Two Agents Stuck in a Loop | Solution Showcase with Donald Pollard

This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Two Agents Stuck in a Loop

[00:00:06] Speaker: Hey, I'm Drex DeFord. Long time recovering healthcare, CIO. Now one of the principles of the 229Project. I'll be your host for today's solution showcase. Uh, first of all, let me say thank you for listening today. We always appreciate your time. And this episode as always promises to be a really good one. Uh,

[00:00:24] Speaker: and thanks to our sponsor today. SHI healthcare with me today is Donald Pollard, the field CTO with SHI international who works extensively with the SHI healthcare team on AI strategy and execution for customers. Uh, welcome Donald. I'm glad you're here. Thanks for having me excited.

[00:00:45] Speaker: Uh, to be talking to you and of course, project two two nine. Yeah. Um, we're, uh, I'm looking forward to sort of having this conversation about, um, all the stuff you're doing in AI, how you're speeding AI delivery and healthcare. More on that in just a minute. The best place kind of to always start is just tell me a little bit about your background and what's happening with healthcare at SHI right now. Yeah,

[00:01:08] Speaker: it's been kind of a passion of mine because I married a nurse and I got to see like how much of a, uh, how much it really takes to really deliver that quality of care that we're every health system is trying to deliver. Like, what does that look like on the late nights and the overtime and all those shifts. So, uh, I was able to try to give back in my own little way of a pretty new IT systems rather be back in

[00:01:32] Speaker: the day from, uh, you know, uh, spreading out security and make sure, uh, EHRs are first being implemented all the way up through cloud computing. Uh, and now with that shipped into AI, how to actually put these cool new tools in action to actually get real meaningful results. So people can end their shift and go home, go to Home Depot, go, go to the park, uh, go on vacation and feel

[00:01:55] Speaker: like they can really like reach rest recharge and come back reinvigorated, take care of people. Yeah. I love the, the mission orientation, um, to this, to all of our jobs, I think, because it really is about not only the providers, but the patients and families too. We hear as we travel around the country and do our events, um, and we do the 229 signal reports,

[00:02:18] Speaker: uh, there's just flat out, not enough resources. There's too much capacity is max. There was a high demand for AI. Uh, what are some of the challenges you hear you see in person when you talk to health systems around the country? I think like probably the biggest one today that I'm seeing

[00:02:37] Speaker: across, uh, across the industry there is really like how to utilize AI and get some returns out of it. I would say my number one, one of two that I have every single week, uh, is going to be around. Hey, uh, I'm being pressured to get AI into my health system. Uh, how do I get it in there?

[00:02:58] Speaker: And what the heck is this thing actually going to return back to me? Uh, versus I think so many people like look at AI as it's a business problem to be solved versus it's just another solution for your actual business problem for the care for again, like the provider, uh, in a small town network, it's hard to recruit those folks. And if you could take away some of that paperwork burden,

[00:03:20] Speaker: because I don't think a single doctor has ever like woke up and said, one day I'm going to grow up and do paperwork. Like this is more admin work to do. Yeah, exactly. Like they want to, they want to take care of people and like, how do we refocus technology to allow that and enable that action and be able to like provide that care that they're looking to?

[00:04:33] Speaker: So I want to ask you about this thing conceptually. I don't know how you really title it. We sort of talked about it as a genteck bootcamp, sort of the make version of this on steroids that y'all are doing lots of efforts across the organization, lots of iteration. Tell me about a genteck bootcamp and

[00:04:52] Speaker: how it works. Yeah. We started looking at across the landscape itself is that the amount of care needed by, by, by just people, uh, out there in America right now, it's just so dwarfs the amount of care we're actually physically able to deliver. Even if we filled every job and every open

[00:05:12] Speaker: requisition out there right now, and we had all the providers and nurses and all the support staff working out a hundred percent, we'd still be 2 billion hours in the hole. we created a genteck bootcamp to really attack the problems that every minute that we give back to a provider is another minute that we can give to people in need and needing of care. Uh, and so that was

[00:05:35] Speaker: the big, big one there. I think part of the biggest problems that we, we really saw was that a patient there, um, uh, is maybe one real person in real life. Uh, but it is seven different people across all of your data systems, uh, rather than be your EHR, your labs, your portals, your this, your that all those it's impossible to tie those, that information together and build

[00:06:00] Speaker: AI systems that can actually communicate, pull the right data, go ahead and actually provide like an actual use case. That's not just, uh, uh, kind of throw against the wall of like AI, POC, like spaghetti and see what sticks. We really wanted to go in there and attack like what we call like the, the, the big hairy gnarly problem in your organization, because that's where you're

[00:06:22] Speaker: actually going to get the most results. So we wanted to go in there from a way of having folks like sit down with your actual charge nurse, watch their actual, uh, uh, click through like, uh, workflow, those types of things and actually identify what is the part that's actually slowing them down day by day and then build an AI system to help them go ahead and reclaim that. It's a

[00:06:44] Speaker: smooth out that rough spot. Uh, so they'll be able to get through the work faster, easier and provide more care. So this is really, uh, interesting to me because a lot of times this sort of the sage advice that you get is go after the low hanging fruit, the stuff that's really easy to understand when it comes to AI projects start there and that's okay. People can do that. I think

[00:07:07] Speaker: that's probably still a good place to kind of think about and a good place to start, but you're suggesting they can go after these big, hairy, audacious problems and get more value even quicker with this model. Yeah. Low hanging fruit. I think, uh, yeah, I mean, that's, that's one we've gone

[00:07:27] Speaker: after since the dawn of AI since like a chat GPT first busted on the scene. There's a way to kind of get people used to interacting with an artificial intelligence. Like let's go after and just summarize our emails. Let's, uh, have a little bit, uh, productivity type of gains. The problem with that is that they're a huge investment and really the return on that

[00:07:48] Speaker: investment is usually pennies. The hard problems is where you're going to get the dollars. That's when you're actually going to get, I have changed like how much time it actually takes to go through an entire patient experience, uh, from, from going ahead, uh, in the rev up revenues, uh, operations all the way through the clinical experience and then, uh, the, the outbound as

[00:08:09] Speaker: well. So I really think like going after low hanging fruit, you're going to get low hanging returns as well. Uh, I think going after the high, the more complicated ones, that's where you're actually going to have value. You can actually say, Hey, our patients are having a better experience because of this technology versus we have this technology. Isn't it cool?

[00:08:31] Speaker: It's so it's sort of like, um, you can't believe that when you're going to start these projects, they're going to be perfect out of the gate. Right. One of the things that you and I had talked about was this concept of forward deployed engineers and how you're using forward deployed engineers, for lack of a better term at SHI healthcare. So tell me about who are these people

[00:08:55] Speaker: and what do they do? What skills they bring to the table? How do they help these big, complicated projects get off the ground and actually become something useful pretty quickly? Yeah, that's, I think this is like the key to it all. Um, I really do like, uh, you know, I know when I was in a leadership there inside of healthcare, I would have loved to have this model

[00:09:17] Speaker: of someone to go into my organization, sit down with my frontline people, the people whose problem we're actually trying to solve, actually just watch them, watch the workflow and watch where the problems are happening and ask the intelligent questions and then actually code up the solution

[00:09:36] Speaker: there on the spot and not wait six months for 180 page document and a PowerPoint that no one will actually ever read. I don't know about you as CIO, but like, I think the old school like version of consultancy, I spent millions of dollars on that. And that typically was my experience. Like you, you would go, they would go through that whole process and it would take six months. And then

[00:09:58] Speaker: we'd be right back where we were when we started the process. Now, several million dollars later, six months later, you're at the starting line and you still really don't know exactly what you should be doing next. It's really interesting. I'm a Toyota production system guy. A lot of folks have heard me say this. What you're describing is in Toyota production system speak,

[00:10:18] Speaker: the Kaizen man, the person who had a production, you know, manufacturing facility goes and watches the work and helps figure out how to take one second out of the work or one step out of the work, make it just a little bit better every day. So that's how you're using forward deployed engineers. Yeah, absolutely. It's that supercharged business analyst that knows Python

[00:10:40] Speaker: and knows how to make a data pipeline. They're on the seat right there with the person. I mean, how do you get past like AI adoptions or being scared of AI? Well, you build it with the person. They actually built this with you. And so when they get done, it's like, no, this is not just a app, AI application that's being forced to use. This is one that I helped create to fix

[00:11:01] Speaker: my problems. And my job is that suddenly, like, those those those are when you were actually attack real problems. And those when you actually get people not just to buy in like, okay, cool, I'll use, you know, this this chat bot, like, I'm going to use this AI, because I believe in it, I helped build it. And I'm going to move forward with that. Yeah, interesting. I'm going to ask you more about that people afraid of AI question here in just a minute. But one of the other

[00:11:26] Speaker: things I want to know about all this is, as we talked about this approach, using the data as it is, right, we're pretty famous for not having well organized or well governed data. And so

[00:11:44] Speaker: again, kind of the sage advice is wait until you have your data squared away and then use, you know, try to use it. How do you get past that? How does that work? you have data that goes across, you know, your one individual patient, it goes across like seven different systems. And it's all has a unique identifier on your EHR that has social security

[00:12:06] Speaker: number on this system and this one, like be able to actually map that and make it clean and actually turn that into an actual business object itself, whether it be a patient bed or a patient themselves or whatever it is, be able to do that in a way that doesn't touch the source system. And I think that's going to be the big clear one. Because a these stuff is usually all locked around inside

[00:12:30] Speaker: those SaaS applications that we have, we can leave it to where it is, then we can abstract that data to a place that we can make connections virtually across the board, we can build AI on top of that. And now we have clean data without actually going through like a 10 year, 80 kazillion dollar like data modernization project, we actually just virtualized it, abstract it, mapped it, made that

[00:12:53] Speaker: connection points and then built AI on top of that. I think that's going to be the key to not having to wait to 10 years to get your first POC off the ground. You're helping to to drive the data governance and the cleanup and all of that stuff as part of one of these projects. How fast does all this happen? From the start of, you know, I signed a contract with you,

[00:13:15] Speaker: I want you to come in and I want you to go into one of my departments and put a forward deployed engineer in there. They've got some challenges, help them figure out how to make work better. How fast does it from there to like we have something that actually starts to work? How long does that take? Yeah, great question. I mean, it depends, I'm sure is one of the, you know,

[00:13:37] Speaker: at the starting point. I it a week. Oh, wow. A week. And I say that like, no one's going to believe me. There's like not a chance like we built this bootcamp to take one use case. And once we kind of identify going, all right, this is the use case, we're going to attack supply chain, we're going to

[00:14:00] Speaker: attack ERP or inventory balancing between multiple like hospital sites or any of those is that once you kind of like define that and what we're going after what we do is we go ahead and like have a precinct and we kind of dive into like, okay, who is the process person that we can talk to and we can sit beside what's all the data systems that are actually identified within that use case,

[00:14:26] Speaker: load it all up, and then start to go ahead and what we call the kickoff, the actual agenda camp itself. We have the FDS come in there, and they're there sitting there learning the workflow, learning where the problem is, linking all that data together. They start on day one with a great idea. And by Thursday, you have system that's in someone's hands. This is a way for us to actually

[00:14:49] Speaker: prove it. And I think that's the big thing is that we've all been part of the marketing blitz. When it comes to AI, like, hey, give us a dollar, we'll start raining hundred dollar bills out of the ceiling. We know it's not true. Or it's so it's such a system that we have where it's on the shelf

[00:15:11] Speaker: solution that's made for everyone. Yeah, those made for everyone generic ones typically don't work in a situation where you have data across all these different systems and you actually have to build something custom to you. Everyone's unique in their own workflows and what they're doing based upon decisions that we've made technology purchasing wise, process wise, you know, that could

[00:15:31] Speaker: be over decades. Having that system there is a way for us to prove that yep, this use case is a great use case. And this one we can actually ascertain how much return on investment we're looking at if we go ahead and scale this out to the entirety of the organization. So yeah, it's a week. And it's

[00:15:52] Speaker: a week of like prove it. Yeah, you will actually have a working system in your hand by the fourth day of our engagement. So going really fast. I have to say the other part of this that I'm sure I'm if if we were live, I would have people asking me questions already sounds really expensive. Can only big systems do this are smaller health systems able to afford this? How does that work?

[00:16:17] Speaker: Yeah, we actually made this to not target big systems at all. We are talking, we are targeting the smaller systems that could really benefit. And again, I'll go back to I come from a small town in Missouri, our regional health system there, it is hard to attract talent. If I can put a system like this in there to actually say, Hey, come here, you'll have a great work life balance. Well,

[00:16:41] Speaker: that gives me a leg up as far as to be able to attract the right talent. And so we price that in a way to go ahead and make it make it palatable for all different sizes of organizations. I want that. One last thing. And we could go on forever. I love talking to you. We've had several good conversations. You think about this

[00:17:06] Speaker: stuff a lot. So more broadly, I'm hearing a lot more about this issue of tokenomics. It certainly applies to this sort of business that you're doing, too. I've heard you say architecture is the best way to save money in AI. So between that and tokenomics, give me some picture of like what

[00:17:25] Speaker: the future looks like. Yeah, no one likes getting that angry phone call from the CFO. And I think right now with tokenomics and tokenomics for everyone out there who may or may not know, it's just the way that we pay for AI. This is just a big word that we just someone comboed together

[00:17:43] Speaker: and like, oh, that sounds cool. It's just how we pay for AI. It's very similar to when everyone first started shifting to the cloud. We started shifting to a consumption based model. And now you're seeing AI that had that, hey, pay $15 a user a month, you have all the AI that you want,

[00:18:01] Speaker: just go nuts. You can tokenmax and go crazy. Now everything's run on a meter. We start running on a meter is that suddenly your 12 month of AI budget gets used in three, four months like Uber did. You got two agents that get stuck in a loop and then suddenly you have a $40,000 bill.

[00:18:22] Speaker: That's not like stuff that's out there as far as like a quick research pulls that together. That's three customers here in Nashville I talked to that they've had that happen. Tokenomics really jumps into like how do we not only like have observability on how much we spend or putting a hard quota, when I kind of go into the architecture side of that is that do we really

[00:18:45] Speaker: need to send what's the weather like today in Austin, Texas to fable five to something that is the most expensive AI model out there. And I think that's the big question. I think as healthcare systems as any industry really like how do we move the routine to systems and AI systems

[00:19:08] Speaker: that are more cost-effective to run that rather be on-prem, it could be cloud hosted and also how to go ahead like the really complex stuff that we do need to have like these more frontier labs, models from like entropic and an open AI. It only takes care of that stuff. And so you go ahead and

[00:19:27] Speaker: you divide your bill and you cut your bill in half just by architecture. It's just knowing that I have an AI router in the middle says, hey, what's the weather like in Austin, Texas? While a good question, a fun question doesn't need to go over here and cost me $50 per million tokens. I can go over here and get it for a nickel for a million tokens and get the same response. Yeah,

[00:19:52] Speaker: now I appreciate that. I'm glad you guys are thinking about that. As you help health systems build some of these systems, it's all part of the business plan and they've got to figure out what they're going to do, how it's either going to help them see more patience and add revenue to the pot

[00:20:12] Speaker: or how they're going to reduce costs. And so using the right model turns out to be really important. We're building all the frameworks out for it right now. I don't think people really like, as far as what tokenomics is and then how to get observability and set guardrails and policies and not just saying, hey, this many million tokens cost this much dollars, but it's actually tied to

[00:20:35] Speaker: this outcome. It's the cost per patient onboarding. It's the cost of XYZ type of thing and how to spend that. architecture makes all the difference. Hey, I want to say thanks. Thanks to the audience for joining us today. It was really great to spend time with everyone. Thanks to SI Healthcare for

[00:20:55] Speaker: sponsoring today's solution showcase. Thanks, Donald, for being here and being a part of this today. Absolutely. Glad to be here and thanks for having us. And yeah, we really appreciate the partnership with Project 229 and looking forward to our next event. We'll go and hang out in person. Yeah, I'm looking forward to seeing you. Okay, I'm Drex DeFord from the 229Project. And as always,

[00:21:18] Speaker: again, thank you for being here. You want to know more about what we're doing at the 229Project and where we'll be. Well, the 2026 tour is nearly finished, but the 2027 tour dates are locked in. So if you're a health system leader, you can find out where we're going to be for the summits, for the city tour dinners, for the regional staff roundtables. Just drop us a note at hello@229project.com.

[00:21:43] Speaker: Tell us what you're interested in and we'll find the right place that is a perfect match just for you. Thanks again for being here. Stay a little paranoid. We'll see you around campus.