Episode Transcript
[00:00:00] Speaker A: Hi, my name is Tim Cohen. Welcome to Vital Conversations presented by Focus. Here we talk to leaders across the healthcare industry about what's going on in the world and today I'm pleased to have an old friend who we've crossed paths for seems like 100 years, Dave Haight. He is the principal of Synergy Consulting.
Dave, welcome to Vital Conversations.
[00:00:22] Speaker B: Tim, glad to be here, nice to reconnect again. So looking forward to the conversation.
[00:00:27] Speaker A: Dave, give people the 2 second version on your background and resume because I will create a lot of context for some of the things we're going to talk about. You've been in the middle of healthcare IT data EMR for a long long time.
[00:00:42] Speaker B: Well thanks Jim for aging me but there's a little bit of truth to that. So I actually started my career on the provider side of the business so ran clinical and financial and administrative operations, both an acute inpatient ambulatory across multi geographies here specifically in the us did that for a big portion of my career. I will not say how long since that will probably age me beyond belief. The second part is in life sciences work. Moved out to Chicago, did that for a number of years, marketing consulting. Two kids wanted to move back. I'm a native New Englander if you can't tell by my accent.
I live in the New Hampshire area, Seacoast area and got involved with healthcare IT a little over 20 plus years ago. I will not say the exact date. That's how we got connected Tim and have done everything in that space in terms of implementations, marketing, sales, integration I would say went through. I'm a legacy idx, worked for GE Healthcare, spun us off, became part of Athena, went through three private equity transactions and got to do a lot of work. I would say probably the last seven, eight years. I would say it falls into four buckets. Bucket one would be your end to end consulting or as I say credentialing. Two remit on the business or revenue cycle side on the clinical side, cradle to cure and that's pillar one. Pillar two would be around managed services, managing a lot of our clients ecosystem both from an application IT perspective and kind of putting that mesh together for lack of a better word, three ways around analytics, taking information from different systems, putting it together and making that useful for our clients. The fourth one, which I think is a topic we're going to kind of get into a little bit today, got involved a lot with new solutions and automation. So it used to be rpa. Well let me back up, I'm going to date myself Scripting, rpa, machine learning, AI, allergenic AI and have learned the trials and tribulations and I'm a big believer in fail first attempt in learning and or some cases maybe second attempt in learning or sail and so have some experience with that. I would say probably the last two years, you know I've really falls into two buckets. Bucket one I spend a lot of time with your traditional healthcare IT vendors, you know, looking at different companies, their portfolio, what makes sense, what are the adjacencies, go to market type of pieces. The second bucket's a little bit involved with life science and it's really kind of interesting in the life science area what AI can do and that's pretty promising. We can probably get into some details around that as well too and then do a lot of work with in the PEBC world as well too. So it's good enough to keep me busy and out of trouble.
[00:03:25] Speaker C: Well it's that long again, we won't
[00:03:29] Speaker A: date it and diverse backgrounds. Why I wanted to talk to you today and specifically I want to talk about sort of this non sexy part of AI. So everybody's talking AI, it's all over the place. Going to change the world. It's not going to change the world.
Even large health systems which were notoriously slow IT adopters are kind of on the bleeding edge. But there's a bunch of things that every scaled organization, if you're not digitally native in Silicon Valley, if you've got any kind of scale, any kind of legacy data and applications, but particularly if you're a healthcare organization and you've got HIPAA and other regulatory things around your data.
There are a lot of places that you could get yourself in trouble just by jumping into the flavor of the month on whatever hot new AI thing
[00:04:23] Speaker C: it is in the moment.
[00:04:24] Speaker A: So let's talk about that a little bit. What's your take on where we are today in and maybe both sides of the equation, Sort of the people that are providing solutions to healthcare, the extent to which you think they understand how we're different, what our rules are and the people particularly in provider organizations that are trying to buy and implement these solutions, their understanding of the unique risk challenges, compliance issues for us as we
[00:04:56] Speaker C: bring data into provider, AI into provider.
[00:04:59] Speaker B: Well you put it into a couple different buckets. Let's see if I can, I'll answer one at a time here. So let's look at it and I could give some examples. So let's look at it on the provider side of the equation. Right. A.I. you know, let's, I think, let me take the plane up a little bit. If you look at the macro conditions that are really driving AI and automation in general, one labor shortage, clinicians specifically. But if you look at the business office, rcm, you know, that's also aging out as well too. A lot of companies offshore, you know, a lot of those operations, either their it, their applications, their revenue cycle management, there's a lot of concern around cybersecurity. You know, we've had two data breaches, two major data breaches I think in the last three years. Change healthcare, very well publicized, Ascension healthcare, very well publicized, impacting patient care. So obviously there's some breaches there. And so if you're looking at from a provider perspective, no, they have a very big ecosystem and you know, they might have one single emr, but they're also using other systems. They have a workforce management solution in there, they have a CRM solution in there. They also have other point solutions in there. And so you know, I think you bring up a really good point. There's a lot of what I call points of failure or success as part of that, you know, you're connecting to pharmacies, you're connecting to labs, right? So there's a large ecosystem but there's not one system today that takes all of that. So I remember the old spaghetti diagrams where you look at that interface engine, you look at the EMR and all the systems and stuff like that and I think I'll pick out EPIC in a good way. They've, they've really reduced a lot of those bolt on solutions. But if you still look at that diagram, there's still a lot of other solutions. You know, you have managed care and everything else. I think, you know, for me in terms of, in real simple terms, you know, big piece is the data. Even, even in within systems that are contained is the structure of the data or the non structure of the data. And when you add in the complexities of those other systems and how they pull the data out once, so then you get into the whole security piece of it in a very big way how that information is deployed, how is it used especially with PHI and hipaa.
And then in addition to that it only works if the data is structured. So you know, you, what I worry about is do you have each of these AI solutions or point solutions structuring the data differently to make it work for their AI tool. So versus a uniform piece and obviously data lakes have been around for a long time and then how do you get into the governance of it as well too. And then some of the AI specific health system I'm thinking of, they used AI, but it didn't have all the data. So not only do you need clean data, but it didn't have all the right data. So in this case, if you had an inexperienced clinician, it has some pretty severe consequences. And so this one health system I'm thinking of, they actually had to, and I put in air quotes, had to put in verbiage into it saying please validate this first before you move forward and put checks in there because they were concerned that it could have adverse effects to patients or you know, how do you know you got the right data set? You know, how do you spell my name? Can go Dave David, although I may have an R and that sort of stuff. So even that basic information I think from a provider perspective is a challenge. And it's the old adage to garbage in, garbage out. And in this case data, you know, having that in a clean manner is very important. So I'll pause there and see if you have any questions on that.
[00:08:17] Speaker A: Yeah, it's interesting because you know, if you go back to where you know there's an, there's an ehr, whether it's, you know, big epic thru system or a physician practice is using an ambulatory module and you want to integrate a bolt on application. The way that process historically worked is some software engineers got in and mapped field to field. It was, we do a lot of that work at focus. Right. It's hard, it's labor intensive, there's a lot of quality work.
But at least you kind of had eyeballs on the field to field migration of the data and you knew how you could check for quality and how you can do validation.
Now we have like some level of mythical magic in the large language model, right. That we don't have the software engineer going field by field, we have the AI doing something. But on the back end, I'm a clinician, I need to make a decision for patient safety. I need to make sure I do no harm. I'm in rev cycle. I need to make sure not only do I capture the data, but I don't inadvertently commit fraud. I mean we've got a lot of places around health care, a lot of. Yeah, and it's a little more serious than making videos of kitties if the AI gets it wrong. Right?
[00:09:35] Speaker C: That's right.
[00:09:36] Speaker B: Yeah, yeah, you're absolutely right. I mean in the old days you do if you were doing a conversion or an interface Right. That you had a very robust process engineering end users to make sure that the data flowed correctly, came across appropriately, making sure that it put everything into the appropriate field. And I think, you know, and there's a lot of checks in that it was very much an iterative process. Right. So in the old days it used to be unit testing and then you do modular testing to make sure it went through the whole workflow and then you do a production simulation to make sure that you could do it at scale.
And you know, I think that still process applies today. And I think the challenge is, is the AI can do the data mapping, but you still need that human in there to QA it. Right? And I think a lot of the automation processes still needs that, that human component to QA to make sure that it works appropriately through the different things and also to look for those points of failures. And I think the biggest concern that I hear from folks is what if you get a runaway agent, right? What happens? How do you stop that? Do you have the appropriate steps in there to slow that down? Or if you're sending information to payers, for example, they're worried about the volume and a lot of these things, the infrastructure can't take that much volume. Let's say I can do 50, let's say I'm a coder, I can do 50 charts a day. Now the AA agent can do 200. But the pass through or the pipe to, through a clearinghouse to the payer may not be able to handle that if they have a lot of people doing that stuff. And so those are the things that have to be put in place. And then worse, because of the speed and how they do this, if you have a runaway process or job or an agent that now gets modified, you could cause quite a bit of havoc. And so really making sure you have those guardrails I think still apply. And I hate to sound old school, but you still need to go through that ITERA process as well.
[00:11:21] Speaker C: So well.
[00:11:21] Speaker A: Well, let's go old school for a little bit because here is, I think the reality when a lot of IT executives are talking to the business executives and the business executives are reading the headlines and you know, somebody vibe coded this over the weekend and stood it up on Monday morning and they're walking in and pounding on the IT team like, we gotta do this faster, faster, faster. How do folks that understand what you just articulated communicate back with folks that are getting pressure from wherever to deploy these things, to go fast, to, you know, take all these cool stories we hear about and how do they have that conversation to say time out, we get you, we're trying to go here. But because we're in healthcare, we do
[00:12:07] Speaker C: have some obligations to slow down, test, validate, lean, normalize. All the words that data people think about is kind of getting lost in stock position.
[00:12:17] Speaker A: How are you?
[00:12:19] Speaker B: It's a good conversation. I did this a few years ago with a client we were looking at.
It was around claims processing specifically. And they had, you know, a hundred plus people that were working in that and they were looking to, they were under severe financial pressure to reduce cost. And you know, they were looking at all avenues of their business. And obviously AI was big, automation was big. How do you do that? And we very frankly had a very frank conversation around it saying, you know, this is no different than the human component, except we need to make sure we get it right. And the way we approached it is really making sure we had the appropriate, you know, standard operating procedures in place. Right. And so just like you would as a human, and making sure that the agent understood those operating procedures and then making sure that we had very rigorous testing cycles to make through. And we, we made sure the client understood that. Because even with a human being doing these process right, the stakes happen. You have a bad hire or somebody has a bad day, somebody calls out sick, you want to make sure you do that. And then any, the other piece too, that was important, anything that fell out of that, you wanted to make sure that it was looked at. Right. So you wanted to make sure that if it followed the SOP perfectly, you had it run, you tested it, if there's any fallout from it, it went to somebody to evaluate or to look at it. And I think when you talk through, you know, my experience for this particular group, very large group, they had 2,000 providers in their, in their network talking through that process. And we took it more from a workflow perspective than we did from a technology perspective. Because at the end of the day, it's not the technology that keeps me up at night, it's really the people and making sure that you have the appropriate policies and standing operating procedures in place. And you could have a very meaningful discussion around that. And people are very sensitive about security data breaches and runaway processes. And I think for me it's been a relatively easy conversation.
[00:14:04] Speaker A: Well, let's take those, those two ideas, thinking about them like humans and thinking about workflow.
I'm sure it freaks some people out
[00:14:16] Speaker C: when the new AI visionaries describe their agents in human language, but we're hearing More and more about digital employees, it
[00:14:26] Speaker A: sounds like you're saying that's not a
[00:14:27] Speaker C: bad construct to think about hiring and onboarding and managing your agents just like you would an employee.
[00:14:39] Speaker B: Yeah, I would say that's it. Yeah, it's the same concept. I mean we. The way the approach that I've done it with different companies is to use that same construct, that same framework because in essence they are doing the work that people would normally be doing. You're just automating the agent and you're also giving them those critical thinking skills as well. And you want to make sure that they understand that and do that. Right. And you know, the example of the health system you know, I was working with on the clinical side, you know, they would recommend a course of treatment, but they didn't have all the data. And so did they have standard operating procedures? Yes, but it would miss certain things. So you know, maybe I'm diabetic. They didn't, you know, they would say, geez, you know, Dave's not had his A1 test. He hasn't had this done. He hasn't had this done when the reality is I did, but it was in some other place. And they might prescribe me the wrong course of treatment, which could be very adverse to my health when I was actually taking that. And they might, might actually take me down a different path. And so it's very important that you have those safeguards in place because even if you follow the steps, you may not have all the data. That's one example or two.
Let's pick on. I'll pick on Rev cycle. I'm thinking of the case that I just mentioned to you with that large provider payer information changes, something changes. And so does it have that critical skills to do that with? And so we had to train the agents for an example. We did claim status. And so we go out to the payer websites, they might move it from side A to B, page three or four. You need to have that critical thinking. And this is what I'm looking for. Maybe they even made a name change. And so you need to make sure that you test for that and then if there's changes and you need to know, have some governance over that stuff in order to make sure it works as well too. And one of the best ways I've seen organizations do this is if they notice the change, it goes to a list and says, you Tim, you probably need to look at this. We just want to make sure this is right. Just kind of doing that basic QA Validation is really important.
[00:16:27] Speaker A: But again, we actually have that model, right.
In an ambulatory setting. We put a patient in a room and Ma goes in and collects some information before the physician shows up. When the physician shows up. She doesn't just take that information as gospel.
No, she brings the higher level of training, looks at what's been done, picks up some additional information as she works through to her diagnosis.
[00:16:54] Speaker B: And she also, she's doing an intake too, right. And she may not, you know, I might say, geez, are you on any meds? I'll pick up my mom. My mom will say no. Well, what that might mean is my mom didn't take her meds today.
Right. She should have. Right. She should have taken her blood pressure and her whatever. And so you know that Ma knows that. Well, they would say, hey, Carol, you supposed to take your, you know, medication every day? You know, you just said you're not, so I'm not. You know, we just gave you a script for that last week. What's going on? And so there's that thoughtful process that comes in it as an example.
[00:17:24] Speaker A: So, so that's one layer, the digital employee kind of motif. But employees do work in workflow, right? That's the way the real work works. So you talk about the second layer of this. It's not just thinking about this AI like you think about humans to be onboarded, managed team, but think about them as doing work across workflow, not a data set. But this is a process of pre authing a patient. This is a process of refilling a scri.
What are you seeing? What are you learning early as you're helping some of your clients deploy this technology around how the workflow construct, the workflow framework is helping with the evaluation, with the deployment, with the realization of benefits.
[00:18:11] Speaker B: I mean, it's. No, you do look at the workflow, you kind of do the end to end workflow and then look at it from an outcome. And we kind of work our way back when we look at it.
I think we treat it as you look at where the handoffs are. And so what I recommend doing is you also want to provision your agents to make sure they have the right permission, just similar to how you would for a human. The information can they access, what can't they access? And then, you know, depending, depending on the workflow and everything else, you want to make sure that they have the appropriate handoffs and have the appropriate credentials. And so when you set somebody up back in, back in the day, you would have certain roles and rights that you could do. We recommend doing the same thing with the agent. Few reasons why you do that. One is so they can, they understand their role, they can do it and you get the maximum value at it. They also can do that critical thinking for that particular part of the workflow, but also to limit any damage they can do versus everything else. Right. And so you want to kind of have that chain in place in order to do that. I think that will change over time. But today you do that to it makes it easier if there's a problem to fix and also reduces your exposure as well. So those are some of the things that we put in and you know, kind of start off with that basic piece of it, with the outcome. So, you know, if you're doing, if you're going out to any sense of leashing, it's a nice way to kind of contain that.
[00:19:25] Speaker A: So, but let's finish with this question and I'll, I'll narrow it back down 25% of the way into 20, 26,
[00:19:35] Speaker C: 3/4 of the year left.
[00:19:37] Speaker A: When we get to the end of the year, where do you think we're going to actually see in this short time frame?
[00:19:43] Speaker C: We've moved the needle on some things, whether it's quality, safety, productivity, production. Early stage deployments of AI are having a positive. Where are you most optimistic?
[00:19:56] Speaker B: I think, you know, I think if I were to look in the next nine months, I think, I think, I think a lot of it has to do with AI is around the architecture and how it's done. Right. And I think that's important. And then I think you come down a level. It's really those, those.
It's really the unique capabilities of the AI to do those features and functions in that workflow. And then if I were to go down a little bit more, is the ability to do that what I call critical thinking and to be able to adjust that. So to me it's kind of multilayered. I think you talked about that earlier.
I think quite honestly, my baseball analogy, I would say we're probably at the top of the third inning in this. I think things are going to progress very quickly. So I do a lot of work with like autonomous or automated coding is very big.
I think the initial results are great.
They can look at a chart, they can do the appropriate coding. You know, let's say I'm a coder, I can do 30 charts. You know, these agents can do, you know, 3, 10, 5, 15X, a lot more than that. I see a lot of productivity gains for that, especially for something that's one highly regulated, has reimbursement implications, either positively or negatively. Because you have to, you know, if you get audited, you need to make sure that's defensible. So I think you're going to see a lot of. Lot of improvements in that area.
I think you also. So I think if I were to take the revenue cycle piece of it and then I'll talk about the clinical piece of it here in a second. I think the rev cycle piece, you'll see that a lot of advances in coding, which obviously is huge labor, very expensive.
I think the ability to go through the chart making sure all that information is there. Dave came in, he had an encounter.
They spent 30 minutes with me. It's appropriate. Level three, by the way, he had flu vaccine labs. I think you're going to see in that simple encounter, I think you're going to see dramatic improvement between now and the end of the year. I think if I were to look at years two and three, I think as you get more complex. Dave comes in, I have a mole, you know, then I get a biopsy, then I have to go in and get my surgeon to remove it. That sort of stuff, that more complexity in terms of the medical care in terms of that cycle, I think will take a little bit longer just because there's more steps involved. I think on the clinical end, you know, the ability to surface things like, hey, you know, Dave's diabetic, We didn't see a note in there for his insulin. You know, what's up with that? You know, those, those warning things that are part of that workflow so you don't have to do it afterwards, or geez, Dr. Tim, you know, we noticed that you missed these clinical elements. Before you close out the encounter or Dave leaves the room, please do that information because this is what we would see for standard of care for this type of patient. So I think some of that is already in place. Obviously a lot of work in the ambulatory listening pieces of it, which is pretty good nuance. Suki, A bridge, I scribe some pretty amazing stuff. I think that technology between now and the end of the year is going to get better. And it's some of the amazing technology. I don't know if you do like call centers, like I called in the hotel the other day and I couldn't even tell it was not a human.
I think you'll see that as well too, the ability to distinguish between the patient and the doctor and really get that comprehensive note. I think that's going to just explode.
[00:22:56] Speaker A: Awesome. It's an interesting time. Dave, thanks for joining us. We really appreciate having you on vital conversations.
[00:23:03] Speaker B: Take care. All right. Thank you.
[00:23:04] Speaker C: Bye bye.
[00:23:05] Speaker D: Thanks for listening to today's episode. For more conversations like this, subscribe wherever you listen to your podcast. We'll be back soon with another discussion focused on what really works in healthcare.
