AH103 - Proactive, Personalized, and Powerful: The Future of AI in Health Benefits with Amit Srivastava

Transcript

Lightly edited for clarity.

[00:22] Justin Venneri: Hello and thank you for joining us for another episode of the Astonishing Healthcare podcast. This is Justin Venneri, your host and Senior Director of Communications at Judi Health. And this episode is a long-awaited one. We always get asked, what are we doing in AI? How does AI work? What are the best uses of AI in healthcare? And we're here to answer some of those questions today. With me in the studio is Amit Srivastava. He is our vice president and head of AI. Amit, welcome to the show.

[00:48] Amit Srivastava: Thank you so much, Justin. It's really great to be here finally. I've been meaning to do this for a while, but we have been building and building and just in time to come and talk. Yeah, it is busy.

[00:59] Justin Venneri: So Amit, before we get into some of the questions here, tell us a bit about your background. Frame things up for us. Give us a little story about your path to Capital Rx, now Judi Health, of course, and your role here.

[01:09] Amit Srivastava: Yeah, so I've been in AI for a very, very long time. I mean, I'm not going to give out dates to age myself, but I started in AI towards the early part of 2000, almost like on the East Coast in Massachusetts. I was working in a US government contracting company, and we used to do sort of far-left or esoteric projects for the US government, preparing for the future. That's what the US government does. They do all these amazing things so that it will bear fruit in 20, 25, or 30 years. And AI was one of those.

[01:41] Justin Venneri: So you're like 32 years old? [laughter]

[01:44] Amit Srivastava: Yeah. And you know, some of this work was really cool. I mean, I know there's a huge bunch of stuff that DARPA has done that has led to amazing stuff. But all the great things you see today in AI, like Google came out of a research project out of DARPA. Speech recognition was responsible for Alexa becoming such a big hit. Large language models, actually we were working with language models and I was doing speech recognition back then. So I was there for a long time. Then I got sort of restless. I wanted to work in the commercial space, I wanted to work with large tech companies. So I moved to the West Coast, I went to eBay, I did conversational voice agents and text agents. A long time back, way before ChatGPT and everything. That was really fun. From there, I went to Microsoft, worked in PowerPoint, and built the AI systems in PowerPoint for Designer and Presenter Coach. Went to Talkdesk for a short while, then went to ServiceNow, which is a large IT behemoth. And I was the VP of AI science and architecture there before I came to Judi Health, or Capital Rx as it was back then.

The main reason was I've always been very passionate about healthcare and I really felt a kinship with AJ's vision where he basically told me that he's going to fix healthcare, and I really want to see that. I would love to see that in my lifetime in this country and whatever I can contribute to it. So this is a great place to be for that. That's how I got started at Capital Rx. Back then the goal was very simple, see if we can integrate AI into the main platform Judi that we have to reduce costs mainly, but the scope has grown as we have gone along. We are doing that plus five different things. We are trying to build AI into the fabric of the whole company, using it for our employees, our technicians, our pharmacists, our customers, and everybody.

[03:29] Justin Venneri: For us in marketing too, right?

[03:31] Amit Srivastava: Exactly.

The Top AI Use Cases in Healthcare Today

[03:33] Justin Venneri: So let's get right into that. You know, just might as well hit it. What are the two or three main use cases that you see for AI in our world to try to fix healthcare, upgrade our systems, improve processes, et cetera?

[03:44] Amit Srivastava: Yeah, that's a great question. And actually, that was the first question I asked when I got in here. How do I help? What is it that AI can do? Well, the first and foremost that I saw was the processes in healthcare are very human-driven, they are very manual in nature. A lot of them are. It's not like everything is. I mean, Judi by itself is an amazing platform. It's taken legacy manual systems to the next level by digitizing the workflows and making certain that things can be done much faster, much more deterministic, but yet there are still a lot of workflows remaining that are still manual because somebody has to read these documents that are coming in and extract all the relevant information out of it to be able to seed that into Judi. Somebody has to reason over many of these disparate sources of information that are coming in to decide whether prior authorizations should be approved or denied. Somebody is actually looking at all of this documentation to answer questions that next-stage pharmacists have to look at to be able to figure out how to make a decision one way or the other, or what to respond back to the customer.

Healthcare is so complicated in our country. I mean, in my opinion, healthcare and tax law are probably neck and neck in terms of complexity, if not healthcare being worse. People have a lot of questions. They're always asking about, why this, why not that, why am I paying this much, what's going on with my benefits, how do I get things to work, how do I make this work for my particular use case? So in my opinion, the way I see AI right now, perfectly poised to help with manual workflows with human processes that can be significantly sped up as well as automated to reduce the load on healthcare. Because a lot of that load results in costs that are then passed on back to the members, to the payers, to insurance. And essentially, it raises the tide of costs everywhere.

So I think that's one of the first things that we are doing. You asked what are the main two or three things that we are looking at? Number one, clinical workflows. What can we do with AI to reduce manual intervention, increase automation, and increase efficiency? Number two, customer service. There's a huge demand there because there's a lot of people calling in, they have questions. Humans are right now answering those questions because this data is very sensitive. These people are all within the United States. They're amazing. We have one of the best customer service departments anywhere. We just won...

[06:06] Justin Venneri: The Stevie - Stevie® Award(s).

[06:06] Amit Srivastava: Stevie award - yes, there you go. We want to keep that. We want our customers to be happy with our customer service, but then we also want, as we grow, to keep our levels the same. I was part of Talkdesk, which is a CCaaS provider out there. And I know there's a lot of AI being built there. I used to be VP of AI at Talkdesk as well. I know one thing, platform providers can do only so much. They need people who are tied directly to the use cases, the domain, and the industry in which they're providing their platforms. The data is absolutely key to understanding the workflow and providing a customized solution. That customized solution makes it really sticky. Otherwise, it seems like just a superficial solution that people don't accept. We can't accept it because we hold our customer service in very high regard. So that's another one we could look at.

And the third option obviously is what can we do for our internal workforce? How can we help AI make them super efficient so that everybody is working with the same amount of time, but producing double if they can or more because the tools are there and they are working.

Streamlining Clinical Workflows and Document Ingestion to Improve Health Benefit Administration

[07:16] Justin Venneri: So let me go back to the beginning part of this on the clinical side because you mentioned documents and extracting information. I know the fax machine still exists for a reason, and I think largely it's healthcare-related. Talk to us a little bit about what AI helps with in terms of that kind of document ingestion process and the whole PA approval process. It helps to clarify because people ask all the time, it helps with clinical decision-making but not denials, right?

[07:44] Amit Srivastava: No, no, not at all. Absolutely not. I mean we are very, very conscious about ethics and responsibility in this company. And in all the companies that I worked, I've set a very high goal on that. It's really funny that people are still using faxes. I mean I actually know what a fax machine looks like. I've seen one in my life.

[08:05] Justin Venneri: I still have an HP hybrid fax copy printer on my desk here next to my computer.

[08:11] Amit Srivastava: Oh you do? I have one of those printer copier fax hybrid machines and my son always asks me, "Dad, what does that fax mean?" You would be surprised. A lot of our documents come in as faxes. A lot of our requests are people printing out on paper, filling it out by hand, and faxing it to us. Fax numbers are everywhere, especially in healthcare, provider to provider, provider to us, to insurance. A lot of faxes going back and forth. Now what happens when you fax a document in?

The document doesn't come in as a PDF that has fields filled in with text that you can just extract using computer software. It actually comes in as an image, as a photo if you might, of the document itself. Then you basically say, well what is the problem? It's a photo. You can just scan and convert it into text using OCR - optical character recognition. But it doesn't always work very well. There are many issues associated with that. These are very low-resolution images because the fax machines are old technology. They basically use telephone line encoding and compression on these images. Sometimes there are a lot of artifacts that happen on the image themselves, then you have to extract out the fields. You don't know which field is what. What are the constraints on the fields themselves. So straight OCR may not work.

As I said, sometimes people just write it in by hand on these documents. And so you have the template of the form itself with all these boxes drawn and then people writing on top, so [black] pixels are mixing with each other. It's a complicated project. There are a lot of people working on this. It's an intelligent document processing project. We worked with providers out there that do this again as a platform solution. Didn't work really well for us, didn't work on our documents.

So then we started building our own solution that could extract all this information reliably. Because for us, if you take a faxed document and convert that into a case in Judi, you need to do it with near 99% reliability. Otherwise, you might as well not do it. You might as well let the human touch it. I mean, you can use the initial output and give it to the human, but if the accuracy is low, humans will tend to reject it and start fresh. I've seen that same problem with speech recognition when I used to work on it 20 years ago. So essentially we decided we had to build our own techniques.

So we're building on top of large-scale, open, closed source, large language models that are out there, but then we are building our own set of algorithms. We learn from our data... not training.

We never train on customer data, we are very careful about that. But we try to learn about the structure of the form itself, which doesn't have any data, doesn't have any PHI or PII. And we try to use that to learn how best to do this. That's what we have deployed. Once the system extracts the information from a form and creates a case, the prior authorization process, which I think you've already explained in a previous podcast, involves technicians answering a bunch of relevant questions about the drug going through the prior authorization process regarding this case for this particular individual, the member. By looking at all of the information submitted with this case, the medical history of the individual, any recent tests, all the notes from the doctor, the system takes all of this into account and the technician answers these questions.

After they answer these questions, the pharmacist looks at it and decides whether it's an acceptance or a denial. What we did was build an ambient agentic solution that mimics that whole process and it basically tries to automate it all the way up to approval. So if it can approve, it approves it. If it cannot, it leaves it in that state. So a human goes in, takes a look, and makes the final decision. Our system makes no denials. It's only if we are confident about making an approval, and we try to minimize the times we make mistakes on approvals because those mistakes are costly to our clients. So we're trying to do as little of those as possible until our acceptance criteria goes through the roof.

I come from different environments where if you were 85% accurate, it was good enough, especially in the consumer space. But here I see my PMs dropping numbers like 99% and I start wiping my brow.

[12:09] Justin Venneri: It's definitely a high accuracy number. It's good to hear.

[12:12] Amit Srivastava: And we meet it. And that's something that I'm really proud of, the team having done that.

[12:17] Justin Venneri: So when you have this process kind of sped up by AI and the information's organized, what are the main benefits of that? Is it all more about time to get paid, or does it speed up notifications? How does that part of it work?

[12:28] Amit Srivastava: Yeah, that's a really good question. So ultimately it is time to response. You know, from the time that a doctor's office submits a prior authorization request to the time that they get back a notification about a decision being made. The prior authorization, I think in the industry, is the process that takes the most amount of time already. In Judi, in our company, with the prior authorization tools that we built, we cut down the time quite a bit because of the automation built into Judi. With AI added in, we are cutting it even further. And in fact, we can take care of a lot of the process without a human getting involved. For all the simple approvals that can go through for cases which have less complexity involved in them, which means people get a decision quickly and no humans are involved, which saves money. I mean, in many ways, if you think about it, you know, a member or client, whichever way you look at it. But our business model is not about—

[13:21] Justin Venneri: We don't make money on drug spend. So that's.

[13:24] Amit Srivastava: There you go. Right.

Related Content

• AH086 - Balancing Technology and a Human Touch in Member Service, with Lisa Ellerhorst and Sonia Pettis
• Health Benefits 101: The Importance of Clinical Programs
• Pharmacy Benefits 101: Prior Authorizations
• Judi® Redefines the Role Tech Plays in the Administration of Medicare, Medicaid, and Commercial Pharmacy Programs

Enhancing Member Care With AI Customer Service

[13:25] Justin Venneri: So yeah, no, it's great. I mean, I know for varying populations, there are pretty tight regulations around turnaround times, and we pretty consistently stay below that 48 to 72 hour mark on turnarounds, especially on easier ones. And then I'm curious, Amit, on the member care and contact center, if you can elaborate on that a little bit. Are we building our own solution there too? And is it sort of like AI, more human-like agents? What are we doing there to deploy AI on the customer care side?

[13:57] Amit Srivastava: That's the next progression of where this AI is headed. AI in customer service is not new. I mean, there are two areas where AI seems to be proliferating: coding and customer service. Those are two areas which seem like the best areas for AI. Like I said, we did look at solutions that are platform solutions that we could use right out of the box that we could integrate with our CCaaS solution that we could deploy quickly. There were a couple of problems with that. The most important problem is that platform solutions are very difficult to integrate with your own data, with your own systems. With Judi, which is a proprietary system with a lot of data behind it, that could have become a huge project by itself. We were very conscious about that. The second and most important reason was that a lot of these platform solutions are basically designed to be cookie-cutter solutions. So there's not a lot of ways you can customize them and make them work for your particular use case - there is a little bit of leeway, but then you don't have a lot of control over what you can do to make these conversations feel human-like and feel warm and really engaging. So we are very concerned about that right from the get-go.

As I said, our customer service department is an award-winning department. Our customer service is one of the most highly rated in the industry. We didn't want to jeopardize that one bit. However, there is a clear opportunity here. And the opportunity is that just like in the case of the clinical workflows, there's a lot of easy cases, there's a lot of low-hanging fruit in terms of people calling about the status of a claim, the status of a prior authorization, or statuses where they can do self-service or asking about simple benefit questions that AI agents can absolutely answer. Anything more complicated than that, obviously we have to take it to the human. Our humans are amazing at doing it. But even there we have an opportunity to work with agent assist solutions so that the human can answer those questions fast. They don't have to wait around looking in Judi, how do I get the solution, how do I make the determination that I can pass on to the human?

So both in terms of automation, being able to deflect a lot of the calls to AI agents, as well as being able to help with efficiency, reducing the time taken for customer service representatives to respond to a particular call or the time for each call, we had an opportunity to address both of those. That's what we are doing; we are building the agentic framework and the voice agents that will help us do that.

The Future of AI in Healthcare: Proactive and Personalized

[16:14] Justin Venneri: Very cool. And I gotta ask about the future state. I've heard it - what I'm about to ask about, and I think it's very cool. So tell me a little bit more about this and the opportunity to have a conversation with Judi herself. That's our adjudication platform, as you mentioned. It's our proprietary platform. But is this like Jetsons, flying cars, and why don't we have this by now kind of thing, or is it right on the horizon?

[16:36] Amit Srivastava: AJ asks me almost every three months, "Why don't we have it already, Amit?" I'm like, we're almost there. So that's a great question. I mean, and frankly, any platform out there has a conversational interface to it. If you look at it, AWS, Azure, Snowflake, they have conversational interfaces. It seems like it's something that you need. We definitely want to go there. I'm very excited about that. Like I said, in customer service, if you look at the kinds of questions that our members are asking, they're essentially asking questions to Judi. It's just that a human interprets in the middle. The human works with Judi, gets the answers from here, and passes it on to the user. But essentially they're asking about stuff that's happening in Judi.

Judi does everything end-to-end, almost all of it modular. The few things that we are building with AI, it already does automatically, automagically, I would say. And so if you think about it, what if we made the middleman sort of optional? They can walk in when they need, but you can directly ask Judi, what happened to my claim? How do I fix this? What should be my next step? What about a CIO or somebody who's paying for the platform asking about how many claims are we getting through the system? What's the trend looking like? Where do you see the issues happening? All of that is absolutely possible. We're building the agentic framework that will allow us to do it. But what really excites me about the future, why did we go from Capital Rx to Judi Health? We went from Capital Rx to Judi Health because Judi Health is about the future. It's about bringing all of healthcare benefit management under a single platform, under a single unified umbrella.

And what happens when you do that - when you have a platform that has all of the data flowing through it, not as silos, but directly through the platform? Just imagine what you can do with AI when you're there. You have a member's medical information flowing through that. All their chart notes, their visits to their doctors, their medical tests that they have done, all the medicines that they have been prescribed, all the tests that they have been prescribed throughout the years. Their personal blood test reports, everything combined with all the medicines that they actually filled, the medicines that they're taking right now, the ones that they have filled at the pharmacy. Same with dental and vision. All of it in one place. What that allows us to do is go one step further and get into this realm of being able to do proactive, personalized AI.

That's what I'm really excited about. Because nobody right now, nobody in the US can do it other than us.

[19:02] Justin Venneri: Yeah, it's awesome. And we're both sitting here wearing glasses, so we understand the pains of having vision as a separate benefit.

[19:08] Amit Srivastava: Exactly.

The Astonishing Use of Paper Claims

[19:09] Justin Venneri: Very cool. And it's all integrated. Okay, Amit, thanks so much for sharing what we're doing and a glimpse at what's most exciting to you about the future. I gotta ask, because I ask everybody that comes on the show, what's the most astonishing thing you've seen or heard or experienced related to the use of AI in healthcare or on our plans and our discussion here today. Tell us a good story or two to send us off.

[19:29] Amit Srivastava: Yeah, well, I think the most interesting thing that I heard, I don't know whether it is true or not, but I do see a lot of paper claims being submitted and a lot of the data that's coming to us scanned in. Like I said, fax boggle my mind still. And like in 2026, we still have fax machines around and people use them. But what really boggled my mind is a lot of vision claims are being sent by people printing the form that they could fill in via PDF, printing it out, filling it up, and either scanning it and sending it in an email or faxing it into our system. And I'm like, wow, we pull our hair about digitization. I'm like, yeah, that is amazing.

[20:09] Justin Venneri: I think I saw a stat the other day. So vision and dental are collectively. I know they're widely utilized, but they're worse than something.

[20:15] Amit Srivastava: You know, what I heard was, at least in dental, I don't know about vision itself, but in dental, I mean, the images that we might be getting for this one use case, I don't know whether that's all the use cases we have, but it can be thousands of claims coming in per month. That's what it looks like. It can grow significantly more.

[20:31] Justin Venneri: So, like thousands of paper claims coming in.

[20:33] Amit Srivastava: I mean, they're not coming from members, they're still coming from the prescribers who are trying to get paid. But even paper claims I've heard are in the order of thousands. Yes. Again, don't quote me on that. I'm not the best person. I'm more of the AI guy, not the stats guy. But I did see those numbers pass my desk.

[20:52] Justin Venneri: That's a lot of paper claims. So, Amit, thank you so much for taking the time. I appreciate having you on the show and hearing your thoughts and insights, and I'd love to have you back on. Hopefully, we can have another great discussion about how things are going with the use of AI for these workflows and processes and maybe share some additional data down the road. That'd be great.

[21:09] Amit Srivastava: Yeah. It was really a pleasure being here and I can't wait to report back on some of these capabilities. I mean, we have many of these capabilities in production. Many are going into production soon. So I'll have a lot to talk about at that point.

[21:21] Justin Venneri: Awesome. Well, have a great rest of your day.

[21:23] Amit Srivastava: Thank you.

‍

‍

Want to stay apprised of the latest Judi Health / Capital Rx news? Sign up for our monthly newsletter!

‍

Interested in transitioning to an aligned and transparent pharmacy benefit partner? Click here to get in touch with our team!

‍