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

 Thanks for joining us on this week in Health it. This is a Solution showcase. My name is Bill Russell, former Healthcare CIO for a 16 hospital system and the creator of this week in Health. IT at channel dedicated to keeping Health IT staff current and engaged. Today, Matt Fordo joins us. He's the head of artificial intelligence at Trace three, which is.

A really cool title, and we have a great conversation about what it takes to get your AI program off on the right foot, what it takes to get the right foundation set, what maturity looks like, who the first hire should be, what the budget should be, all sorts of questions like that. Great conversation. I think you'll enjoy it.

Special thanks to our influence show sponsors, Sirius Healthcare and Health lyrics for choosing to invest in our mission to develop the next generation of health IT leaders. If you wanna be a part of our mission, you can become a show sponsor as well. The first step. Is to send an email to partner at this week in health it.com.

I ran into someone and they were asking me about my show. They are a new master's in Health Administration student, and we started having a conversation and I said, you know, we've recorded about 350 of these shows, and he was shocked. He asked me who I'd spoken with and I said, oh, you know, just CEOs of Providence and of Jefferson Health and CIOs from Cedar-Sinai, Mayo Clinic, Cleveland Clinic, and

Just, uh, all these phenomenal organizations, all this phenomenal content, and he was just dumbfounded. He is like, I don't know how I'm gonna find time for, to listen to all these, all these episodes that I have so much to learn. And that was such an exciting, uh, moment for me to have that conversation with somebody to realize we have built up such a great amount of content that you can learn from and your team can learn from.

And we did the Covid series. We did so many great things, talked to so many . Brilliant people who are actively working in healthcare, in health. It addressing the biggest challenges that we have to face. We have all of those out on our website, obviously, and we've, we've put a search in there. Makes it very easy to find things.

All the stuff is curated really well. You can go out onto YouTube as well. You can actually pick out some episodes, share it with your team, have a conversation around those things. So we hope you'll take advantage of our website. Take advantage of our YouTube channel as well. Just a quick note before we get to our show.

We launched a new podcast today in Health it. We look at one story every weekday morning and we break it down from a health IT perspective. You can subscribe wherever you listen to podcasts at Apple, Google, Spotify, Stitcher, overcast, you name it, we're out there. You can also go to today in health it.com.

And now onto today's show. Today we are joined by Matt Fordo, who is the head of artificial intelligence. Matt, welcome to the show. Thanks for having me, bill. Appreciate it. Well, I'm looking forward to this conversation first of all, because you have the coolest title in history. We had a, a short phone call leading up to this, and there, there was so much great content that I'm looking forward to sharing it with the community, give the community an idea of what is the title about, what do you do at three as a head of ai?

Yeah. Uh, a little bit of everything really. My background comes from a PhD program in industrial organizational psychology, and, uh, somehow that segued into a lot of data science roles. Had my own data science consulting firm and, uh, then ended up joining this space on. Um, the, the solution provider side.

Yeah. And we're gonna, we're gonna delve into all of those areas. Industrial organizational psychology. Is that, what, is that what you said your major was? Uh, yep. , that's a, that's a, for grad school at least. You know what, that's a great major. We, we've talked a bunch on this show about the fact that ACIO is less about technology and more about change management and really moving an organization from point A to point B, which is a lot of industrial organizational psychology, I would imagine.

Yeah, and I think that's why I got lucky and uh, was successful in this field was primarily integrating the fact that everything with AI is foundation based on the data that resonates with it, which is based on people's attitudes, behaviors, and cognitions. And the advent of all of that mixed with understanding how people operate and about transformation and change within an organization helps combine those two together.

Well, let's talk about the business problems, the, the specifically the business problems that AI can solve. Now, you, you work inside of healthcare and outside of healthcare. So let's, let's start with a discussion of outside of healthcare. 'cause oddly enough, people inside of healthcare are looking more and more to the outside to say, you know, what are they doing with technologies outside of healthcare and then bringing it in.

So outside of healthcare, what business problems are?

I mean, you know, I, I say take a step back and look at the 30,000 view, right? Because. Want ai, and I don't necessarily know what that means. And, and I'll, I'll say AI is the umbrella term can be data science, uh, machine learning, deep learning, all these different, uh, combinations of, uh, nomenclature. But AI at root is, is there a business problem and is there data that can help answer that business problem?

And if that's the case, then AI can help solve these things, right? Identifying patterns, clusters, et cetera. And, and so when we're looking at outside of ai, what are the. Where, I guess where is the market going and who's investing in this? Who's seeing value out of this? The primary places right now are automotive, especially with self-driving cars and financial services, so fraud detection, likelihood of default, and those kinds of things where it can make much better decisions than humans can in regards to if a credit card purchase was uh, fraudulent or not.

But it's segueing into a variety of sectors as well. You're seeing retail use this, and when you go into a store, there may be cameras with heat map to identify where customers are going, or facial recognition to detect your sentiment and see if you're actually having a negative or positive shopping experience.

So ai.

Like credit card transactions and those kind of things. Look for patterns. These are acceptable transactions, these are questionable transactions. And then act, you know, essentially kick off a, a series of a workflow or of some kind based on what it's seeing. Because when we think about credit card transactions, if you were to do that with people, the financial institutions.

Look for fraudulent or acceptable transactions. So that's where the, the computer really steps in. And, and what you, you also described is it can, because computers now have things like computer vision and they're able to make sense of images and those kind of things, we're, we're starting to use it in other areas where we're, we're.

You know that that person's not happy. That person who just walked outta the store is not happy. We, we can tell those kinds of things based on the image. Yep. Yeah. Dead on. Right. So there's, there's three main types of use cases. There's prediction in regards to say, what is my home worth based on the number of the square footage, number of bedrooms, bathrooms, et cetera.

Right. And obviously there's a ton of features. There's classification of is this fraudulent or not? And there's clusters, right? Who are the types of different groups that if we go to healthcare, who are the different types of groups that doctors see? Right? There might be the young people with chronic issues and then older individuals who are just coming in for checkups.

Yeah. So how, how are we seeing AI applied? Healthcare there, there's, you know, it's interesting we're sort of playing recovery because one organization came in pretty, pretty loud and proud into healthcare saying we're gonna do all sorts of crazy things, making a lot of promises, and, and so we had to go through that and realize that there's a lot of clinical applications that are too fine at this point.

We're talking about life and death that we had.

Before we start applying it to, I don't know, diagnosing patients and those kind of things at this point. Yeah. You know, I, I, I, I think of it as where's the, the value add, right? And I'm sure we'll get into this today, but you know, the, the people that have to drive AI within an organization are the people that understand those business problems and business use cases and the, the cost or potential value associated with them, right?

Because if you just had a, a data scientist, uh. So sign they can do that with data. Um, mathematically it's very easy to do that, but is that really going drive, you know, cost savings or better optimization and efficiencies, et likely not. Within healthcare, I'm seeing the trajectory go in a couple different patterns.

The, the insurance organizations are investing very heavily in it, have tons of data scientists and you know, are being able to more accurately and rapidly assess claims and approve or, um, not, and, but then if you look at the hospital systems. There are some that are very progressive and some that are more laggard in regards to it.

And, and I, I see it always as a crawl, walk, and run approach, right? Because organizations, let's take, uh, billing and ICD nine and ICD 10 codes. For years and years, it had always been focused on manual. Everything has to be done manually, and that means it has to scale linearly with more labor. Then we started getting to OCR, which can hopefully detect some of those aspects within billing.

Now we're getting into deep learning, which to your point with computer vision, and that's only processing, now it can read and understand the text, right? And automatically identify what was done in regards to some sort of prognosis and, and then identify what the billing codes would be associated with that, based on the computer vision, reading the actual page, and.

The natural language processing, being able to understand the context of that. So the long and short answer is it's really going across the spectrum for the organizations that are moving further and further along, um, in their AI journey. Also seeing it as things like using AI to augment doctor's success.

For instance, using AI software applications and models for x-rays and MRIs to detect anomalies so that doctors can be more efficient and spend more time with their patients. And all. So leading ultimately into personalized healthcare, right? Because at the end of the day, end of this, of the what.

Ultimately stop treating the disease and treat that patient holistically. And if we can do that using augmentation through ai, we should be able to reduce cost, uh, reduce the number of treatments and other associations for a diagnosis and get a patient in and out the door faster and happier. Yeah, that personalization's interesting.

I mean, in healthcare.

Johnson Johnson vaccine. Generally it's, it's safe. It's a very safe vaccine except for a handful of cases where it's not, and you know, it's this kind of stuff where early on in the medical record we had the, the Vioxx issue where people always ask me, it's like, is there a study that shows that the EHR actually digitizing the record has helped outcomes?

And I, I just said, yeah, Viox. Essentially because we had all that data, data scientists went in and looked at it and said, Hey, look, everybody we're, we're giving Vioxx to is having adverse reactions. And they were able to, to service that and change habits across the, uh, across all of healthcare. But when you're talking about an N of one, when you're talking about that level of customization.

You're talking about, you know, three point something million people just in the US alone. That's a lot of processing. That's a lot of, I don't know. That's just a lot of data that's gonna be coming in and I, I would think at some point it's just gonna overwhelm individuals and we're gonna need that ai, machine learning, computer vision, whatever it is, to come alongside of us and say, Hey, you know, I'm not gonna diagnose, but you might wanna look at this, this, and this.

No, I, I, I completely agree. Uh, because that's, uh, where I think that the state of the industry and, and not even just healthcare, but the state of all industries is going to be, is how can we relieve people's essentially mundane task, right? Or the things that we do on a. Aren't quite effective. Right.

Ultimately, hopefully we'll get to a point where we don't even have to do emails anymore. Oh my gosh. You know, AI for email, sign me up

But, uh, yeah, I mean, I think that there's, there's huge, huge opportunity, um, especially in this space and I think with the way that GPU computation works with parallel processing, we're getting to that point. And with the amount of data collection we have, we're getting to. We can get smaller and smaller clusters of people to hopefully drill into that end of one that you talked about.

Well, and we're gonna get into the technology stack in a little bit 'cause I, I really do want to delve into that a little bit. But, uh, one of the things I found fascinating in our last conversation is you start talking about an AI maturity framework that you guys utilize at Trace three as you go into organizations to really help them to think through it.

Can you share the framework with us and give us an idea of how it's utilized? Yeah, so I, I built up this theory and, and then built a model from it on five levels of AI maturity. And really they are clusters to help an organization define where they're at in their AI journey and, and, and their maturity and where they want to go.

So if you look at a level one organization, they don't have data scientists, or they might have one or two. Just came out of a, a bachelor's degree, right? And those poor kids have no idea how messy real world data is, how disparate, how distributed, et cetera, right? And, and so it's kind of a, a wake up call, but it's for those organizations to do something more, a little more meaningful than just Tableau and Excel reports, right?

Trying to get some insights from the data. Try to get something predictive, try to understand some value. Of the spectrum, the level five, Amazon, Microsoft. The companies that are doing things to completely innovate or renovate society, whether for good or bad, is deterministic, but they are doing things that are, are, are so monumentally transformational that they have tons and tons likely hundreds of data scientists.

They have an executive level AI champion that is not only evangelizing this out. But really promoting it within the organization to articulate and show the value of data and these models. Because one of the biggest gaps that I see within organizations is if there's not a, a senior leader that is passionate about data, that is passionate about the value of what data can provide.

And mind you, like, as much as I love ai, AI can't exist without the the data. Right? It's at the end of the day math. Science with some art mixed in, but it's that data that's really telling. And so if you have an AI champion evangelizing and promoting that out with the right team framework, with the right hardware and software applications and the right processes for that organization to be successful, then they're going to do so.

Levels 2, 3, 4. Basically sit at different clusters within that journey. Lower level AI champions that move up through the cycle through the organization. More and more data science teams, seeing models productionized and seeing the value, revenue, um, or cost savings come from those models. And those models should ultimately, and.

Probably by level two actually start paying for that team, for the infrastructure, the software applications, all the revenue that that group is making should pay for itself. Uh, because in all the analyses I've done and from what I've seen from. Other research teams, the value of every dollar invested in AI is conservatively five to 10 x and up to 30 x for every dollar spent.

This question is probably gonna be more revealing about me than anything, which is do I have to start on level one or can I start on level three? Okay. And just say, alright, that's the foundation. We wanna start at level three. Yeah. You, you can, and you know, I think it, it takes a higher level AI champion and it takes a budget to do so.

AI is fortunately or unfortunately not a cheap venture, but it is. What's going to separate organizations in the next five to 10 years. If they're leveraging data in a meaningful way, then they're going to, uh, make a much larger impact for those that aren't leveraging data, still doing things manually, it's gonna become more problematic over time.

And, uh, so like what I'm seeing currently and with a lot of customers that I'm working with. Involves helping build a, a center of excellence, right, which is more so centralized data, centralized model repository so that data scientists, machine learning engineers, even the analysts and whatnot, can be more successful because that data is more accessible.

People are understanding what's being done. And if you have that AI champion promoting and evangelizing the wins to the rest of the organization to start leveraging that data science or analytics team, then you're going to see a lot more success. So yes, absolutely. There's plenty of organizations that I've seen that, you know, decide to make a big investment and move forward pretty quick.

It just, just takes some bucks to do that, which is not, not always easy. Talk about the data a little bit before I. Standing up an let's talk about the data first. You know, as I think about it, you know, I'm very familiar with healthcare data. Actually. I've worked in a lot of industries. I'm familiar with, uh, finance data and banking data as well as manufacturing and, uh, retail data.

And it's just cleaner. I mean, so it, it just, people a lot of times will say, you know, why can't healthcare do this, this, or this? I'll say, you know, a transaction on Amazon is, is a very clean set of data. I mean, there's, there's very little to clean up in that respect. I mean, they create all the discreet data elements and they sort of own that whole cycle, whereas in healthcare.

When somebody goes from one health system to another or one practice to another, you end up with different data entry clerks. We call them doctors and nurses, but you end up with different data entry clerks, and that's what we've turned them into. And quite frankly, it's not what they went to school for and, and some of 'em aren't, aren't all that good at it.

So you end up with, you know, notes that are just loaded with tons.

So give us an idea of what. That lead to that, or at least the foundation for a good AI program. Okay. , that's a fun loaded question. Yeah. My, my, my, am I, am I leading the witness? Is that what's happening here? I'm sorry about that. . Uh, okay. We'll see if I answer this appropriately, but ask for clarity if you need it.

So there's. Two main mechanisms, and I'm going to be a a little overtly broad here. There's machine learning and deep learning. Machine learning is basically structured data. Technically they're model frameworks that can work for other types of data, but we'll say machine learning for structured data, so sql, Excel type files, and then deep learning for unstructured data, audio, video, text, et cetera.

80 to 90% of data science projects and workloads and use cases. Are still on structured data right now. So in that component it means that we need, uh, a better way for data engineering and processes to move and transmit that data and ensure its accuracy for data scientists to leverage. And that can be done for a variety of components, especially like let's say front desk receptionist inputting patient's data, right?

That's structured can easily feed into a data warehouse. Great. But to your point, when you get into deep learning, then you need data science talent that understands linguistics and natural language processing and or understand deep learning and computer vision. Because if you're trying to read these hand.

And not only trying to read that, but being able to parse apart, understand it, um, and extract the relevant features or data from that and either be able to put that into a table so that you can do machine learning on certain variables or features within that, or just leverage that as an unstructured document and do deep learning on that.

Require a lot more lift, both from the talent side, the process side, and the infrastructure side. We'll go back to our discussion in just a minute. Today's show is sponsored by Dell and Trace three Dell, EMC. VxRail, powered by VMware. vsan gives you a fast track to hyperconverged infrastructure, HCI, without risk, while lowering it costs and providing an agile solution ready for future hardware.

Cloud and application changes. To learn more, visit trace three.com/explore. Dash dell dash emc dash VxRail. Now back to our discussion. All right, I'm gonna, I'm gonna hit you up for some free consulting here. Let's, I wanna set up my organization for success in ai. Where do I start? Who's my first hire? The, the, the person that I need to be the, the, the first AI person within our organization.

Yeah. I, I think that you have to curate that AI champion, right? Somebody that caress enough about data science and AI to help promote and see that takeoff, an AI champion is my first hire. That's the category. No, no, before, before, before the hiring itself. You want to have somebody that believes in this, right?

And it can be as low level as like a.

If you're actually looking to get traction, productionize these models and see success with ai, then you're going to want to have somebody, and it might be ACIO, it might be ACTO, it could be a VP of or director of infrastructure. This oftentimes comes from itd, it oftentimes comes from a business unit themselves, right?

So the, the chief operations officer may want better cost, better efficiencies for the organization and knows what those use cases are. That's why you need that AI champion because they understand the business and they understand that there are problems that need to be solved. So from there then you, the first hire, you probably want to have.

Is really contingent on, uh, what your data looks like today. So you might want a data or machine learning engineer to start getting the right data curated together in a meaningful way because unfortunately, most data scientists are really only trained in. Statistics, machine learning, maybe deep learning, so they don't know how to ingest the ETLs, APIs, all those fun processes, they can't do the front end and they can't necessarily productionize and scale those models to large, large scale ecosystems.

So you need a data machine learning engineer for both those pieces. And a data scientist is leveraged in the middle to build those models. And so it can go one of two ways. So if you have good data already collected, accessible, et cetera, then you probably want to hire a data scientist, ideally, one that understands healthcare, because feature engineering is a key, key component of data science.

Feature engineering means. Creating new variables from other variables that you already have, and those often drastically increase the accuracy of a model. So if you hire a data scientist fresh out of undergrad that doesn't know anything about healthcare, they don't know that, let's say smoking and drinking increases, the interaction of those two increases your risk of cancer.

If somebody in healthcare does know that, then they know to create that variable and include that in a model for predicting the likelihood of someone getting cancer. So I, I'd say it's usually one of those two roles that really starts helping drive success. Would it lead to a higher degree of success if we were able to get a clinician who is passionate about AI to really be the champion?

I'm working with a few healthcare companies and hospitals and the research. Doctors that know what their business problem is and, and so some of them also even have, oddly enough, like engineering backgrounds and so they're doing some of this building themselves, right? I don't like necessarily promoting or evangelizing a citizen data scientist because if you build.

Models without understanding the implications of what that can do to the data, then that can create very strong biases or very problematic things. And, and healthcare obviously being a vertical that is very prominently needs to be accurate in regards to their models for, for patient health and longevity.

But there are, there are a lot of frameworks that do exist, whether those are software applications or even. Whether that's GitHub, Nvidia, et cetera. There are a lot of models that already exist and there's something called transfer learning. So if you were trying to say, detect bone breaks or, or build a model for bone breaks and X-ray images.

Those models already exist on the web. And so you could use that initial model and then ingest your own data, the transfer learning, uh, to essentially optimize that model. And so some of these things have basically already been built. And there's also, uh, point solutions, software applications as well that doctors can use that like already have AI embedded in those already have thousand trained.

And so that's. To start this journey. You know, it's, it's interesting as I, as I hear you talking about that there's a, there's a danger in just being clinical in, in not really understanding how the models come together and how bias plays, plays a role in it and not really understanding the, the data science of it.

And if you just go in the data science route and you get somebody who really understands models and building those out, you might build great models, but you may not be connected to the business. That need to be solved. And, and so it's interesting to hear that and say, you know, are, are we trying to find a needle in the haystack here?

Or is this, are these people readily available? Do we have people who are data engineers, who have a clinical background who can step into those roles? Yeah, I, I definitely think there are, and you see people from it segue into engineering roles or even data science roles. You know, my perspective is the, the more math or the stronger the math background you have, the more adept you're going to be as a data scientist.

There are plenty of software applications that can auto build models for you. So, so there's not a huge, huge gap in regards to the complexity of being able to program that. But the reality is that you really do want to have people that understand the data and more importantly, and the thing I care most about is, and I guess that goes back to my psychology background, is how do you make this actionable, right?

If. I can detect something with 99% accuracy, but that doesn't do anything to improve patient care, reduce cost, et cetera, then it's basically a workless model. Right? There needs to be something that can be leveraged within the outcomes of that model for it to be relevant to the organization. Right? We, we hear a lot about AI ethics and bias.

How do you organize in. You know, you're, you're creating ethical AI models and that you're not introducing bias into these models that are gonna be pretty important in terms of making decisions for how we deliver care. Yeah. And, and that's where I think the statistical components comes in, right? If you look at, say, uh, so part of a data scientist time, actually 80 to 90%.

Of a data scientist. Time is not actually spent building the models. It's spent, uh, getting the data and the right data together. It's visualizing that data, it's exploring that data. And so if. You were building models and noticed, I'll, I'll go back to my sociology degree for instance. There was a company probably 10, 15 years ago that was trying to basically build an AI judge right for sentencing people, and I'll keep it.

Brief, but essentially they fed a ton, a ton of cases into this model and built it out. And they found out that this judge bought was essentially racist because it incriminated different ethnicities at a three x rate compared to white. And so when you look at that, and then you look at it in healthcare, you know, how do you, the data needs to be understand in a meaningful way and needs to.

I don't even know where we

the aspect of ethics within this.

Organizations need to be cognizant that race, gender, um, age, the interaction effects of drug medications of those that may have comorbidity of different diseases means that we have to think about these things. And that's where it gets into the of one because we're not thinking about that. Drastically miss something or like administer some prognosis that could endanger a patient's life.

Uh, so that's what I'm most concerned about within healthcare is that a governance group then that we stand up who, who can look at the data and look at the models that we've developed to try to, to, you know, suss out the, the bias and the, the ethical issues. Yeah, I think, uh, there needs to be ethics and governance and, you know, hopefully that's done internal.

It's one of those who will watch the watchers and there's a few different AI groups that are, are working towards that. Credo AI is, is one of them at the forefront, and really that's what I. Care enough about that. I want to make sure that organizations are, are protecting their customers or their patients and, and so yeah, I think ethics and governance around that, continuously looking at those models because they can change over time based on, you know, patients just changing over time.

It's gonna be a critical element. So we're gonna talk about the technology stack, but before we get there, talk about engaging the organization. How do we introduce the organization to AI projects? What leads to a successful program, AI program within a healthcare organization? I, I'd say it's across the board, but you want generally the business unit leader, or if it's a data science team leader that really knows the use cases or the pain points and what the associated cost or value is with those.

You also want it involved because we're seeing more and more data sprawl and data science teams can go rogue where they'll just upload all the data into the cloud, build their models, and it is not being able to track or monitor any of that. So then you have to worry about HIPAA compliance. You have to worry about is the data being deleted?

Is it safe for that data to be there, et cetera. And so, you know, it's good to have that leader involved or someone who really understands the use cases. And it to be in that process, and we're this more on front started.

Have someone who understands the data and understands the math to build those models and just start exploring that, seeing what value comes out of that, the accuracy of those models, and if they're good enough, work on scaling that into production to see what value comes of that. As long as all that goes well, then you can continue to iterate, build additional models, scale up hardware, software, applications that.

All right. Well, we've skirted this question to this point, but let's talk about the technology stack a little bit. But what core technologies do we need to have in place to support ai? Quite a few. Uh, if, if we're talk, if we're talking about full scale production, right? So you need fast storage, you need fast networking.

Or switches and you generally, if you're getting to that deep learning that, uh, computer vision naturally across et cetera, then you're going to want GPU compute. It's still light years ahead of as six FPGAs, any of the other ones that you've likely heard of before. And so that basically means you need NVD cards.

Right now, they're really the only player in the market. Linox switches are great, but you also have Juni, you're. And, and then there's a variety of storage vendors. Dell's been prominent in it. They have also been working with Nvidia and VMware on their VxRail to do the virtualization of all that. Beyond the infrastructure components, then you're also looking at software applications because you're likely going to want a data science platform.

You're gonna want AGPU orchestration tool going to, uh, want something that can handle all of the processes from end to end, from ingestion to production. Uh, a lot of the data science platforms have that now. And if you have great people, then you also may not need a data cleaning platform, but otherwise, there are some automated data cleaning platforms to help on that engineering side.

That's the rough aspect, but I'd say every customer is also different. Do organizations lean towards the cloud in this case, or do they lean in the on-prem direction, or does it really depend on what their current infrastructure, uh, strategy is? Yeah, no, that's a great question. It's multifold, right? So if organizations are already in the cloud, it obviously makes sense for them to stay in the cloud if they're OnPrem and, and all of their data, and they're going to buy GPUs on.

That's perfect, and I often recommend it just because it allows for the data science teams to iterate, right? If you want to continue to try to build models with different parameters to increase the accuracy of those models, then doing that OnPrem is great cost-wise. If you're doing it that in the cloud, you're paying for the GPE compute every hour.

Right. But for organizations that are just starting off or more novel in the approach, if they are on-Prem, they may go ahead and we see this a lot as well and do their training in the cloud. So building those models first with the data science team will be done in the cloud, and then we'll productionize that in the on-prem.

The last piece, I guess, is really that we're seeing hybrid infrastructure where some people have some workloads in the cloud, some on-prem and, and then there are also GPU Colo facilities for offloading those workloads too. GPU Colo facilities. That's interesting. Are, are there any key players, vendors, technology players that have taken the lead with regard to AI in the technology space?

I mean, and NVIDIA's definitely taken off, right? We're, I think, believe they're number two partner, likely because, uh, of the number one sells to all the cloud guys. And, and, and it's because that GPU compute is so prominent. There's, you know, a number of storage, uh, vendors that have been successful in this space.

All the big players that you guys already know. And, and I'd say same for networking. When you look at data science platforms, if, if anyone on here wants to do research or investigate, the, the big players right now are DataRobot, um, an H two ai, which both have auto ML platforms, which means auto machine learning, as in.

It can auto run thousands of models and identify the best one right to Productionize. But beyond that, there's great data science end-to-end platforms, uh, from Converge IO to Paperspace data I to Iio, et cetera. And, um, and, and those are really what most organizations are, are needing right now besides the, the engineering components or a center of excellence to be successful.

Well, you know, one of the things I started to do on these interviews is throw out a goofy question at the end, which is essentially, what's the topic we haven't talked about? What's the question I didn't ask that you're kind of surprised and, and you think the community would benefit from, uh, a short discussion about.

You know, I think that big takeaway is really that as, as many organizations are, are getting into this or wanting to get into this, it's, it's a heavier lift than most people see. You know, when I was a, a data scientist, starting out, the number of models I built that never made it into production, which is.

Sound with the industry is around 80 to 90%. Wow. And when data scientists cost six figures, you want to ensure that you're getting value from that type of investment. And so, I mean, that's part of the reason I, I find these processes so important. But you know, really I just look at how can we continue to iterate and demonstrate value because I see the healthcare space as.

Huge, huge, huge opportunity to, you know, not only increase efficiency for patients in waiting rooms and, and doctors getting to have, you know, better care for those patients, but, you know, my significant other works in, in healthcare and is exhausted every day, um, coming home and I think. What could be done to make these, these doctors, nurses, rts, uh, front desk receptionists, like how do we make their lives easier and how do we improve patient care in the process?

And my go goal or hope is really that leveraging AI in meaningful ways in the future is going to help drive us there. And, and I just love getting to talk in this space. It's, uh, really exciting and love getting to talk with different customers because every single one has, uh. Something unique that they're wanting to do.

One follow up question. How do you, what leads to 80 to 90% of the models being thrown out or not used? Is it a not engaging correctly early on? Is it no champion within the organization? Is it, uh, a lack of belief that AI can really do the work? Is there a cultural barrier that we need to, so, yeah, great question and hit most of.

Usually a small handful of things. One is that there's not processes for a data scientist model to be handed off to, let's say, a machine learning engineer to scale and productionize that model. So that tends to be one of the largest gaps, which is more so just a process, um, an efficiency type thing. But the, the other prominent one that I tend to see is data scientist, and that's again why I always promote that internal AI champion.

Is data scientists love building and loving exploring data. But again, if there's not actionable impact or requests or ask around it, then that can be detrimental. So quite a few times I've also seen that data scientists built this fantastic model for marketing, right? Let's say, uh, customer churn. Okay.

This model, if we put this into production, we can reduce customer 3%, which would be million in revenue, have c. I didn't tell you to do this. Don't tell me how to run my business. And, and so obviously it becomes problematic if, if there's not synergistic alliances and things operate in silos. So really it tends to be data or people process gaps and, and that's about it.

Which size an organization have to be in order to really engage with ai? Do they have to be a certain size or a certain budget or a certain sophistication in order to engage with ai? No, I, I, I'd say not. You know, you look at a lot of these AI unicorn startups and they were investing, I mean, they're built a lot of it built foundationally on AI and, and.

They with five 10 people are investing extremely heavily in it. Meanwhile, you can have Fortune 100 organizations that don't even have data scientists. So it's not really a, a people element gap. It tends to be a, a, a leadership buy-in or, or budgetary thing. And you know, that's where I think, you know, even just that starting at.

Level one in the AI maturity with an AI point solution, some software application that, that solves a unique case, whether that's, you know, using EMR or you know, and to help with billing or these x-Ray and MRI imaging diagnoses. And, and start off with something like that, if. You don't have the, the budget or have constraints to hire a full scale data science team and, and engineers to productionize all that and, and then yeah, see what kind of takes off within the organization from.

Fantastic. Matt, thank you. Thank you for your time and thank you for sharing your expertise with the community. It's, it's really been a fantastic conversation and really appreciate your perspective on it. This great Bill. I. What a great discussion. If you know of someone that might benefit from our channel, from these kinds of discussions, please forward them a note.

Perhaps your team, your staff. I know if I were ACIO today, I would have every one of my team members listening to this show. It's, it's conference level value every week. They can subscribe on our website this week, health.com, or they can go wherever you listen to podcasts. Apple, Google, overcast, which is what I use, uh, Spotify, Stitcher, you name it.

We're out there. They can find us. Go ahead, subscribe today, send a note to someone and have them subscribe as well. We want to thank our channel sponsors who are investing in our mission to develop the next generation of health IT leaders. Those are VMware, Hillrom, Starbridge advisors, Aruba and McAfee.

Thanks for listening. That's all for now.