AI Automating B2B Healthcare Conversations | Ankit Jain, Infinitus Systems

This transcript was generated automatically. Its accuracy may vary.

Mike Koelzer, Host: [00:12:00] Ankit, for those that haven't come across you online, introduce yourself and tell our listeners what we're talking about today. 

Ankit Jain: My name's Ankit Jain. I'm one of the founders and the CEO of Infinitus Systems. We're an automation company focused on healthcare. We've been building a machine that can make phone calls on behalf of providers, on behalf of pharmacies, on behalf of patient access programs to payers to do things like benefit verifications and to verify the status of prior auth. I thought it'd be very interesting today to talk about not only what we do, but really focus on what's been unlocked in healthcare, using

ai, what's possible

today, what will be possible in the coming months and years, and what maybe never possible, 

Mike Koelzer, Host: Boy, Anke you came at the right time. Because I've been in pharmacy for a long time, and we could fill twice as many prescriptions back in the day because there was no prior auth. There was no blockage because of this phone [00:13:00] call needing to be made and all this, so we could just fly through prescriptions.

 just before Covid. I was out of the store for about three years. Just various reasons. When I came back, I was astounded even in those three years how difficult it was just to get a damn prescription to go through.

It almost seems like the PBMs put up these big walls on purpose, just for the hell of it. As long as they're screwing us, they might as well really screw us. So half the time it sounds like we're on the phone with people.

Ankit Jain: I Think the last part of that is absolutely true. I do think half the time we're on the phone with people and that's unfortunate 'cause it's a waste of time. but it's important to recognize that the provider or the pharmacist or the pharmacy technician or whoever's calling is on the phone, but there's also someone on the payer or PBM side that is also on the phone.

So it's a cost to both sides. When you're doing that, it's a cost and time. It's a cost and money. And the person who gets screwed in all of this is the patient. 'cause their therapy is being delayed. I think everyone [00:14:00] recognizes this now. I think there is a belief in the ecosystem, I'm relatively new to healthcare about five years in. And from day zero, I heard folks say the players and PBMs want to put up these roadblocks in order to make it harder. For people to get on therapy.

 at the surface level it's very believable. it's the story in David versus Goliath that we

all want to understand

that the big bad corporation is making it hard. As someone who pushes .Hundreds of thousands of phone calls towards the payers and the PBMs. I've had the pleasure if you may call it to be on calls with them, where they go, why the hell are my humans talking to your machines all day long?

And I said, it doesn't have to be that way. You could give us your data digitally and we won't call you. 'cause our customers at Infinitus don't care whether the data has gotten through an API call through a phone call or through a fax. They just need that data in order to get the script [00:15:00] filled. And in the process of these conversations, we. Ask the payers and PBMs, can you give us this data digitally? Some of them said let's look into it. And what they realized is that most of that data isn't available easily even to them inside these big corporations. So when you call a call center and you say, for this specialty medication, is prior authorization required?

And what is the benefit? Is there a copay or coinsurance?

That agent might go from between three different systems, between six different PDFs, dig out the answer and give it to you. So the first thing I'll say is the power of large language models. All the stuff we're hearing in AI is to take all of those PDFs, suck them up, extract the right data, organize it, and create an API so it can be surfaced to everybody else.

And we're working with a couple of players on going and experimenting with exactly that, which is how we can make this data. [00:16:00] Available more instantaneously, but in the absence of that, you need a phone call. The second thing that I learned which was fascinating to me was with a major national payer. I said, Hey, you know, everyone on the provider side says you guys are putting up these prior OTs too, to delay therapy, to all these things that seem bad. Why are you doing it? And the lady I was chatting with said, you realize I'm a patient and my family's also part of the patient population of this country. I wouldn't want myself or someone I love. To be delayed because of some idiotic process somewhere.

So I said well, can you get rid of prior auth,

just hypothetically?

And what this lady said, I will never forget, she said, I could get rid of prior auth, but 40 to 50% of the patients that we administer insurance, two are covered by self-insured plans. And we are not the ones determining whether prior auths should be [00:17:00] required or not.

It's the employer is and

the whole ecosystem. And so it's not an easy fix to rip the bandaid off. It goes pretty deep

deep. And so then I decided to go a step further. I chatted with some folks that were at UPMC, which is, they have their own plan. They've got their big health system and they had run an experiment for six months where they got rid of prior auth. And the providers ask for it back. 

And the reason they did was they found that people were not being as careful. They were over-prescribing and their resources were being strained in ways they had not previously 

predicted. So instead of prior auth, they've rebranded and they've brought something back called clinical decision support

 So prior auth in its worst form was by default you deny and you really make it the burden of the provider or the pharmacy to give you

all the proof of why it should be approved.

A clinical decision support, as I am seeing it evolve philosophically different where the goal isn't to default [00:18:00] deny it's to ask questions and say, have you

done X?

Have you done Y? But

the provider or whoever's going through the system can override and say, I. Think this is the right thing to do, but it's guiding folks in the right direction. And we're seeing the bigger players go in that direction and also do things like gold carding where they're allowing providers who have a good track record and have shown proof of doing things the right way in the past to go right

past it. So I think there's a lot of interesting motion in this space. It's an interesting time for sure to be anywhere around prior authorization

Mike Koelzer, Host: It seems with that statement that everybody wants efficiency. Our fun is picking on the PBMs, trying to slow stuff down. Everybody wants efficiency. I know lawyers don't want efficiency, but everybody besides lawyers want efficiency.

Ankit Jain: I mean, I couldn't agree more. Maybe the RCM companies don't want efficiency either.

Mike Koelzer, Host: What is RCM?

Ankit Jain: Revenue cycle management companies.

'cause they manage all these back office processes and if [00:19:00] everything was digitally

connected and everything just worked and there were no denials, therefore no appeals, they'd be

out of business.

Now the other kind of ethos, the other thing that I've heard over and over again is a lot of these systems have been put into place to protect against the fraud,

waste and abuse that happens in any system with humans in

there. fraud, waste and abuse that is happening because one to 3% of the population is penalizing the other 97%.

But that's, any system.

Mike Koelzer, Host: I listened to this on your website, there is an example of what you were talking about, but do a little role play here of what you're talking about, what your system can do. Give us an example of that. 

Ankit Jain: sO let me set up a situation.

You are an ophthalmologist and I am a 60 something year old who's suffered from diabetes most of my life. And I walk into your office after waiting for three to five months, because specialist appointments are hard to get. And I've been having difficulty with my vision. [00:20:00] And I walk in, you have a look at me and you say, oh, you have a history of diabetes. You're having difficulty seeing. I look into your do some tests and you say you've got something called macular degeneration. And I go, that's a big word. What the heck does macular degeneration mean?

Ah your body degenerates over time.

In this case it's your eyes

that are parts of your eyes that are degenerating, which is why you have difficulty seeing,

I go, doctor what can I do? Does this mean I'm gonna go blind? Oh no. This is pretty common. It happens amongst diabetics more often than we'd like to admit.

 And there's medication for it. Ah, great. So can it be cured? It can be managed. How does this work, doctor? The way it works is you come in every two to five months, depending on the medication, and I inject a medication into your eyeball. Wait,

what? You're gonna inject something into my eyeball?

 Yeah but you'll get used to it. The first time will be a little weird, but then you'll get used to it. And, people have had this for decades and it keeps them going amazing. Can we get started? Because I'm really, [00:21:00] it's really hard for me to do live

my day-to-Day life.

Wait a minute. These medications cost anywhere between 10 and 50,000. Oh wait, what? 10 to $50,000? I can't afford it. There's affordability programs, there's copay assistance programs. We can find all that out. But first, before we do anything, my team needs to make a couple of phone calls to your insurance company to see. Who will pay for what and how much you will be liable for. And if you can't afford it, then we'll look at the affordability programs and make sure you're set up. Okay. So now that you know what the problem is, what the therapy is, when

Could I get started? Somewhere between two and four weeks. 'cause if it needs a prior auth, we need to submit the paperwork, go through

that, and then my team will call you to schedule it. That's the conversation

Ankit Jain: hundreds of thousands of patients have.

And then the diagnosis gets put into the EMR. It gets into a work queue to say you need to go process a prior author, do a benefit verification, and then someone picks up the phone, [00:22:00] calls the payer, goes to the IVR, waits on hold for half an hour, 45 minutes, and then asks 80 questions. 'cause you need to figure it out. Deductibles out of pockets. Prior auth requirements, step therapy requirements, whether you can buy and bill, or you have to go through a specific specialty pharmacy who's in the distribution network. All these questions. And then that person's job is to just do these benefit verifications all day long.

Once they figure out their prior Roth requirements, fill out the paperwork, submit it through a portal, or fax it in, and then every 24 to 72 hours call to check the status of it. What Infinitus does is we say, Hey, provider's office or patient access program or specialty pharmacy, your people joined healthcare and joined your organization to serve patients. They didn't join healthcare, to wait on hold to collect the same information over and over again

instead of picking up the phone and making a phone call. Let's make a digital [00:23:00] connection to our service and my machine, the Infinis digital assistant, Eva, will make a phone call to the payer, go through the IVR, wait on hold, talk to the agent on the other side, collect all that information, extract it from the call, and return it back to the customer in a very well organized format.

.

. It seems the next step will be to have two of these things, so they're on that side with a robot listening and your robot's talking. I imagine that would be cool, but at that point then you wouldn't really have to use words, right? this could all be electronic, back and forth.

Mike Koelzer, Host: Talk about that a little bit like where this goes then.

Ankit Jain: Yeah, I mean, it goes back to the thing I was saying earlier, which is many of the payers and PBMs tell us why the heck are our people talking to your machine?

 One of the largest ones in the country said, can you build us a machine that can answer the phone calls? And

 We chuckled. And we said, [00:24:00] sure, but can you give us the data because a machine

can't talk without having the data. And we

said, by the way, if you give us that data that we can then have the machine talk and answer questions. Any calls that we make, we will get rid of the call because you've got two machines that can talk to each other. Through an

API and what we found again is today that data isn't available digitally as we go down that path and create those APIs. The ideal end state is getting rid of these phone calls,

is getting rid of the delay because if, even if two machines have to talk to each other, there's still a 20 to 30 minute conversation between those two machines as the data

is exchanged. But if you can do it all electronically, that can be milliseconds. And just think about the fact that that provider that I talked about, that ophthalmologist could say, Hey. Patient, just gimme for two seconds. I'm gonna run a couple of things, click a couple of buttons and right away I will re see if your prior, if prior auth is required, if I can buy and bill, and whether your prior auth [00:25:00] is approved.

If all of that could be instantaneous and that patient could go home with their first dose that same day, rather than needing to go back home, come back from wherever they live a, few days or a few weeks

later, and feel better right away, that's a win

for everybody. 

Mike Koelzer, Host: So Onca remember about five years ago, I'm an Android guy. And anyways, Google came out and they said, Hey you don't need to spend all day making appointments for yourself to the barber shop and that kind of stuff.

And they said, let us show you what we can do. And so the barber says, ah, this is Harry the barber. What do you want? And the robot then is saying, Hey, I need an appointment for Mike. When are the openings open? And then Harry says, I got something tomorrow at five 15.

And the robot says, as any robot, would go, like a real person. Hmm. And they say, that's not quite good for Mike. He can make it at uh, four 30. Would that work for you anyways Google. Is making all of these appointments [00:26:00] for the customer. Now, you worked at Google, not only worked there, but the director of I don't know what the hell you were the director of, but something pretty big there and now you are going off and this kind of seems a little bit familiar to that phone call I had.

What's the connection in all this?

Ankit Jain: It's interesting you bring this up, Mike. At that time I was running Google's AI Venture Fund, gradient Ventures. And it was about a year after the Transformers paper. The transformer stuff is what is now the GPTs and the Bards and all these that we hear about every day in the news. The paper was published in 2017. In 2018, Google announced at Google io this thing called Duplex that you could say, Hey, Google, make me a reservation at a spa, a salon, or a restaurant, et cetera.

Mike Koelzer, Host: You were in charge of. The fund? Is that what it's called?

Ankit Jain: The Google AI Venture Fund.

Mike Koelzer, Host: Fund? Does that mean you were involved with the financial part of it?

Ankit Jain: I

was [00:27:00] investing in startups that were building AI 

Mike Koelzer, Host: gotcha. Okay. Keep 

Ankit Jain: So they were making appointments for spas, salons, restaurants, and the idea behind the scenes was actually quite beautiful. They said if a user says, make me a reservation, if OpenTable is available, it would do it right away. So if there was an API, it would use the API, but if the

API was not there, then the Google assistant would make a phone call and then talk to the person on the other side.

 the

The demo was mind blowing. Everybody who saw the demo was like, this is a marvel of technology.

And I remember going home that night and showing it to my wife. And I'm a deep technologist and so I was talking about speech recognition, natural language processing, transformer models and speech synthesis. And my wife told me to hold on and said, wait, so these brilliant Google engineers came up with a machine that can have coherent conversation. They made spa reservations.

And I said, what do you mean it's incredible? She's no,

but seriously, that's the best thing that they could [00:28:00] think about. I'm like where would you apply it?

In this, in almost this little competitive uh, mode. I'm like, where would you apply it?

way of background. Shelby has been in healthcare for a long time. She was at a home health and hospice clinic.

She was at Stanford in the oncology group and for the last six years has been in the patient access group at Genentech. And she said there's millions of patients waiting to get on therapy, but they're waiting on somebody in the healthcare system to make a

phone call for a benefit

verification, for a prior authorization for a referral to be sent. And so if someone could use this technology to unlock time in healthcare, Reduce the time to therapy, and the cost of therapy by automating these phone calls, it would change healthcare. And that was the inspiration for us starting Infinitus. that Google demo was the starting of that conversation with my wife,

which eventually led to us starting Infinitus.

Mike Koelzer, Host: When you say that, you were in charge of the investor fund. What does that mean? Because when I hear that, I'm thinking accounting and all that, and I know you're not that.

So how does that all fit together [00:29:00] with what you said by the fund?

Ankit Jain: Yeah, absolutely. So I've worked at Google twice. The first time I joined was when a search engine company that I was an engineer at was acquired in 2010. And I helped build, launch and scale something called Google Play. So I ran all of the search infrastructure, AI and, and discovery for Google Play for a number of years.

And then I went off, did a startup in the mobile space. And then the second time I came back I helped start Google's AI venture fund. So what Google's AI venture fund does, it's called Gradient Ventures, is they take money off of Google's balance sheet and invest it into companies that are building AI technology, either technology that others can use or apply that to specific verticals.

It's just a way for Google to support companies, but also get a pretty impressive financial return by being early investors and supporters of some of the most cutting edge companies out there in ai.

Mike Koelzer, Host: I'm always thinking Google is always making this stuff, but it sounds to me like [00:30:00] some of the brains behind it is that they're helping the companies that are doing it. That will then, I don't know, do they put the Google stamp on some of that stuff or 

Ankit Jain: So there's two kinds of corporate venture funds, some that are very strategic in nature, where the goal of investment is to get to know the company better and eventually maybe acquire it.

And some that are financial in nature, just because of their unfair advantages, they're able to figure out, Hey, this company is doing something very special. Let's be an investor and get a great financial return and help them along the way as appropriate. And Gradient Ventures and Google Ventures, which is another venture fund that Google has, are in that second bucket where

They make financially driven investments in companies that they think are gonna do things that are special.

Mike Koelzer, Host: Gotcha. Alright, so you talk to your wife and you think that there might be something, not that getting your hair done is not important, but there might be some things out there for the masses of something like healthcare. So [00:31:00] what was your choice then, and maybe that's too simple of a question, but was it like, alright, Google's not going this way and I think we can go this way and here's a great time to do that.

Or how did you decide then to go out and do this versus not going out to do it?

Ankit Jain: No, I think there's many parts to that calculus. There's the, do I stay at Google? which of course was a very nice, well-paying job. Being able to run Google's venture fund is quite the privilege. And there's the other side, which is everyone that I've ever looked up to, has bet on themselves at some point. I did a previous startup in the mobile space. as I mentioned it was a nice exit, but I think I did it more to do a company rather than because I was so passionate about the idea. the

I dug into this opportunity and the potential impact it could have, and I started this along with my co-founder Sean, and we got to know each other in middle [00:32:00] school.

We were middle school and high school together. We were college roommates. This is our second company together. both of us found a little bit of our calling where we said it's great to build incredible technology, but being able to do that while impacting healthcare

It Was something that was Very meaningful. And both of us have had our healthcare journey since my daughter who was born about two years into the company when she's born, her blood sugars dropped to 12 to 14, which is

Very low, about 18 hours into life. no one at the hospital we were at UCSF, had any idea why that was happening.

And turned

out she had a very rare disease

Called hyperinsulinism. And there's very few kids in the world that are born with this. there's two centers of excellence in the country that know how to deal with it. 'cause the last patient

at UCSF that had, it was a year and a half to two years before us. And so even in the nicu, they did not have the experience, even though it's a level four nicu. Incredible folks, we had to either go to Children's [00:33:00] Hospital of Philadelphia or Cooks Children's in Texas to get treatment for it.

But it was a three week ordeal to go through the test, to get the prior authorization, to get on an air ambulance, to get the baby over there to then be

enrolled in a clinical trial, get prior auth for a surgery. Every single part of the journey

was delaying what was inevitable in

getting her the cure she needed. 

 She's very fortunate and we're all very fortunate that she's fully cured. But through that, the journey that we went through is a journey that almost everybody and their families go through. And the journey to start Infinitus was. Something that we said, this is a higher calling for us, but it's become very personal. It's to help families when they go through their journeys in healthcare which have three parts, right? First is the diagnosis. It's that patient who's saying, macular degeneration and an injection to my eyeball.

That's scary. What does that

mean? The

The second is the industrial complex of healthcare. Who's gonna help you navigate[00:34:00] 

all the complexity that exists? Most people

I don't know what a prior auth is. Most people don't know who to go to for A second opinion. And the third one is the financial journey.

Who's gonna explain to me how much I owe and why I owe that much? I thought, I have insurance.

Why do I have to pay 

all this money? Wait, I didn't choose an out of network. Person to transport me across the country. The hospital

just put me

in an airplane, like, why do I owe a hundred thousand dollars for that?

 I can't afford that. And so that's where, by being able to create time for healthcare by automating those routine, tedious tasks, our hope is that we're able to get those people that joined healthcare to serve patients, to do that. 

Mike Koelzer, Host: 

What has been the biggest? Stumbling block in this where you said, we never knew this part was gonna be that difficult, or in that vein of thought. 

Ankit Jain: I don't think there's one or two things that are hard, the entire [00:35:00] journey of entrepreneurship

and especially entrepreneurship in healthcare is humbling. there's moments of ecstasy and there's moments of pure pain. moments where you can't believe that you are able to get some of the largest companies in the world to even have a meeting with you, let alone become a customer, 

because that gives you distribution and the ability to impact millions upon millions of lives, which

is very fulfilling for me, for Sham, for our team, for our

families that have all sacrificed so much on this. And then there's the other side of it, which is that sometimes it takes 18 to 24 months to go through a sales cycle. And some people say healthcare is slow. But I think I've come to really appreciate the diligence that is done. It's good

to make sure that data is handled correctly.

The security reviews are making sure that all of our data is being handled correctly.

It's good to make sure that if a large company is going to take on the services of a startup, that [00:36:00] startup can survive the test of time because it's gonna be painful to

integrate, it's gonna be painful to

scale it up. But if that company runs out of money, then all that time and energy just goes to waste. So in some ways, the long sales cycle is proving to the company that this uh, vendor is going to be around. And the third, the other part that I, I don't think enough people necessarily talk about especially for founders

is the sacrifices that their families make on this journey. Going through this, you go through lots of ups and downs. You take your teams through it in some ways but no one wants to be panicking. So there's a lot of emotion that is built into a founder's mindset and their mind ultimately is dealt with by their families.

 as I talk to my co-founder, sham, or other founders that I know, that's the part that's probably the hardest which is making sure that you're mentally at your best as you go through these hard times and the good times, and you become kind of [00:37:00] even keeled through it. And about a month ago the CE of Nvidia was doing an interview and someone asked him if you could go back in time and knew everything about what you knew, 

Mike Koelzer, Host: remember that interview. 

Ankit Jain: remember? Yeah. And what was his answer? His answer was, I don't know if I'd do it again. 

Mike Koelzer, Host: I dunno if I'd do it again. It was too hard. It was too much on the 

family, too much. 

on me. He didn't know if he would do it again.

Ankit Jain: Yeah. And I think, when you think about the lessons and people, quote my parents, my family, my wife, others quote this to me, they're like, when you're on your deathbed, are you gonna think about how hard you worked or the time you spent with your family?

And like,

where is your calling and where do you want to do it? And I think that's

a constant struggle for every founder. 

Mike Koelzer, Host: You were talking about making sure a vendor is gonna be around and so on, once in a while someone . Heard me on the podcast or something, they'll say, Hey, we've got this technology and we can do it in the pharmacy and it's not gonna cost you anything.

We just want to have a site to try it and things like that. I'm like, no, it's gonna cost me a lot. It's gonna cost me time. [00:38:00] It's gonna cost the trust of my employees who trust that I'm not gonna throw some technology at them, that we didn't think of something or other. And you got customers, upset and that because they're waiting on this and that.

Even if something's free or even if they would pay me to do it, you gotta be really careful about who you're getting in with.

Ankit Jain: absolutely. The one thing I will say that I've also learned through this journey is how earnest almost everybody in healthcare is. Everyone is pointed in the right direction. We kicked out this conversation talking about the roadblocks on the prior AU side. I consider myself very fortunate because of our business, we get to interact with. Every part of the ecosystem. I tell my team that there's the five P's of healthcare patients, providers, pharmacies, pharma and payers. And one of my investors reminded me, there's the silent s, the five P's, and the silent S. The silent S is the sponsor. 'cause they're driving everyone in different ways, whether [00:39:00] it's the government or the employer, the two major

sponsors in this country. But we've had a chance to spend time with every single one of those five pss and the Ss. And everybody wants to do the right thing. And that is pretty unique. Compared to any other ecosystem that you

go work in. And

so if there's that basic alignment, sometimes, timing might be off or the expectations of cost may be off, but

Everyone wants to do right by the patient because, and why is that?

Because they themselves are patients. Their loved ones are patients. So

They want to create a system that serves themselves better. So that selfishness ends up becoming selfless in some ways. And it's it's really, really powerful and really

encouraging. 

Mike Koelzer, Host: On a smaller scale of that, years ago I had to do an inventory or something in our pharmacy, and it had to be done by the next day or something, so I had to have my staff come after work and after about an hour in their first break, I felt like I had a hard hat on, like walking around a construction site saying, all right, [00:40:00] get back to work.

But it's like during the day, it's not hard to motivate the employee because they're staring at the customer in the face and they're putting themself in their spot. They don't wanna let them down and that kind of stuff. So it's nice to motivate people on the healthcare side when it turns into counting the beans on the shelf, because people don't have either the benevolence of not having someone staring at them at least, that can be hard to motivate that way.

Ankit Jain: Absolutely. I think back to, I. Many of the friends I've had that have gone into something like the ads business, you can

make a lot of money

in advertising, whether online or offline.

 At some point you start questioning yourself, which is, am I doing all this just for the money or is there something that's larger than that that I can be putting my time and talent and efforts towards?

Mike Koelzer, Host: So on. We in the outside world have all had fun with the large language models, chat, GPT and all that kind of stuff. And I'm gonna say it's been around, I don't know, we've known about it for, what, a year or something like that. 

Ankit Jain: isN't that crazy? Like [00:41:00] we've only known about it for a year.

Mike Koelzer, Host: Yeah, it's crazy. 

Now someone like you though, who was on the inside track of this, That wasn't a surprise to you, I take it. When you started this, you knew that was out there and that's incorporated somehow. Did anything come to you in the last, let's say, year when this came out to help your company? Or was this just old school to you by that time?

Ankit Jain: I Think a bit of both. I think the technology if you're being part of the ecosystem, it was a linear progression to this

point. Eight years ago, nine years ago, there were teams at Google that I knew that were trying to answer questions and do reading comprehension the way a fifth grader would. And so it's natural that makes progress and how we can take the MCATs and we can take the

LSATs and a machine can do it. The thing that. I think it surprised everybody how quickly [00:42:00] became usable by the masses.

With most technology, the interface is so important. I

Remember in 2008, Google had something called voice actions. And a dear friend of mine was a product leader on voice action. Then you could, say, I forget what they called it, but there was a way to invoke it and say, make a reminder for me to call mom at this time. It was on the

Android devices back then, on the G ones,

uh, that the HTCs back then, but no one used it. It took the new interface of Siri that came out and Siri was never very good.

But that was the start of the. Revolution for voice digital assistance to say, ah, this is how we should do it. You can say, Hey, Google, or you can say, Hey, name of assistant to

then actually be able to converse with it. So, the user interface is almost as important, if not more important than the underlying technology.

This technology has [00:43:00] been

brewing for a while,

and I think chat, GPT was that for large language models. another good

For example, if you go back in history and kind of think about that, I was on the Android team and I saw the Google assistant. All these things came together, but it was Amazon that won the race with Alexa, with their little devices.

And then they had the fire stick. That was a new interface to how you can connect multimedia. And

internet enabled media to your tv. And then

Of course Google had Chromecast, they had Google Home.

But it was the innovation there. It wasn't the technology, it was

putting it together and the interface.

And so I think what chat GPT did was show us all a new kind of interface that we can communicate with knowledge in a different way. Whether it's information gathering, whether it's creative knowledge or creativity, we now have this interface. And so now you're seeing within the EMR you can ask questions and say, [00:44:00] find me that this patient had this diagnosis and it is able to figure that out.

Or let's put together an appeal for a prior auth denial where it is able to go get the data from the right places, stitch it together. And then have a human review it and submit it. Or the person who receives the prior authorization to review the nurse on the payer side today, they get 150 pages. There's 17 requirements.

They're looking for needles and haystacks tomorrow. It's very believable that a company that comes out with an implementation using a large language model, will understand each of those requirements. Go into the 150 pages, find the paragraphs that are relevant, and show just 17 paragraphs so that the nurse can then quickly look at them and say, this makes sense.

This doesn't make sense. So something that would take three hours, can take 10 minutes. That is efficiency that can be unlocked with this kind of technology.

Mike Koelzer, Host: Hey, here's what I want.[00:45:00] I got probably half my time at the pharmacy spent in, my staff spent with these kind of verbal conversations, and it'd be damn hard to put it into a, someone fill out a chart or something like this. So I'm, how far are we away? I know Google has it with their uh, Google

lm, I was kind of waiting for that to send all the Google Docs through it and Anthropic accepted a little bit longer documents and now chat. GPT has caught up with the reading, the PDFs and things like that. How far are we away from a corner drugstore like ours to have the language models get inside of our stuff, able to interact with our pharmacy system, which is a national system, and then take a phone call from Mrs.

Smith who's called four [00:46:00] times that day. Either she's forgetful or looking for her pain medicine or something. How far is that away? Because I know . Some of the people are doing it now, the at and t's and things like that. How far are we away from a little store to do the APIs and all that stuff and then be able to come out with something?

Ankit Jain: we're not very far from that. It's actually I think the hardest thing there, Mike, is not the technology to have the conversation, it's the interface into your system. It's the

ability to take data out and put data back in. Because when Mrs. Smith says, Hey, I'm gonna come pick this up tomorrow at 10:00 AM.

To make sure that it's documented there so that everything is ready for her when she comes in there. That's the part that needs to be figured out. It's the workflows, the conversational AI part of it, that's mostly a solved problem at this point.

 the interfacing and the workflows where a lot of investment is going [00:47:00] into.

There's companies that do exactly that today. And it's only gonna get better.

Mike Koelzer, Host: Anki, if I define AI as machine learning, progressing machine learning that keeps cycling through, how much of that is involved with your product, and does that keep getting better because of the ai without the humans going in there and doing a bunch of it, but just the AI teaching itself.

Ankit Jain: No, I think different models have different ways of learning. In the early days of machine learning, you had lots of humans that labeled data, which was then fed into a machine that could understand the patterns and be able to mimic the exact same patterns. And then there were some pretty big breakthroughs with something called reinforcement learning that had the ability to reward or penalize the system if it was right or wrong.

And that reinforcement got it to improve itself over time.

And in the last few years, there's something called RLHF, reinforcement Learning with [00:48:00] human feedback.

So you have those incentives and those penalties, but you also have the ability to have a human go in and adjudicate whether it did it right or wrong and these systems are constantly getting better. I

I think there's gonna be some areas where a pure reinforcement learning based method will do very well. There's other places where you do need to have humans in there in order to make sure that we're making progress the right way.

If you remember a few

years ago, Microsoft released this bot called te. It was on Twitter and used to just respond to people. And people started trolling Tay. And Tay learned from the conversations he was having with people. So if there was inherent bias, whether it's racist, sexist, et cetera, in how

people were talking to Tay. Pay would learn that and mimic that right back. That's a place where Microsoft said, Nope, we're gonna stop this right away for many reasons. 

one of the reasons being, oh, I guess we did not think about all the controls. I think the power of human feedback is that [00:49:00] allows us to direct machines in a specific way. And so I think especially in healthcare, there

will be that human feedback loop for the near future. 

Mike Koelzer, Host: Sometimes it might be as much as, like you were talking about with the macular degeneration. It might be that the doctor says something and the person just gives a little bit of a, a, fright, like a little bit of a breath in or something, and that means everything to a doctor or somebody who's listening.

But will AI pick up somebody kind of, moving their head one millimeter because they're upset about what they heard.

Ankit Jain: It's very interesting that you bring that up. I don't think we're quite there yet, but there was a paper in jama, JAMA, the journal a few months ago where they did a study where when people messaged their providers on MyChart they had one population where the provider responded, and the other population where an AI [00:50:00] responded. And there were two outputs from the study that I remember. The first one was. The accuracy and they found that they both were pretty comparable. And

the second one was empathy as judged by the patient who

messaged. And the AI was deemed more empathetic than

the providers.

This was very controversial.

I thought it was very interesting. I happened to be in a room full of Chief Medical Officers the week after this paper came out and I was giving a talk and I said, why do you folks think this is, do you believe it or do you not? And about a 10th of the room said, ah, this is bull.

 There's problems with the study, et cetera. The other 90% said, of course this makes sense. Our providers are burned out. They're getting to respond to these patients at 10:00 PM We call it pajama time. After

a hard

day's work, they're just

trying to respond, get over it and move on. 'cause they feel a personal burden in unpaid hours [00:51:00] to respond to these patients.

On the other hand,

an AI can say, dear Mrs. Smith, I hope your day is going well,

and then give the answer. And so of course it's gonna seem more empathetic. That's after a draining day that the provider had at the office.

I think technology will get there. I think the bigger question to ask is what will we want technology to do and where do we want to unleash the power, which could go in many different directions, but we need to have concentration of time and resources towards specific problems. Which problems do we want technology to solve first?

Mike Koelzer, Host: We have this program at the pharmacy we're using to help with some delivery stuff. And it's a rather new program, I think. I think we're one of the earlier customers with it. And whenever we write to the support team before chat, GPT comes out.

they would just say, okay, we got it. It wasn't terse. It was just like, okay, thanks. We got it. We'll think about [00:52:00] that. So then the last five or six months, you know they're running it through JGPT or something because the first like three paragraphs is like, they understand our pain and they appreciate us bringing this up.

After a while, I just jumped right to the third paragraph. 'cause I knew the first couple paragraphs were just language model sympathy things.

Ankit Jain: Yeah. I think we're still in the novelty period of it.

A good parallel to this is San Francisco is full of self-driving cars. I took my five-year-old for his first way more than a year and a half ago. So it was very early in the trials. My wife was kinda like, you're gonna take him in rush hour to go in

A self-driving car? I'm like, yeah he has to grow up in this world. And so I, and he's now six, but he was five at the time

and I told him about the lidar and I told him about how these things are there. We got inside the car, there's no driver. We're driving around and I'm like, how does this feel, rayon? He's 

it's [00:53:00] cool, like

this car is driving itself, right?

And one of the things I think they got right in the interface, going back to talking about interfaces of these self-driving cars is they made it as natural as possible. They didn't do anything funky. 'cause they could do a lot of really funky things.

They could have the front row and the back row face each other because no one's using the steering wheel. In the future designs, that's how it's going to be. But

they're transitioning us into that world by making it as natural as possible. I took my in-laws, I took my parents. They all kind of go the first minute, two minutes, you see the screens showing you all the pedestrians, all the cars, the buses, and you

feel like you're in this other world, but within two to three minutes it just becomes normal. And that the power of interface is so important because at some point you want technology to disappear. And with these language models,

I think there's still a lot of [00:54:00] novelty where we're putting in all these niceties and sympathies. But if are as good as we all believe them to be, they will learn with us and get rid of that crap because that's,

 we don't care about it.

Mike Koelzer, Host: There might be a term for this but it was something about . wHen they ask a hundred people, let's say, and they ask them, they say, rate your driving skills. it comes out where everybody thinks of those a hundred people, they all think they're in the top 25%

They say I'm the top 25%. And that's mathematically impossible. You have to have it every quarter. So I just come out and admit that I suck at driving for various reasons. A computer driven car?

I would take that in a second. I know there's a lot of different ways they do it, but they got 32 cameras on this and that, and it's a speed of light and all this kind of stuff. There's no way in hell I would trust myself over [00:55:00] that. Once it comes along a little bit, 

Ankit Jain: It's one of the safest things I've been in. I

I wouldn't take my 5-year-old if I didn't think it was safe.

Mike Koelzer, Host: Alright, so Ankit, we talked about all the, uh, how's of the business and the why's and so on. Let's talk about the business itself. How is it going for you and is it where you'd like it to be and where are some areas that you thought were gonna come along and they haven't? That kind of stuff.

Ankit Jain: I've learned a lot. I think there's parts of healthcare that are very good at adopting technology. There's parts of healthcare that are slower than others that adopt technology. And you have to hold hands a little bit more, go through a lot more hoops, and understand how you balance these different groups so you can keep growing your business.

So at

at some point you have to have the big wins, but sometimes you fill those with a few pebbles so that you've paved the forward. recently. One of our investors introduced me to [00:56:00] one of the early employees at Veeva which I think is a fascinating story in the world of healthcare technology because of their exponential growth. did a few things that were fundamentally different

than most healthcare technology companies. There's a lot to learn from them. They were different in that they were selling to life sciences and pharma more than they were to health systems and the like. they also had a 10 year exclusive exclusivity with Salesforce to go build what they needed to on top of 'em. So they had a lot of things going for them, but when they went public, they had less than 200. Customers, which is crazy. When they went public, they were doing I wanna say 150 to $180 million of revenue and profit.

That's not the crazy part. They were profitable when they were doing 4 million in revenue. 

they built the company upon their own [00:57:00] profits. They only did one round of financing. They never really used it. And now they're doing, I think, close to 2 billion in revenue. As you hear the story or you look at the numbers of a company like Veeva, they have a couple of things, right? They went after a market that was ready to pay, ready to adopt technology, and they had the right ecosystem forces and a lot of luck behind them to do that. And so when you are an entrepreneur that wants to go into any industry, in our case healthcare, it's important to understand that there's many facets of healthcare. The way payers and PBMs adopt technology and pay for technology is different than the way life sciences and pharma does different than the way pharmacies do, and definitely different than the way health systems and the provider side do. And so

Being strategic about where you enter, how you sow the seeds and understanding that the germination times are going to be different and you don't put all your eggs in one basket can be the difference between life and death.

Mike Koelzer, Host: What is Veeva? Is it [00:58:00] software?

Ankit Jain: They're a software company. They've basically built a CRM. So they were originally built on top of Salesforce. and they had a bunch of applications that they sold to the pharma and life science industry to run different parts of their process, whether regulatory documentation or to manage their customer relationships or to manage all their clinical trial data. But they started in certain areas and they expanded all through pharma.

Mike Koelzer, Host: Anki, tell me about your surroundings. I know you're in San Francisco. How many people now in the company, how many people are you seeing day to day? iF you could divide up the company, rough percentage, how many are in, software, how many are in sales, how many are in this and that, and so on.

Ankit Jain: Yeah our company's about 120 people. around 60 of them are here in the San Francisco Bay Area. The engineering product and design teams are here in the Bay Area.

We've got a six person [00:59:00] sales team. We have a three person marketing team, a 10 person customer success and implementations team. And then a number of folks on our operations and data labeling teams.

Mike Koelzer, Host: The interesting thing about voice to me is that nobody has to be a willing participant on the other side because you're using voice. And so the person on the other end, like you say, if they don't have the APIs matching up or whatever, the person on the other end, they could be really any company in the world as long as they speak English.

And then your system basically interfaces with them only because you both know English.

Ankit Jain: Absolutely. And this was part of the magic of, google Duplex back in the day, 

they said if there's an API, we will use that. if

not the [01:00:00] API will be the English language. 'cause today there's two humans that talk to each other in English, and

 We can also have our machines talk in English. that becomes that common interface for

anyone to talk with. Anybody else there's many companies that have come, some succeeded, some not succeeded in healthcare that have said we will be the future API of all things in healthcare. And the reason the ones that have failed have done so is it's very hard to convince everyone to adopt your system at the same time.

So by being able to decouple the two sides and say let's show ROI on one side, and then as you grow it becomes a network effects game, you go to the other

side and you say, Hey, I'm coming with so much traffic your way. If you do this, it's a big win for you as well. You

eventually get rid of the phone calls. 

Mike Koelzer, Host: if you can wave a magic wand, where does the verbal . Conversation that your system does, where does it [01:01:00] have to do any better? Does it have to be able to tell a joke once in a while? qualities does it have to have, I guess, to make it more maybe you don't wanna make it more human, but let's say the goal was to make it more human.

What has to happen still or what would be a dream to happen? Still?

Ankit Jain: 

I don't think becoming human is a dream. I think the goal eventually is to get rid of the phone call. The way we measure success, Mike, our mission is to create time for healthcare. So on one side, we are reducing the amount of time those making the phone calls are on the phone, 'cause our

machine's on the call. But we also measure how long we're on the call because there's another person on the call.

 if we can

drive more efficiency on their side without them doing anything, just by understanding that United healthcare systems like an agent at United Healthcare can go faster if you ask about prior auth before copays.

the agent at Humana goes faster if you do copays before prior

auth, just because of

how their systems [01:02:00] are laid out.

And that's

a win for those companies when we call them. And sometimes those discussions are explicit 'cause we know the team's there and sometimes the system is able to experiment, learn and get better.

And to me it's about eventually getting rid of the phone call, but along the way, expanding the knowledge base to cover more and more kinds of phone calls, standardizing them and making it easier for us to eventually remove them.

Mike Koelzer, Host: My dad, God rest his soul, but he always said that. He was never concerned about chain pharmacies opening up around the store because we were different from them. We were a different beast. He said he would always be concerned about a little independent pharmacy coming across the street. How could you put yourself out of business with this company?

And

 What would be something that would even supersede the APIs talking together or something? 

Ankit Jain: Oh I think the question becomes when those APIs are ready, [01:03:00] what's the role that we play in that world? Are we the ones enabling the different constituents in healthcare to exchange that information? Do we become that router of information or do we get taken out of the whole ecosystem, right? Because everyone just connects directly with each other.

If you look at every industry that has gone through this kind of transformation, there have been companies that have to build new infrastructure that take them into the future. When I think about people saying the administrative cost of healthcare is very high and it's increasing. I wish there was an insurance company that was cheaper so more people could afford it. That's not gonna happen overnight. It's gonna happen as new rails are built.

And eventually the underlying infrastructure will allow for a newcomer or an incumbent to offer a product which is significantly cheaper.

[01:04:00] Now, we saw this happen in the financial services industry. It took 25 years of building new infrastructure before Robinhood came out

and said, I will allow free trades. You don't

have to pay per trade. that then had a snowball effect forcing every incumbent to go down the same place as well.

They can't charge the five or 10 or $20 per trade fee because, then why would you trade with them? You'd go to Robinhood.

I think something similar will probably happen in the next 10 to 20 years in healthcare where these new rails are being laid by.

 Dozens, if not hundreds of startups 

somebody will say, cool.

 if I put all these pieces

together, 

I can build something that is way more

efficient. And that will be a great product to bring to market, but it will force the incumbents to do something similar to.

Mike Koelzer, Host: ANke, golly, thanks for joining us. Really cool stuff because I know, and our listeners know that if we've got 10 people in a period of time that come in, complain at the [01:05:00] pharmacy, they're not complaining because of something healthcare wise that they didn't like.

They're not saying typically that my cast has been on a day too long, or I wish this medicine tasted a little bit different. Something like that. I mean, those are one out of a hundred. What we hear is. All the complaints of, I was on the phone with so-and-so for so long. And they hung up on me, and then this happened and that happened.

And so that's the pain in healthcare. So boy, you're a hero if you get rid of some of that for people. And such wonderful work you're doing and I'm looking forward to seeing how you guys progress.

Ankit Jain: Thank you so much, Mike. Absolute pleasure. Really enjoyed this conversation. ​