Revolutionizing Healthcare Management with Advanced AI Technologies

Unlock the secrets to transforming healthcare with cutting-edge AI technologies in our latest episode featuring Alex Zikoff, co-founder and CEO of Thoughtful AI. Discover how Alex's journey from aerospace ERP systems to democratizing AI for mid-market healthcare providers offers valuable insights into overcoming significant challenges like the slow adoption of cloud services. Learn how advanced AI agents, powered by OCR, NLP, RPA, and large language models, are revolutionizing revenue cycle management, allowing healthcare providers to focus on enhancing patient experiences and tackling higher-value tasks.

Explore the transformative power of OCR technology and Retrieval Augmented Generation (RAG) as Alex breaks down how these innovations optimize revenue cycle management. With an emphasis on accuracy and reduced human intervention, Thoughtful AI's unique approach promises better health outcomes and financial stability for providers. Dive into the intricacies of data security, the role of serverless containers in safeguarding PHI, and the seamless integration of AI within existing healthcare systems. Understand how a master automation plan can turn standard operating procedures into efficient AI agents, empowering healthcare providers to make more strategic decisions.

In this episode, we also delve into the broader benefits of AI and automation in healthcare, from reducing staff burnout to improving job satisfaction. Hear the success story of Signature Dental Partners and how implementing AI agents led to reduced sales outstanding days and scalable growth without additional hiring. Looking ahead, we envision a future where AI engineers can deploy solutions on the same day, enabling healthcare providers to efficiently manage their revenue cycles and enhance overall patient care. Don't miss this enlightening conversation with Alex Zikoff and the potential future of AI in healthcare.

Transcript

Transforming Healthcare Providers With AI

Speaker 1 0:02

Hey everybody , super exciting chat here with Avera Health . Today we're talking about transforming healthcare providers into AI-powered organizations . Really important topic , alex . How are you ? Great Evan Appreciate the time appreciate your time and you being here really intrigued by your mission and your purpose so important . Maybe introduce yourself and the team at Thoughtful AI and how you're planning to transform healthcare organizations .

Speaker 2 0:37

Absolutely , I appreciate that . Great to see you , Irma . Alex Zikoff , co-founder and CEO at Thoughtful AI , Really started my journey building ERP systems for aerospace companies about 15 years ago . I then moved to Japan for two years to help a company set up a large aircraft organization . After that , I was sponsored to go to business school . I went to Berkeley for a couple of years studying entrepreneurship , design thinking , venture capital . After that I basically went back into aerospace , went back into AI enabling technologies . We can talk more about that .

Speaker 2 1:07

Ai is a big umbrella term for many different technologies . Started building what we called bots back then . We called them AI agents now about eight years ago , and I just sort of had this vision that if you actually count up everyone that operates software , moves data on an annual basis , that's about $7 trillion of labor . That's just pushing data , and I sort of kind of correlated that to . If you look back to the 1910s , Henry Ford's assembly line you see actual people on the assembly line . You fast forward to the 1980s . There's robots . That took 70 years , and in software it's probably going to take about 10 to 15 years , and so with that , I left Deloitte in 2019 with the mission to democratize the technology Very expensive , what we're delivering in the Fortune 500 context , and our mission was to basically bring this to mid-market health care providers that don't have access to this technology To help them enable incredible outcomes save money vis-a-vis cost to collect day sales outstanding , basically getting paid more for the services they're providing to patients and to date , we've raised a lot of capital from venture funds to enable this mission .

Speaker 2 2:16

We're really excited to connect today just to learn , you know , to share what we're doing and what we call apply to AI in healthcare . We're actually delivering AI solutions in production seeing the roi that everyone's sort of talking about in the news and we've got some really great case studies on that and excited to get to connect today and share what we're doing in that space fantastic before um you tell us how you solve these issues in healthcare .

Speaker 3 2:44

Maybe you can outline some of the biggest challenges that you're tackling with thoughtful AI approach .

Speaker 2 2:53

There are so many challenges in healthcare . To begin with , I think one of the largest challenges is healthcare technology has typically been sort of 10 years in the past , and what I mean is right now you see a lot of healthcare providers just getting to the cloud , which many Fortune 500 are already in the cloud , so it's really hard actually to deploy AI if you're not in the cloud , and so that's sort of challenge one . So we see a very different spectrum for AI readiness across the providers we interact with . For those providers who are in the cloud , we are delivering AI agents . Basically , the providers are hiring our AI agents to augment their revenue cycle team so they don't get backlogged with the work . So what's an AI agent ? An AI agent is a combination of many technologies OCR , optical character recognition , nlp , natural language processing , rpa , robotic process automation and LLM large language models , most featured by ChatGPT going live 18 months ago . When you combine all of those different what I look at them as sensors into a container , you can actually replicate a human up to 100% , and so it's really the orchestration between the different technologies and how they integrate with the healthcare applications and portals that drives that value . If you just deploy an LM , that's great , but you only might be getting a 30% productivity increase , but when you combine it with RPA and OCR you can get to 100 . So we've been at this for four years , architecting a platform that builds AI agents . So we actually have AI that builds AI . So we're sort of in this like second meta layer where the AI constructs it so we can deliver this faster .

Speaker 2 4:35

For us , the thing is time to value . So if you hire an employee , it might take 90 days Recruit , hire , train . We can actually deliver an AI agent in the same time it takes to hire and train an employee . So now it's shifted , this whole paradigm of you can actually hire software to do the work . We call this services as software , not software as a service .

Speaker 2 4:55

It's a completely different paradigm shift . I'm not training you , irma , to do the work with a tool , I'm actually just delivering software that does the work . So you can go focus on , in healthcare , the patient experience . You can focus on exceptions . You can focus on higher value work , not just clicking buttons and denials , denied claims and trying to get collect the money that should be collected . So it's really going to be great for the healthcare industry at large and we're really excited because this is a multi trillion dollar problem we're going after and it solves . I think one of the most fundamental things is our health institution here in the US , which we all know is broken the model across payers , providers and patients , and so really just streamlining the data across all three of those constituents to ensure that patients are actually getting better health outcomes and providers are getting paid more for the services they're rendering .

Speaker 1 5:48

Wow , really important work . Such a yes , that's a blockbuster insight there . And let's pick one of those areas , an important one revenue cycle management that you talk a lot about . Important because hospitals are going broke and it's a really sad situation . So , specifically around RCM , how do you help the healthcare system ?

Speaker 2 6:11

So we think about the revenue cycle puzzle as sort of key pieces . You've got your front end revenue cycle , which is eligibility checks and prior authorizations before the patient even gets the service . So basically checking that the patient and the provider is going to get paid for the services rendered . Then you've got sort of the mid revenue cycle , which is claims processing , scrubbing before that gets sent to the payer and then you've got the claim comes back . It's denied . You've got claims denials management . Last up they get paid . There's the payment posting aspect and that's a huge team .

Speaker 2 6:44

So collectively across those core processes and revenue cycle , let's say you're a hundred million dollar in revenue provider , you might have 50 to 60 full time people collecting money for you . That's their full time job , versus in other industries that might be three people . So when you scale to it's a linear problem . So let's say you want to go from 100 million and patient patient revenue to 300 million , you're probably scaling that revenue cycle team to from 50 to 150 . So instead of having to scale heads , you install agents . You basically turn everyone on the revenue cycle team into superhuman . So instead of them being able to process 10,000 claims a week as a , as a team they can process 100,000 claims a week . That's the beauty . It's actually more of a human in the loop system of leveraging the technology to do more . And what does that mean then ? In outcomes ? Reduction in day sales outstanding . So instead of not getting paid for 60 days , getting paid less than 30 , that's huge from a cashflow perspective .

Speaker 2 7:43

There are providers that are getting denied up to 10% or more of their claims . So they bill a hundred , they collect 90 , 10 millions just sitting there and denied claims and they're just it's going to sit there because they don't have enough team members to actually go through that denials process . So we can actually help get seven , 8 million of that back . So you're collecting closer to 97 , 98 on a hundred million that back . So you're collecting closer to 97 , 98 on a hundred million . And then really , lastly , is cost to collect .

Speaker 2 8:09

It can be upwards of five , six , 7% to collect on a hundred million dollars with that revenue cycle team versus less than 1% in most industries . So you've got six , seven times more money going towards collections process than actually that money going towards doctors collections process than actually that money going towards doctors , patients . You know better R and D for better solutions . So we look at this as if you look at the entire us healthcare system . Three and a half percent of our us GDP is spent on back office healthcare administrative work like this and that's going to be growing to five 6% of our us GDP .

Speaker 2 8:43

Okay , and it's something that people aren't going to be solving because people primarily if you look at revenue cycle employees now their typical age is somewhere between late 30s to late 60s Gen Z is not going into data entry and repetitive work jobs like this . There's going to probably be a 6 million job gap with revenue cycle and healthcare admin work . So this is mission critical to ensure the rails between all three of those stakeholders the patient , the provider and the payer are moving fast and efficiently .

Speaker 3 9:21

That's a great use case you just laid out . Let's step back into the technology behind those solutions . You gave us a preview , kind of like all these different components that comprise AI . Can you tell us specifically which AI technologies power your automation solution ?

Speaker 2 9:40

Yeah , so we have built . We basically use all the technologies . We built our own RPA solution , which is Python-based code . So think about this as the way you interact with the system now as a person is you log in via a GUI , a guided user interface . If there's not an API , which is just direct-to-direct system , we can log in just like a person to get that data . So right now , if you're a revenue cycle employee and you're going to go check an eligibility for Florida for Medicaid , you might log into the state Medicaid eligibility website to check the verification and then bring that back into the HR . We can log into the website just like a person , comply with all the terms of service , Go , get the data , bring it back . That saves 10 , 15 minutes , but we can do that at infinite scale . So instead of a person having to log in and get the data , we use that . That's sort of the RPA technique .

Leveraging AI for Revenue Cycle Optimization

Speaker 2 10:33

Now let's think about OCR , optical character recognition . We have our own solution here . We parallelize the ability to process thousands of documents at scale . So when we're reading payer contracts or we're reading different formats , we're just collecting that data that people are using their eyes to ingest into systems .

Speaker 2 10:52

What's been great the last two years , 18 months is the breakthrough with large language models . So before that came , you know if something would break then a human would have to fix it . On our team Now we leverage large language models to fix when we don't see , let's say , a payer mapping rule in one of our documents . What that rule set is , it's a bunch of if-this-then logic . We can use a technique called RAG retrieval augmented generation to predict the next mapping rule with about 99.9% accuracy . The person on the revenue cycle team can train that retrieval augmented generation to get smarter and then the AI agent can make an intelligent decision when it runs into a block . This has been the unlock . It's truly acting as if a person ran into a block and said what would I do ? And then it's actually accurate enough to make the decision going forward .

Speaker 2 11:41

So when you combine all of that and we put it in a serverless container so we don't store any PHI data , which is very important no data is ever stored . It's like a . It's a serverless container that spins up , does the work and then destroys itself . We only store metadata on the analytics on the process that you , as a potential executive , would want to know about your business and the revenue cycle maybe where there's revenue leakage , where different locations are not collecting as well as other locations , where different locations are not collecting as well as other locations . And that's the layer we get . We record every single click and action that the agent performs . Imagine if you had your 150 person revenue cycle team and every day they reported every single click they did . We have that level of transparency and they can run really amazing graphs and figure out where the process is maybe broken or where we can optimize . And that's what we train revenue cycle teams to do now is look at the analytics and then make more strategic decisions , not more operational decisions .

Speaker 3 12:36

Very interesting .

Speaker 1 12:38

Fantastic approach , yeah . So anyone who works in healthcare IT knows the challenge of integrating new services and applications and tons of technical debt and spaghetti code and et cetera , et cetera . So what are the integration challenges you face when going into an existing healthcare system and how do you sort of overcome those ?

Speaker 2 13:02

So we have a unique model where we don't actually sell our platform and then the provider builds the agents . We just build the agents and license the agents directly . So we have years of training with our team on how to go in and first we start with process reengineering . So we'll look at the as-is process and we'll actually say , hey , we actually have a best-in-class way to do this . Are you open to moving to this future state 2B process ? All providers say , yes , that is going to save us , you know , from 100 steps to 50 . Then we say we create a master automation plan , which is a technical document and code that basically says , well , we're going to turn your standard operating procedure into an AI agent that can make decisions across that entire process . We then put that through all of our technology in the backend , which is about four different AI generated products that build this basically master automation plan into an AI agent , and that's a bunch of different architecture decisions we made

Automating Healthcare Processes With AI

Speaker 2 13:57

.

Speaker 2 13:57

We kind of think about it like Legos . You construct these things called work blocks and a work block . Think of it as like the login to that state Medicaid eligibility website I talked about . We've already built that . So every time the agent uses that and many providers use that same work block . It just uses that same architecture to log in . So we have a lot of repeatability and so we're just kind of assembling Legos at this point , because we have thousands of work blocks in our code repos and then we can deliver in 90 days . It's in production .

Speaker 2 14:24

We have a front end called empower . That's where you interact with your ai agent . So that's that human in the loop , you you basically can click it or it can run on a schedule and it just works like the work you would have been doing . And then you look back and you kind of say , well , there are some exceptions I need to process for the day . So if it took , you know , to run verifications the team the entire day , they can run verifications in five , ten minutes now and they can use that time to run verifications the team the entire day . They can run verifications in five , 10 minutes now and they can use that time to run exceptions , train the agent to get smarter . It's incredible . So we're freeing up , you know , 80% of people's time plus to really focus on well , where's the system broken and how can we fix the system , not just operate the system .

Speaker 3 15:07

All right . Well , let's dive deeper into Thoughtfulai's approach to data security and compliance in healthcare settings . You already told us as far as PHI protected health information . You don't store anything , so that's good . But you did mention a few times there is a human in a loop and anytime there's humans involved , there is more potential for security breaches and things . So tell us about your approach to data security and compliance in general in healthcare .

Speaker 2 15:38

Great question . It's a huge topic in healthcare . There's been some cyber attacks recently that have put the industry at guard , have put the industry at guard . So one we that master automation plan , is our alignment with the IT group of things that we'll let the AI agent do versus what we'll let a human do . So , for example , some in some industries we can't actually submit all of the claims , so we'll prepare the claims but then a person will review them and then submit to the actual , to the payer or the clearinghouse . And that's an important step is there's got to be human-actual viewing before there's actually a live production submit . And so we built that based on the requirements for the provider .

Speaker 2 16:18

Second , again we've talked about we don't store the data . So we're HIPAA compliant , SOC 2 , type 2 , and we don't put large language models in your data set . All we're using large language models to do is make intelligent decisions when there's breakage points , but we're not saying , hey , we're going to train all of your PHI data to make a recommended thing . That's a very scary concept because large language models are known to hallucinate and we don't actually know how they work in neural networks . We just know they give us outputs , but they're a little bit of a black box neural networks that we just know they give us outputs , but they're a little bit of a black box . So we try to take the black box out of AI and just say , hey , what we're using a large language model here to do is predict the next if this , then that rule , and then does that look good ? Yes , and then make the next step . So the human still is involved . We're not just letting the AI take over , and so , especially in regulated industries like healthcare , it's super important that piece that we're not just letting rogue decisions happen , because I think there's really a potential snowball effect there Because it's being trained on maybe the wrong decision it made . Without that human loop training it , it's going to produce outputs that are incorrect , and then we're going to see more denied claims , so that would compound the wrong way outputs that are incorrect and then we're going to see more denied claims , so that would compound the wrong way .

Speaker 2 17:30

So , applied AI in healthcare is really about doing the work getting in with providers , understanding their pain points and realizing that every provider is unique . It's not a one-stop shop system . You actually have to go in and really listen more to what their pain points are and really build a unique solution and platform that's going to solve that provider's specific pain points . I'll tell you , we work across a lot of specialties dental vision , behavioral health . They all operate differently . All the different CPT codes , they all have different rule sets . You can't just go in and sell a behavioral health solution to a dental practice . It's not the same thing . What is the same , though , is the processes are pretty consistent . Everyone does have some form of eligibility or prior auth , some form of claims processing and some form of payment posting that should be automated , can be automated and really just understanding where the gaps are in their current process . You know some providers we see , should you know , let's say , they're running 10,000 verifications a week .

Speaker 2 18:31

They actually should be running 20,000 , but they don't have a large enough team , so we're actually allowing them to run verifications on every patient versus just the amount they can keep up with . So that's the superpower of this , and that's why we're really excited about what we're deploying right now .

Speaker 1 18:45

Live with healthcare providers we're deploying right now live with healthcare providers . Wow , very exciting . Congratulations

Impact of AI on Healthcare Revenue

Speaker 1 18:55

for that . So clearly the financial benefits the ROI is there today , but what are some of the maybe non-tangible benefits of your approach to AI and automation ? Things like job satisfaction or productivity . What other others that I might be missing ?

Speaker 2 19:08

productivity . What other others that I might be missing ? Yeah , that's . I mean , if you look at , there's two main core issues in revenue cycle . It's the reimbursement issues we talked about , and then there's staffing for revenue cycle teams .

Speaker 2 19:18

Churn can be up to 40% on an annual basis . So , again , you hire 100 people . 40 are turning over every year . That is just if you're in recruiting , that is just from you know , if you're a recruit , if you're in recruiting , that is just an , a leaky bucket , that just . It never ends . And then so you're never carrying over . You know , within three years you might've turned over your whole team . So you're constantly losing that tribal knowledge that people have . And so , without the documented process , it's just constantly .

Speaker 2 19:46

Teams are trying to hire enough catch up . There's not enough people to hire for these roles . I mean , if you interact with any revenue cycle leader , they are . It's the same as building a tech startup . They are working 80 to 100 hour weeks . They're just trying to catch up . You can't we come talk to them and say , hey , we could 10x your team , and you know it ends up being like , yes , we want all of that because we are behind , and that's why the revenue cycle team is , I think , one of the most important teams at a healthcare company . It's because they're getting the money so they can generate the profits needed to make sure that company can grow . So it's definitely a staffing problem . The churn rates are incredible and we try to listen there first , as what are the human problems that we can help with ? Ai ? The staffing problems , and that's where we typically see the highest ROI . Highest ROI is , again , not having to burn your team out , not having people to work 80 , 100-hour weeks .

Speaker 3 20:45

That's the burnout problem right there . Yeah , burnout is definitely a huge issue in healthcare specifically , and you mentioned staffing shortages . So it's not just on the backend of the office where we have shortages . There is a shortage of physicians . So the front end , the front lines of healthcare , so the front end , the front lines of health care . So you offered , really told us about a variety of different solutions you have . Can you share some specific , maybe success stories ? I don't know if you can share names of your clients , but kind of give us maybe impact to patient care or some other specific impact from your client base .

Speaker 2 21:25

Yeah , I'd like to highlight one of our VPs of Revenue Cycle , kara Perry . She works at Signature Dental Partners , a fast-growing dental services organization , and when we met Kara beginning of last year , pretty much everyone starts with a core problem and we delivered one , basically AI agent . Everyone starts with a core problem and we delivered one basically AI agent and it ended up moving sales outstanding something like less than 20 days from like 60 or 45 , whatever it was . It was like a 3x improvement within two months . She went to her board of directors and her private equity firm and got funding to basically go from 1 to 30 agents in one year . Now let's think about the holistic solution . All of her team is now practiced using these AI agents . So they used to be able her team used to be able to serve these pods five locations . Now each pod can serve 12 , plus I think we're going to get them up to 15 . So three extra locations they can serve and that's important .

Speaker 2 22:21

And they're also working the same or less . So less burnout , more locations . Now Signature Dental can really scale . They can really grow that business to three X year over year if they want , because they're not having to worry about hiring 100 more revenue cycle people to collect money . That is an incredible outcome and we're driving really it's driving growth at these healthcare providers so they can actually have the core tech ops needed to know they can collect the money that money's coming in and then they can scale Again . They can pay the dentist more and attract better talent . What happens then ? Then we have more incentives to go back into healthcare because we can pay more to doctors and providers , and that's really the case is we can have more money back in the system . We can change really the negative trend of people not going into health care .

Speaker 3 23:07

Right .

Speaker 1 23:09

Nice , nice mission . So I almost dare not to ask what we can expect from you over the next two , three years , given you're already sort of living in the future today . But what is your grand ?

Speaker 2 23:26

vision over the next few years for your role and that of AI agents . We are working on some R&D projects now that I'm very excited about , so how we make all this work I mentioned is that we have AI that builds AI . Well , some of the AI is not fully human capable , like our healthcare products , but some of the products we'll call it like an AI engineer we will probably be fully human capable in the next three years with an AI engineer , so we'll be hiring our own AI engineers internally . If you think about that , it's kind of crazy . We we hire people AI engineers now but we're going to be hiring our own AI engineers that we're training and that will just allow us to build incredible products and services for our customers . Like we can't even I'm trying to like model that out now .

Speaker 2 24:08

It's even hard to model the next five years , but let's just say we can deliver an agent in 90 days . Think about , like Amazon they work to get same day delivery we might be able to replicate that agent deployed in the same day Same day hiring , same day deployment , same day production , and that could be in the realm of possibility in the next three years , and so that's our radical mission right now we're hiring a lot of people , a lot of really smart people , to solve these problems . I'm fortunate , as a founder , co-founder and leader , to get to work with everyone on these problems and serve these customers . So it's an exciting if you're in AI right now problems and serve these customers . So it's an exciting if you're in AI right now , especially in applied AI . It's an exciting time to be a founder and entrepreneur For sure .

Speaker 1 24:49

Say the least . Well , irma , I know you were a software engineer of 20 plus years , so maybe we can clone a few of you to get more done at Avira Health . That would be quite an objective .

Speaker 3 25:02

Let's work on that together . There we go .

Speaker 1 25:05

And Alice , what are you up to personally , professionally , in the next couple of months ? Hopefully you have some R&R time , but what's on your personal radar for you and the company at any events or travel or trips coming up ?

Speaker 2 25:17

Yeah , I'll be speaking at a couple of conferences . There's one coming up in Boston in June . I'll be at an RCM conference . There's one in San Diego I'll be at . So I'm on the road a lot speaking , evangelizing the AI agents in healthcare . We've got a big developer summit in Brazil , so we'll be down in Rio kind of sharing our R&D products , research that we just talked about and , very excited , we're just hiring a lot of people this year . So really focused scaling the team . And that's it pretty much . Yeah , I try to rest Like I said .

Speaker 1 25:48

I just got a kitten last weekend , so that will be zero rest , but a nice try . And yeah , thanks so much for being on the show . Amazing mission , amazing vision , and we can't wait to see the story unfold . Thanks , alex .

Speaker 2 26:05

Awesome Thanks .

Speaker 1 26:06

Emma .

Speaker 3 26:06

And thank you everyone for listening and watching take care .