Rewiring the Enterprise: The CTO’s Playbook for Governance, FinOps, and HLS with Bob Heyl

We are thrilled to welcome back our first return guest, the architect of digital acceleration at CareFIRST Bob Hyle, who now wears dual hats as both chief technology officer and interim chief data officer. With the career forged at industry titans like Leidos and Etna, Bob is spent the last year moving CareFIRST beyond the pilot era and into the phase of true enterprise wide transformation. He returns to the show to help us unpack the evolution of his A3 framework, the powerful convergence of AI, automation and analytics that's currently driving value across his technology ecosystem. We've widened it back to review the hard one hard one lessons from the past 12 months and to explore the new high stakes frontier of multi-agent orchestration. Bob, it's a privilege to have you back from Bob. Thank you David. You know, I think you did a nice job summarizing, you know, what I do I would just say it's great to be back. I still have the pleasure of serving as chief technology officer CareFIRST and really I would say the things that occupy my time right now are how we continue to stay on top of and on the forefront of, you know, a lot of emerging technologies, not just AI, but AI being the one that is that seems to be the North Star Center point, you know, force of gravity if you will that's real lining or aligning everything else that we are doing and definitely have had an exciting year. So happy to dig into it. Well, we're excited to have you back, Bob. You're again being our first return guest, like to be able to go over some of what you talked about last year, see some of the acceleration, deceleration, potentially of some of what you've done. I think it will be really, really interesting and just to be really frank with our audience, you have the most listened to episode of all time on the podcast. So I have a feeling round two will be pretty good. That's a lot to live up to. I look forward to it. All right, I'm going to start with something like you mentioned this last time. It was kind of how you were thinking about time savings and metrics last year or a little over a year ago. Now, right, I've been going to customers and now telling them, you know, time savings is fantastic, but no longer is it a metric that we can really abide by or go to the business with. I'm wondering how you've evolved kind of your time saving metrics. If you've scaled this old metric that we talked about kind of this week and time per person or per piece of AI that you are looking at, if you've scaled that or if you've rethought it altogether as you're thinking about these metrics of the future and how you accelerate that resource. So the short answer is it's still important to us productivity is important, but what we really focused on over the course of the last year or really honed over the course of the last year is I'll call it a set of metrics that really get to what I'll call the hard value versus the soft value related to what we are pursuing in AI. And the reason behind this isn't just because, you know, I'll just say everyone's job is to ultimately yield or produce value. The demand is so great in being able to actually prioritize what you're focusing on or where you're deploying capacity also as a lens that if you're not careful you can be in a perpetual chase of proof of concept and never realizing something in production or another potential risk is spending a lot on ideas without necessarily pulling it all the way through in production. So what we really started to focus on were I'll call them three very material metrics that were the basis of our ROI. The first was specifically around where can we materially affect the cost structure, the cost to perform a specific service function, etc. within or across our operational areas. So that was really a hard you know, S-GNA savings, you know, operational savings that we had to realize, right? Not just, hey, it's great that we freed up some time, but are we actually realizing a cost reduction around that? That was one. The second was for our industry medical cost is also really important. So the trend or the cost of care is important. So how are we affecting the expense relative to basically the cost of care or medical expense? There's a lot of opportunities where efficiency can really materially affect that across the entire supply chain. And then I would say the third was around where we could actually achieve a call at risk adjusted or durable recurring revenue as a growth opportunity, but also bookable because of the opportunity. And what I should say is in each case, it was the solution or the AI solution as the cause, not just a happy corollary, but actually the cause or the driver of those material benefits. And then in doing so, if we could prove that, then we got the benefit of all of the other software savings and we kind of look at our software value metrics, things like experience improving or quality of something or productivity or kind of pick your favorite kind of software value metric. We got benefit for all of that and we could take credit for that, but only if we had one of the other three or some of the other three as the lead. So in other words, to gain permission in booking productivity or to highlight productivity, we had to have one of the others. Absent that productivity is nice, but kind of immaterial to the operation of the business, so to speak, or the business case. So I would say we've scaled our metrics, but I would say most importantly what we've done is refined our upfront starting process, which was really around how we organize a proof of value and how we get to determination of that value in a short two to four, maybe six weeks at the latest, depending on how complex cycle. And if we did not have clear line of sight, not to necessarily throw away an idea, but push it into the backlog again and take on the next. So being very ruthless about that was really, really critical. I love that, Bob, because I'm constantly talking to customers where they're in POC Health, as I like to call it, of I want to see how age of fours or it could be any AI product on the market reacts in a sandbox with a few IT nerds tinkering around to make sure that the results from AI is perfect. I'm hearing a lot from customers around perfection of getting to perfection of some kind of result from AI. I'm curious from your standpoint, when you talked about value and that line in the sand, did you work closely with the business to kind of decide what is the most high value productivity hacks here? Because I am having this conversation every IT person listening right now is thinking, okay, how do we figure out what is actually productivity using AI? Or what could just be, I'll speak Salesforce terms, could be a flow? Why are we using age of fours here? So curious, Bob, on how you approach that? Great question. I would say it has to be in partnership with the business areas. It stated differently. No one likes to have something done unto them, right? It's one of those, it's a collaboration. What I would say in that is for our business partners to really understand or be able to help in the equation, a lot of what we had to do was I like to use the term show not tell so we could show here is the potential. Here are things that are possible. Not we should go and do X. This is the possible. Now if that's possible, how should we actually think about not what we do today and do it better? But what would we actually design different? What would we stop doing? How would we actually enable kind of this technology? Why would we value it? That's a very different conversation than saying, oh, you do something today. For example, I think we talked last year around one of our first use cases where a lot of people started around document processing. It's like, well, great, we can take it, we can read it, we can disposition it, all that kind of stuff. But maybe we need a different mechanism to maybe reduce the amount of stuff that's even coming in in a document. Maybe we just redesigned the service, it would be better and more efficient. One of the reasons why we tackled things in this idea of AAA, of AI, automation and analytic was one, there's a lot of relationship across the three. But also when we start to think about how we put our data and information to use and how we employ these technologies to be very intelligent or insight driven, a lot of times what it does is it leads to a redesign of what we're doing in the first place, start there and then pull through the technology. And you can't do that alone. It has to be in partnership with our business areas. And I would say we have very strong partnership across our four main operational areas. I love that. That's music to my ears. Unfortunately, though, Bob, the reality is a lot of organizations have silo internally between IT and the business. Any advice on how, let's say there's another CTO listening, any advice for the CTO on how to approach that, how to better partner with the business so that AI is valuable and not just a science project. I would say when we were first started, in this journey, now a couple of years ago. We've been doing AI for a while, but I would say when the whole Gen AI front basically hit and everybody became excited and thrust upon us this whole new hype cycle of everybody going down this LLM, a gentick path. We were still in a very traditional, you know, it's a big initiative, very waterfall-esque type of execution, very siloed by business function, and we, like everyone else, started in terms of use case, and a lot of times use case was very specific to a function, not necessarily a process. What we did on that though is it really required rolling up our sleeves and really a lot of one-on-one or kind of dynamic in terms of interaction. It was a heavy time investment on my behalf, on my team's behalf, and a lot of it again came down to, on those use cases, it was again showing, so we were actually able to very rapidly prototype or do those kinds of things. Like we didn't take it to full POC, it was just prototypical, and it sparked the conversation, and then what we asked for was permission to demonstrate. It was no material commitment at a production level, right, at that point, but what it required was a constant engagement and communication strategy, and then it was ultimately moving from those use cases to saying, okay, if we did this here, what other pressure is that going to create in your operational or business area? So for really, really efficient on intake, the next step, are we going to flood them as an example? Or maybe you have a peer in the organization, a totally different operational area, are they going to be upset that you're flooding them, right, in an unpredictable way? How do we actually start to think about this cross functionally? And it just started to raise those kinds of questions, but what it really required is not a conversation about the technology, it really required helping articulate things at where and how this fit into a business process or delivery of a business service, and the pressures that everyone, you know, kind of feels in their particular function or area, discussing how it fit, and then, you know, starting to bridge those gaps, and that was a lot of time investment. You can't get past it without that time investment. That's why I say right now, the probably the biggest limiter to this technology is the human-to-human engagement to permit the technology. You know, it's funny because I love the way that you're thinking about it, right? You are taking it back to the business, removing whatever technologies on the table from the equation, if you will, until we understand the problem. Even when you talk about these metrics, right, removing cost from your bottom line, medical cost trending, or cost of care, overall curbing the cost curve, if you will, achieving risk-adjusted recovery. These are metrics that I feel like we've been talking about in the pair worlds for two decades, right? Let me make the business go. These are the metrics that make the business go. Now we have this new ability to kind of curb these cost curves or bend these cost curves, right, in unique ways, but it's still a function of doing or focusing on those old school things, those legacy things that run our business house. So really, really nice to hear that. One of the things that I'm curious about, I think you described basically like vibe coding before vibe coding was a thing with this rapid prototyping. I'm curious how your team now is kind of doing that across the board, especially with what you're saying, right? You can't give one business unit, thousand X scale, and not give the next downstream business unit similar scale to deal with that new flooding, right? So how is your team or your teams of IT folks kind of committing to this utilizing new tools and really delivering new value to the business? I'll say this is an area that I would say is part of some of the lessons learned over the last year. The best way I can talk about it, and I think even before we started this conversation, David, we were having a background around how far must we go kind of across the organization around raising the bar of AI literacy and AI agency, and there are certain competencies that must be developed, and that is across the board, right? And then it's how does it relate to your function? I'll say IT or technical product delivery is no different, right? And so part of it is helping folks through how does it actually fit into their process and how should we adapt or adopt new processes to help us kind of in the execution? I think most people generally were pretty comfortable out of the gate around the ways in which these tools can help with documentation or creating kind of I'll call it the stub code if you will for some of these things. Those are like the safe ways of getting people to acclimate, right? They've been necessarily trying to always search for something like that. It has like that kind of stuff. It's like they didn't necessarily trust, you know, how far, you know, these things could go. I think though in the process of introducing it, and there are a lot of, I'll just say, engineers who are wired with really need to understand how this works and really need to understand how far I can push it and they did, right? And I think the combination of making it accessible and safe in, you know, in everyone's hands while giving some folks, you know, who are, you know, more wired to push the envelope, you know, safe place to do that and showing what again was possible started to actually drive a lot more adoption. And we did it in areas that were, we started in areas that were maybe a little more advanced and able to absorb before kind of rolling it out further kind of across the organization. And I don't mean advanced in terms of being critical of someone's, you know, importance or their skill level. It was more in terms of where it was right at the heart of some very advanced engineering concepts that sometimes it was just impossible for anyone person to know all of the parts. And so it was a really good place to introduce it because some of the things is great about the AI is that I can pull in all of these other things that you may not be as skilled or aware of. And if you get the team composition, right, you can validate, you know, kind of across that versus just trust. You know, so it's a trust but verify, you know, every step along the way. I would say it has caused us to change some of our controls, you know, in, you know, in that out of necessity, because they were easy to automate away. And so how do you actually force them in? But the, the adoption on the engineering side or the development side is a little easier than the adoption inside of a business workflow or inside of maybe an experience that we're offering to our customer because there is so much more nuance, I guess, you know, to those types of interactions or what makes it good, then I'll just say structured code. But let's take it like one step further, right? Like one of the, I think you're describing like one of the issues that Apple and I see all the time is the science project folks are so deep, right? They're like, they're so far advanced. And the business folks are like, can you just make it so that this field like automatically populates for me, right? Like, like how are you all balancing this or like bringing the IT or the advanced folks I should say, kind of back to earth in a business sense or in an enterprise sense while maintaining kind of like pulling the other folks forward as fast as you can. I'll give you kind of two, kind of two different things. The first I'll just kind of use an example. And then I will talk about kind of a bit of a framework that we've kind of applied. And it continues to evolve, but what's good about the framework is there's parts of it that are actually bearing out as really routine and consistent. And then we're evolving it with the technology. But in the first part, the example, like I'll go back to just even how we enable, you know, kind of that we'll go back to what we talked about a year ago, intelligent document processing. Can we actually take in a highly effective and high quality way any piece that comes in that's been digitized, right? So an email, a document, a fax, you know, any of that kind of stuff. And can we actually a determine what it is, be extract the important information relative to that topic and then see populate somebody's work basket or go do type of, you know, thing off of that, right? That was a very basic function. That started a lead to not all of the really cool stuff you can do with this technology. It really became, okay, if you can do that, can we actually disposition it against? Is there a case that already exists around this? Is somebody already working it? What is this in response to? Where should this be in terms of the work process? Do you always have to start over? It could be more advanced and actually jump to the right queue, right? And drive some automation. Now you could actually say, maybe on the disposition, you could actually create some recommended action. Here's the next step that you should do related to this based off of all the knowledge that you have. And if there is none, start it, right? And kind of continue to work, you know, that is though, it's a new intake. That type of automation and that type of rethinking the workflow instead of being this very sequential step wise, I handle and I pass, and I handle and I pass to being a much more dynamic based off of an interpretation of what's going on with the right kind of verification. Again, it was not even about the technology, it was about how do we organize ourselves to perform this kind of work in the book. what if and then the technology backed it up. We were able to prove that the technology could do it. And we're not done on that front. I mean, that's an enterprise-wide problem. What I would like to say though, is we're scaling quite fast on something is basic as that. Now the framework that I was referring to in the way in which we've approached this in order to kind of balance out all of the different perspectives. Most of the types of interactions that we're talking about around this technology for us anyway is falling into a couple of archetypes. One, and it started when it was just purely a chat-based type of technology to now being much more agentic. It still holds. It's AI working for me. AI is my assistant. Then it's AI as my coworker. It's a team member. It has a responsibility to perform as a responsibility and accountability to me. I have a responsibility and accountability to it. Then it was AI as a coach. AI helping me do better at my job or staying within the performance parameters or standard operating procedures or those kinds of things. It was very informative. Then there's the fourth tranche, which is it just takes work away through brass tax automation and just changes things. That's a lot of the, some of the IT-related stuff like development, sometimes around our environments, other kinds of things. It's not just IT, but it was also in some of the business process too. But if we could say, all right, what are we unknowing? What are we unlocking in each of those archetypes? We started to get the patterns of, here's how value surfaces. Here's how you have to approach the change management or the engagement. Here's how you have to think about some of these kinds of things and it starts to normalize the conversation. Whether you are the most advanced person or person who's new to it, most people can relate to those kinds of interaction patterns. And that has been really, really helpful. The interaction patterns that you just shared those four buckets, funny enough, Bob, I have, we have not spoken until today. I have said that exact same for archetypes to another CTO. So I love that you folks already have thought through what type of agents, what type of work and what type of value that brings. The other big elephant in the room is governance. We work in healthcare. And with the demand of AI and as your team becomes more mature in using, I'm speaking about the business more from here, but mature in using AI. How do you kind of told the line between, okay, the business is asking for this cool use case for AI. But at the same time, we want to make sure that it's reliable, it's safe, and it's governed in a way that makes sense within the proverbs of healthcare. So how do you kind of balance both? - I would love to say there's a pure science to this. I would say I still think we're in the realm of art around this. Here's what I will say. We started with, when we really started to meaningfully engage in all this, we kind of said there's a task force that's related to this. And then we basically said, okay, well, it became a rate limiter, right? And then it was, I'll call it decentralized, put it in all delivery areas. And that was, do we have enough visibility around that? Do we have enough control? And now it's kind of coming back to, I don't want to say centralizing, but embedding in core governance processes where AI is now a central theme and accountability, responsibility and every existing enterprise governance process from risk management through cyber, through data, through architecture, through all those things. What we've established are key accountable individuals and then key, I'll just say gaps that we had and how we're hiring for those gaps and those competencies are training up for those competencies to make sure that all of the intents around this core group or this hub like structure that supports these processes is equipped to deal with them at every step of the way. And so this is from idea intake all the way through kind of execution and those controls that hub is supporting and is embedded in all of those processes. That seems to be working better. I will say, I'll probably come back to you, you know, a year from now and say, we've changed it again. But what I will say is we always start with our responsible AI principles and how they apply. And of course, a big one for us are going to be all the legal compliance elements. But then that's, how do we get started? And the big shift for everyone in this is not to chase perfection, not to chase all this deterministic testing. It was getting to a risk-based testing approach and validation, doing releases into, controlled releases into a subset population, almost doing a different way of approaching A/B testing, but with one being this AI and using it as a way to refine, to train, to improve, but always having, you know, an understanding of what are the thresholds that you have to manage within, how do you dial it up or dial it down, how do you kill it, if it is a problem? That all has to be designed in, but what you're doing now is live learning and live reinforcement under control. I think the other thing that these processes have had to do and learn, which we are still in the process of doing is when you do the traditional enterprise governance in the context of an initiative, it's easy. When you're doing it on an ongoing dynamic basis around continued observability, it's different. And so the shift on these enterprise processes is the accountability of ongoing performance and what does that look like and how are we actually making that visible to those governance processes is where the challenge is right now. I would say there are some things that tooling has helped, but it has been a bit of a gap in terms of, I would just say enterprise-ready solutions that are out there to really help fulfill this gap. We're seeing some promise, you know, there, some are interesting kind of start up, but there's also things that you all are doing that are making it easier to observe. Like I highlight all the time with our friends at Nulesoft, like the agent fabric is a really interesting way to drive some visibility around this and observability from a totally different lens across a variety of technologies. And so I would say there's promise there, but it has not been solved. - You know, Bob, I think what you're highlighting here is like as we shift more and more of our labor more of the work that gets done at the enterprise to AI, to people who kind of understand or manage AI, there has to be a different level of transparency, different level of metrics that maybe we apply to how we measure those things. You mentioned to me before in Big Salesforce meeting, of course, right, that like how we are even coming to you all with our audit trail, right? And how we're supplying you all with different technologies. And you know, it all becomes important to some extent. Can you walk us through and maybe some of the other CTOs, how you think about these like reasoning audit trails, whether they come from your vendors or you need to provide them to end users because they need to do or have the ability to do evaluation kind of at a finite level. - A lot of this does come down to, this is where actually use cases matter. So a lot of this does come down to the use case. So the depth of verifiability and I would say it more importantly, consistency, where like there's some things where we always need the same answer. It'll use a case in point for us. Medical policy must be applied consistently, cannot afford to have it applied differently for the three of us here, right? For the same, we present the same way, we have the same need and we applied the policy in three different ways. That's a problem, right? So we need to make sure that all of the different paths that we're going through interpreting kind of what's needed to being able to then apply the policy. It's where you make in the handoffs between the interpretation and that path to the actual determination. The act of determining is where we often will have to separate the line of, I'll just say different thresholds of quality that must be applied and must be able to defend and be able to evidence. And I would say the reasoning or kind of decisioning trail is really, really important around, here's how we get to a conclusion. But it's, what do we do with that conclusion? Are we automating a decision based on that conclusion and actually applying a determinant or a determination or are we taking it a step further and now applying something else? In some use cases, we'd be fine with automating or applying that, right? But when it comes to some of these bright lines where it's like, we're going to do no harm or we're not going to deny someone access to care or we can't afford to have medical policy or, I'll just say, the science of, that are supporting these various clinical areas corrupted in any way. Those are the scenarios that become the bright line in those use cases. And we need to actually separate the reasoning to support, hey, here is a potential next action from the actual determination or automation of that decision. And so it changes where and how review comes in and who actually is performing or what's performing the decision. And that's part of the automation process. What I will say is, so that's in terms of like, if we're just doing the automation and I would say a lot of the tools are actually really good at providing the evidence needed to support the reasoning. Where it falls down is sometimes in the granularity the consistency of the determination or conclusion. and how it's applied. Those are things that we often take on and actually use our own kind of logging or observability or other kinds of things to augment. And we have to piece it together. What I will say though is there's a different piece to this. And the other piece of it is the evolution of how we are prompting or learning or other kinds of things within the model itself. We also have to be very careful that we are not, I'll just say creating an echo chamber or artificially reinforcing learning or I'll use the term of the day glazing ourselves. Kind of thing. How do you actually smash through that, break it deliberately? There's an art form to this to make sure that you're getting the full perspective because as the models improve sort of their memories, and it's like one of these things that you can start to create kind of an artificial reinforcement if you're not careful. And so that's a different level that we're now having to confront that has been a learning and have a green answer on that one. - It sounds like to me, it's just as important to keep curating your data as it is to apply it to the right place. I'll go back to the policy example, right? The three of us here on the phone. If we were to give it all demographic data, it would say, oh yeah, you all presented differently. Even in the same disease state, same state that you went to, same hospital, blah, blah, blah, blah, but it would say, oh, well, you're a male of 45, but it will start to use all of the things that we give it, and so this concept of, oh, just give it all the data that we have, break all the silos and just let it make all these determinations. I think is this really like false hope, if you will, for having AI run our businesses? Nobody says they want to turn their business, right? But when they give the disease the types of ideas, I think to myself, oh, you want it to run your business? Well, I think on that, it's funny. What's the old analytics saying right around, you got to separate the signal from the noise. I think the problem with this, when you throw all data, what these models can quickly do is actually take all the noise and give you false hope that they are the signal. People's cold, if you will, of the cold and nuggets, right? Yeah, 100%. And so it's how are we challenging that and protecting ourselves from that and continuously learning from that is the part of the art form here, which is why I said it's not quite a science yet, it's still an evolving art form. What I will say though is, what I've been most excited by, actually in this is, and this is not just because I'm on a Salesforce driven podcast, but I actually love the fact that you all are talking about enterprise general intelligence because one of the things that has been challenging is the rate at which all this stuff is evolving in the hands and in the eyes of consumer is great. The tooling in the capability is great. Adapting it for enterprise is challenging. It's all of the things that you have to do that you don't necessarily think through to be truly enterprise grade or enterprise ready, but for us to actually make it truly scale, truly take on a job of its own, et cetera, ask to have those qualities. And so what I'm excited by is, with some of that type of orientation or language, it's like where can we actually start to trust that we are enterprise ready or enterprise grade or we have those enterprise controls at our disposal? That's an exciting evolution. And so what I'll say is, well, it's great to be able to go native to a lot of these models or develop our own type of novel approaches with them, which I think is important, but where we're actually affecting a lot of this kind of core business workflow, I need that predictability and that stability. And so having partners that are actually bringing that to the table at enterprise grade is really, really critical for us. I need to do both and I need to be able to weave them together. And so that's where I'm more excited about what's happened over the last year as the shift to, okay, everyone's starting to recognize to be really effective in something like healthcare. You gotta take on enterprise controls. Okay, if they take on enterprise controls, here are the things that we're gonna do to now filter some of that capability in a different way, really, really important. - I completely agree. And I think to your point around the consumer facing agents and we've got unthropic in OpenAI and in arms race right now with pepper and some Gemini and Grock and Lama. And I can keep name dropping every single model out there. But this is a slight tangent, but related around, you know, having enterprise grade agents versus what's available on the consumer market. So before we hop down folks who are listening, we were talking about our fathers. And my dad uses me actually as his password keeper for everything in his life. And it is quite frustrating because I am a busy person. I may be on this podcast. He actually texted me and I am not going to reply because on my personal laptop, so don't worry sales for his gods. On my personal laptop, I created a digital assistant. And it's me with all of his passwords. And whenever he pings me about what is my password for my YouTube account, the digital assistant actually replies to him. And he loves it. He's a little bit mortified that I've done this to him at this point, but I do love my digital assistant. And I'm curious. I know you folks are forward thinking. Have you thought through actually having digital teammates execute across the organization from an enterprise grade perspective? Or are you folks still on the earlier, I guess, side of things of this learning screens or summarizations or just the basic use cases? So curious what your take is there from a maturity perspective. Yeah, I would say, so we're in different states of maturity. One of the things that I talked about last year and I'll just kind of revisit and we're still on this type of trajectory. We kind of bucket our efforts in the three buckets. One is what are the things that we're putting in the hands of everyone, right? The tools to just help you do you and do your day better. And that's where we put a lot of the chat-based type of stuff in that equation. And that's more consumer-ish than enterprise. Then we have the biggest bucket, which is all the things we're trying to do around affecting our operations and redesigning our operations. That's true enterprise grade. And then the third bucket is what I'll call pure innovation, novel uses, the next horizon of things that we could be doing. And that's a mix of both and sometimes just new stuff we're doing on our own. Sometimes with partners or not, right? So we're kind of bucketed in that way. And I would say the maturity is different in each. I would say we have not really permitted the use of agents yet at the consumer level in the tools partly because we're not, I don't feel like we're at a point yet of maturity or sophistication or agency or AI literacy where it's like individuals don't fully know the or understand the consequence of what they do in their work setting different from their personal life. And so I think there's a, there's, we will get there, but we're not there right now. I think we are further along in the enterprise one where it's actually like for a particular enterprise function, it's easier to actually structure and create an agent for everyone's benefit. And I think we're often running in that space. And then I think in the, you know, in the new novel arena, I think there's a lot of potential exciting things there that I can't really get into. But I would just say that might, that some of those things may end up being, you know, potential game changers for, you know, for the industry. And I'm not saying that it's just us doing that. There's a bunch of things that I'm seeing out there that are quite interesting. And if we can find a way to make them easier to digest in an enterprise, it would, would go a long way. One thing I will say on all of this is the thing that we always come back to as a grounding principle is just because we can do it with AI doesn't mean we should. And we have been really trying to be intentional and purposeful in terms of where we are applying it. Having a focus on not violating any of our core values or principles as an organization, really as a tie to our mission, is it improving our business performance, all those kinds of things. And the list of things that we can do there is immense, right? And that's where we're trying to focus most of our energy. At the same time, we want to empower individuals, improve agency and create force multipliers within our own organization and capacity. So we don't want to lose sight of that first bucket that I was talking about. But I think there's a different way of enabling that, than enabling the enterprise kind of writ large. I want to take that even one step further. We've been talking a lot about prototypes. I would say AI proliferation across your org and other orgs, probably every vendor that you work with in the last year or two, or maybe even three or more, has switched their pricing model three, four, five times. We're in the era of consumption pricing. How have you all adjusted your FIN-OP strategy to kind of, let's say, adjust to the market or fit into what appears to be kind of now the new table stakes or status quo, which is consumption pricing across the board. It feels like-- Yeah. So this is-- Super exciting question, Tim. No, no, no, no. Oh, no, no. No, no, no. It's a very important question. And as a matter of fact, I think everybody, including my account team with all of you, have heard about this over and over again. Because I'm constantly harping on this. We need to make it easier. in general. And what I mean by easier is like take the pricing stuff aside, you know, I'm not debating pricing models. What I'm talking about is it took a while in the shift to, you know, kind of cloud in cloud-based consumption to kind of norm around your three typical drivers of cost, right, which is going to be your compute, your storage, and your transport of data, typically egress, right? Like those are the main drivers. And it took a while for folks to actually understand in their applications or in the platforms that they were creating or using. What was driving those three things? And as they started, and now we're kind of at a point where it's pretty much science at this point, right? It's very tight. We manage it and we can we can we continuously optimize it. And it's a it's it's it's really truly a science is very easy to understand what's going on. You wanted the reason I want to ask you specifically is like I've seen how you observe everything that's going on. Yeah, very sophisticated way to do it. And I it just it's just a great. Historically, our cloud consumption is typically within plus or minus 1% of our forecast, right? Like we are very tight on that. And and that's even and that's after factoring in optimizations. So we'll give ourselves like stretch goals, you know, on that. So like we're tight on the cloud on AI and the learning around like what's actually driving. Well, what are the consumption parameters first off? Everyone's a little bit different on that. Then the second is what's actually driving the consumption. And how does that apply to the stage that you're in in life cycle from like idea and dev all the way through production and what is it going to mean at scale? Because some of the things do like behave like cloud and actually reduce at you know your unit price. As you scale and other things marine flat and so as you scale it's literally you're just adding increments. And so that's part of what we learn in our proof of value is actually what are the cost drivers in the value hypothesis and how we actually try to think about taking it from. Kind of idea into pilot and then we try to actually evaluate that in production. What we write on those cost drivers and has the cost of value equation changed. What it is right now is it is a bit of we're getting more visibility in the AI front and tools are helping drive more visibility around that. Marrying the two together is on us. No one's actually solved that right. And so it's teams kind of doing a bit of. I'd like to say brute force but I think in some ways they're actually using some of this technology to actually help marry some of it together. Which is interesting in in in of itself but it's like the data piece can be can be expensive relative to AI if you're not careful. It's are you moving data around are you trying to consume data in place what is that actually triggering within that particular cloud environment etc. Are there rules or other governors around that like they have to be understanding if especially if it's a purchase platform that is not in our own infrastructure or own private cloud like so. All those things are in the learning process. We do have a lot of visibility more so than we did a year ago. So I like the transparency and the visibility around it but piecing it together and right now it feels a little bit like three dimensional chess. And I think over time it'll get back to checkers but it's not quite there. We hear you Bob everyone everyone feels the same. I'm hoping for not even checkers. I'm hoping for like you know maybe tic tac toe like it's very simple of how price it works. Except for whenever you learn to act to it's always a tie. Like. Oh no. I want to be able to win. I still want to be able to win. You want to win tic tac toe got it got it got it. And it sounds like you know for vendors for Salesforce right we can't present consumption to you all in the absence of outcomes differentiation. Like like you want or you're expecting at some point that will come to the table with that math and not make you all kind of tear point brute force it together. Yeah I mean I would say yes that would be great if we could have that conversation and I think you all and and a few others actually do attempt to do that. But the problem is you don't always own the full stack right. And so the your visibility only goes so far right and so like when when I've had conversations you know with with you all around what like why wouldn't we just create you know an agent and agent force to do this and I'm like well if I did. Like think about like here are the costs that I'm going to incur just in getting data to the point where it can actually even be accessible. Or usable in something like data 360 you know there's a cost component there that you're blind to right and then it's all of the stack relative to the Salesforce stuff that would be a consumption driven to be able to produce that and then it's the pricing of the agent itself. And it's like how does that actually fit into the value of the use case. I actually look at that in totality and be able to say okay here's what I think the cost of value ratios actually going to be should we do it right now sometimes we have to try it in order to inform. That's okay like I'm I'm not saying it has to always be like a front tight predictive but your your lens only goes so far from doing everything in Salesforce it might be easier but when I'm in a heterogeneous environment it's not as easy 100% And I think this is something that every vendor ourselves included or having we have so many teams working on consumption right now Bob and trying to figure out the math years ago back when. Data 360 was genie I famously internally released an enablement video for essays to figure out how genie worked and consumption pricing work then we have advanced considerably from those genie days to now data 360 and age of force and I think it is a work in progress but feedback like this is invaluable and Bob we really appreciate the time today thank you so much. For coming back to the pod to this episode of AI explained for healthcare and life sciences thank you listeners for listening to today's episode and this is apple on your chronelius signing off with David Tamar we will see you all in the next one.