re trying to make sure patients are getting the best health outcomes, and that they’re helping providers who are overwhelmed by the number of patients and diagnoses they see. And two, they’re trying to keep premiums low for everybody, which means spending less out of pocket. Obviously there are many different sides to that, but that’s my plan hat telling you why prior authorization exists. Why is it a challenge was the second part of your question.
Yes. Because in theory it’s a great thing: it brings predictability for everyone. But in fact, the numbers show that up to 82% of appeals get overturned. So why has it been a challenge? What are those inefficiencies? Let’s touch on that.
Yeah. I think the best way to highlight why this is a challenge is to use our data. In six months we saw 416,000 unique combinations of drug and plan. That’s insane, right? Think about the complexity for a single provider. Obviously a single provider isn’t prescribing every drug, but the point is you have 416,000 combinations, essentially 416,000 unique prior authorizations. How can you keep that straight? You can’t.
I think two things cause the challenge. One is that it doesn’t feel predictable to providers and patients. For them it feels inconsistent and hard to understand. Sometimes there’s a prior auth, sometimes there’s not. Why? Why not? The second is that the criteria are varied enough that they often don’t understand what’s expected or what’s being asked for.
A good example is a lab result. You’re going through your authorization, you need to have a certain lab, and you put in the results of the lab and the date it was taken. Well, maybe for this plan that lab needs to have been done in the last three months, and your lab was done in the last nine months. Clearly you don’t meet the criteria, because of timing, because you didn’t know you needed a fresh lab in order to submit and get access to your medication. So in that case, to your point, you get denied. Then you understand, oh, this lab needed to be done in the last three months. You go back, you get your lab done, you submit for appeal, and of course you get approved, because your labs are showing the same thing. You’re still within the eligible criteria; you just needed to check this procedural box. And I think that’s where the frustration lies: I think I’ve done everything right, but there’s something small in the administration that causes this denial and appeal loop, and it is burdensome and time intensive.
So basically it’s adding another loop to an already hundred-times-complex system, another loop that people are required to go through. And not surprisingly, there’s in fact a demand to get rid of it altogether. Of course, you are building solutions and helping prior authorization really be more useful, but what is your take on neutral ground? Is it really helping? Is it too complicated? Or does it require a complete overhaul? Where do you see it currently?
Good question. I think it’s a mix. Like any complicated problem, there are a lot of things that probably should change. There are some instances where authorizations persist that probably don’t apply anymore, that should have been taken away a long time ago because now we know more about the medication. I do think at an individual level people are trying to do the right thing. The P&T (pharmacy and therapeutics) committees within plans are really thoughtful about what they’re trying to do. That being said, when you have these massive plans, things slip through the cracks. So on the plan side, I think there should be some simplification, and there’s a reason to take a close look at what has an auth and why.
What I’m excited about is that providers have kind of been takers. They just have to take what’s given: this is the question set, and I’ve just got to get it done. What I’m really hopeful about is that our technology is finally in a place where we can flip that a bit and make it work so much better for providers. Until now, the only real innovation in prior auth has been ePA, electronic prior authorization. And that’s fine; it’s faster than fax. But there’s still no real support for the provider. You are still answering these questions, you still don’t have visibility into the requirements, and you’re still getting pushed into these denial and appeal loops. So while it is a little more efficient, it’s not actually lightening the load. That’s where I’m excited. I feel like we’re finally in a position where we can. And I think as…
Exactly. Solutions like arrive ACCELERATE come in and simplify prior authorization. So tell us more about arrive ACCELERATE. What’s your thesis around building the solution, and how are you approaching prior authorization differently from anybody else?
Yeah, good segue. The first thing that is really core to what we do is that we believe the best auth is no auth. So it really starts with how we can help doctors avoid prior auth entirely: what is the formulary, and what are the other drugs a patient could be on? Think about the pieces of this tech, Ayush. You have real-time benefits. Real-time benefits out of the box are fine, but they don’t work great, mostly because our pharmacists in the pharmacy setting do so much to make those claims pay. So the first step we’ve taken is to use computers the way they can be used: we need to build models and an intelligence layer that replicates that work and makes sure the claim is paying, so that we’re delivering real options to the provider at the moment of prescribing, options they can switch to. What’s most exciting is that when it’s done well, we’re seeing providers avoid 48% of prior auths, which is wild. When there’s an opportunity to switch, they really want to switch. So that’s part one.
Part two is how we can bring a solution that gives them the answers up front and tells them exactly what they’re going to have to fulfill. We’ve built a series of large language models, using Claude Sonnet and a few others, to extract from the coverage requirement documents exactly what the criteria are, summarized in a way that’s easy and quick to read and digest, and that tells you at the beginning of the prior auth process exactly what you’re going to have to do. So in my example of the lab test being nine months old instead of three, you know that before you even start the question set. That’s the second piece: bringing intelligence in with our generative AI tools to bring that clarity and transparency, so that people can cancel. At that point you can say, “I’m not going to start that prior auth. I’m going to cancel, tell the patient to go update their labs, and come back when I know I can get approval,” so I don’t have to do this deny and appeal nonsense.
And the third is that we really believe there shouldn’t ever be a denied claim at the pharmacy. As soon as you have a denied claim at the pharmacy, you start all these extra loops, with pharmacists trying to help and all these other people getting involved in the process. If we can truly get to where the script is written, the auth is resolved and the script is sent to the pharmacy in hours, then we don’t need that. The pharmacists don’t have to carry that burden, and patients don’t have to be confused and delayed by a denied claim at the pharmacy.
So that’s what we’ve built, and it’s been a wild ride. We have two awesome health systems that have been really good partners throughout the process, and it’s working. We’re getting to resolution in about four hours. I’ll give the plans some credit: they’re good at getting decisions back in a timely way. So between a good workflow within the health system and the PBMs and plans being really responsive, I think there’s a real path forward.
Right. Great. One thing I’d like you to elaborate on more is AI, because you have been prominently using AI to solve some of the prior auth challenges that have existed. As a practitioner yourself, what has been your experience with AI in terms of the results that are coming? What do you find excellent currently, what still needs to improve in an AI implementation for prior auth, and where do you feel AI is still far behind, at least as of now? Of course it keeps changing, so I won’t say two years or five years. What is your take on AI implementations?
Okay, yeah, good question. Before I give you an opinion: some of this is the tech itself, but some of it is also organizational readiness to accept or implement the tech. So it’s a little bit of both, because some of the constraints I see are just our collective willingness and readiness to let AI do some things.
So what do I think it’s good at? I think it’s really good as an assistant. I think we have proven that it can really help, essentially as a sophisticated augmented intelligence.
Yeah, no doubt.
That is a proven and well-thought-through use case. In healthcare, I think we need to better understand how we feel about PHI (protected health information) in our models, because there continues to be conversation about it. I think we’ve gotten past a lot of it, but there’s still the question: am I going to put PHI in, or let the model see PHI, and therefore truly be…
If you can elaborate: how have you handled the PHI aspect for your model training, if that’s…
We don’t today.
Okay.
We don’t today. We’ve avoided it. We’ll have to tackle it at some point, but for now our model has been put against the coverage requirement documents, which are the policies and plan documents that are universal to the plan. The model output is then what gets inserted into the normal, PHI-rich prior auth flow. But it’s totally separate: the model did its work over here, and the summary is inserted separately. So I think that’s a spot we’ve got to figure out.
I also think we’re getting better at quality. We’re learning a lot, from a prompt engineering perspective, about what loops to put in post-action to validate the output. When I think about the model we’ve built, we’ve got three subsequent loops with other LLMs (large language models) that validate the output and make sure it didn’t have any hallucinations and that it got everything it needed to. That’s a spot where, as those quality loops continue to improve and our prompt engineering continues to improve, we’ll need less clinical or manual oversight to validate the quality of the output. So I’m really optimistic about that.
And the last piece, which I think is probably farthest away: I don’t think we’re going to let it make decisions by itself in healthcare for a long time. There’s probably some good and some bad in that. I’m not sure there are, and I’d love your thoughts if you can think of a scenario, but I’m not sure there are cases where we’re going to say, “Oh yeah, let the AI agent go do X, Y and Z.” At least at this point, it’s a human-assist tool, not an autonomous action tool.
Right, right.
Do you agree with that?
Yeah. From our work and implementations, we are seeing pretty much the same, again trying to avoid touching the PHI aspect as much as possible. In fact, we have been building solutions more around patient education rather than care plans themselves. The technology exists to build on it and be helpful in that space, but it’s a very gray area in the sense of how you really manage PHI if it’s getting fed to the AI. And I think for all innovators, it’s better to pick the low-hanging fruit and get started rather than waiting for…
That’s right.
…some perfect day to come. But I would really like to dig into your experience of taking the solution to market, because that’s one of the biggest problems for most digital innovators. Tech can definitely do a lot of things, but how did you get providers and payers to actually implement it? You mentioned two health systems using the solution. Any personal experiences of how you got there on the commercial side?
Yeah. On the real-time benefit side of our business, we do about 22 million transactions a month, and we’ve got hundreds of thousands of providers. What we learned there is that what we don’t need in healthcare is more apps, interfaces and portals, specifically for the provider workflow. So if you’re seeking to have influence on the provider workflow, you need to be as natively embedded in the EMR (electronic medical record) as possible. And that has been both good and bad, Ayush. It has meant that this has been a long growth journey.
Yeah.
What’s that?
It’s a painful journey in itself.
It is. It’s long. It’s long. And as a product person, it hurts a little bit, because you have no control of the interface. You influence it, but it’s not yours. So what I think we have gotten really good at is tight feedback loops with our health system partners, to understand the interface, how it’s working and where it’s not working, and to improve the input data (garbage in, garbage out) so that it actually works with the EMR system and we get the outcomes we expect.
So when you talk about adoption, how am I going to get a provider to use this tool? We have sought for our solutions to almost not need training. That’s a stretch; you’re going to need a little bit of change management and training on anything. But truly, our solutions are so native in the EMR that when the system decides to use it, it’s their workflow. With ePA, we have it set up in two different ways: in one there’s a centralized team, and in the other it’s distributed at the clinics. In both cases, that solution is the new workflow. It’s part of the process, embedded within that EMR. And like anything, there are a few bumps at the beginning, but as that change happens, it’s been fantastic.
The other thing I will say is that part of what’s really helped is creating urgency around prior auth by not sending the script to the pharmacy: saying, hey, the script is being held for you to resolve the auth, and then it’ll be released to the pharmacy. There are some rules to make sure there’s no patient impact, of course. But it’s really created a sense of urgency for adoption: I’ve got to do something, I’ve got to pick this up and keep it moving so the patient can get their med. That was a little long-winded, but hopefully it makes sense.
Smart way to navigate.
What’s that?
That’s a smart way to navigate. Every solution has its own journey, but in health more than any other industry, building is one thing, and getting to market and making people really use the product is a skill, or a mountain in itself that has to be climbed. So, great.
Maybe the last thing I’ll say on that is that we have used the standards where they exist. That doesn’t mean they’re perfect. They’re not perfect out of the box, but they’re a great starting point. Getting really deep in those standards and then learning how to make this work better, how to augment it to make it unique and drive our own differentiation. But those standards are a good starting point for interoperability, to unlock opportunities, reduce custom builds, all that kind of stuff.
Right, right. Yeah, definitely. In fact, we share the same philosophy, telling our customers that clinicians would not use a software if it is standalone software. It has to be embedded within the ecosystem. But as I said, building a solution takes a few months, and then the integration in itself is a long journey, so definitely a big struggle.
Let’s switch gears a bit, Christie. Talk to us about your leadership philosophy. You have built and led high-performing teams. I’ve seen the journey of Arrive Health, and you have been doing a fantastic job; that’s the least I can say. So what’s your philosophy for building your team? How do you manage your team’s performance? What are some of the things that you do?
Yeah, yeah. I think about my job as the leader in two ways. One is to really unlock and help people perform at their best, whatever that looks like for them. That’s a combination of making sure folks are in the right role, making sure they have the tools they need, and making sure the teams are coming together and functioning cohesively in a way that allows the whole business to move forward. Cross-functional communication is often a spot of focus, because we’re all so dependent on one another to reach an outcome but come with vastly different perspectives on what’s most important. So we spend a lot of time on that.
The second is that I really think about my role as being a thought partner. So I spend a lot of time in the weeds, not because I think I’m better at solving the problem, but because I can help the team by asking questions and bringing a truly outside voice to the problem we’re trying to solve. I think the team would say that’s usually helpful. Sometimes it gets us down rabbit holes, but mostly helpful. And in that way, I would also say… oh, go ahead.
I didn’t want to interrupt your thought, but I was just thinking that as leaders, most of our work eventually becomes connecting the dots. People pick up different dots, and then you come in and enable the connection of those dots, which eventually creates the magic. So that resonated with what you said about bringing an outside perspective to help the team.
Yeah, yeah. It’s a good point. That’s another thing I think is so important. I spend so much time trying to share context, and I really do have this philosophy that there’s nothing that shouldn’t be shared with everybody and anybody, because information is the only way we’re going to make the right decisions, all the way up and down the organization, to your point. So part of being a thought partner is immense context sharing and making sure everybody has the same information.
Right. Great. So Christie, this has been a fantastic conversation. Thank you for sharing your insights on healthcare innovation and your journey. Before we wrap up, where can our listeners learn more about Arrive Health and follow your work? What’s the most active social network for you?
Yeah, I think follow us on LinkedIn, and go to our website, arrivehealth.com. On our website you can find all of our thought leadership, blogs and podcasts like this, and on LinkedIn you can see our more recent activity.
Thank you so much, Christie. To our listeners: if you enjoyed today’s episode, make sure to subscribe and leave us a review. Stay tuned for more conversations, and until next time, take care and keep innovating.
At the time of recording (published March 2025), Christie Callahan was a senior executive at Arrive Health, a Denver-based company that brings real-time medication benefit information and electronic prior authorization into the EMR prescribing workflow; the episode introduces her as its CEO. She trained as an engineer and began her career in manufacturing on the Puffs tissue line in Green Bay, Wisconsin, then worked as a consultant at Bain & Company. CVS Health recruited her onto a team that stepped in to fix operations, starting with the SilverScript Medicare Part D plan and then Medicare Part D coverage determinations and appeals, along with work across CVS’s specialty and mail-order pharmacies, call center and prior authorization operations. She joined Arrive Health during the COVID-19 pandemic, working remotely. LinkedIn: https://www.linkedin.com/in/christie-callahan/
Working out how prior authorization or AI fits inside your clinicians’ EHR workflow? Tell us what you’re building and we’ll walk you through how we’d approach it.
Partner with us to design, build, and scale digital solutions that drive better outcomes.
Global Tech Teams LLC, 525 Washington Blvd, Industrious at Newport Tower, Jersey City, NJ 07310, United States.
Let’s discuss your goals, workflows, and next steps in a focused consultation call.