Enterprise AI Success: What Separates Results from Expensive Experiments

Speaker A: Foreign.

Speaker B: So what separates successful enterprise AI from failure? With stories of token maxing budget blowouts and nothing to show for it making the news on the nearly daily basis, it makes sense to be asking hard questions before you start another initiative. In this episode, we explore what separates those outcomes, the good from the bad. Drawing on experience, leading enterprise AI programs across industries, as well as lecturing on current AI to. Sorry, on AI, I should say, to current and future technology leaders. We examine the organizational decisions that shape long term success from the principles of governance right at the top, all through, all the way through to operating models, team structures and the development of AI capabilities over time. Welcome to to another episode of the AI Forecast, proudly sponsored by the incredible folks at Cloudera. I'm m, your host, Paul Muller, and uh, quick call to action. If you are a regular listener, and I hope you are, please don't forget to like subscribe and share with others who might find the podcast valuable. It helps us tremendously. Well, joining me today is Mark Rizzi. Mark brings a quarter of a century. Gosh, that sounds really impressive when you say it that way, doesn't it? He's got more than 25 years. I like that, don't you? Mark brings more than 25 years of technology experience, um, across a variety of roles. But in specifically he's held AI and automation leadership roles at Carrier Cognizant, Capgemini, the sort of big consulting names that I'm, uh, sure you know and trust yourselves as listeners. He's led enterprise transformation initiatives across financial services, manufacturing, energy and a range of other industries. He also teaches or lectures on AI and machine learning at the, uh, University of Texas at Ox at Austin and the University of Chicago. And he is vice president of AI and automation delivery at Leighton Bridge, where he helps organizations build enterprise AI capabilities. Welcome to the podcast, Mark.

Speaker A: Thank you so much, Paul. It's great to be here.

Speaker B: It's fabulous having you. Um, we're going to get into the topic of AI transformation initiatives and what works and what doesn't. But before we do, I've said this before, we're a bit of a dessert before dinner kind of podcast. We do our lightning round at the beginning. Why not, right? It's in Australia. The water goes down the plug the wrong way and we do everything the wrong way around. Yeah. So, uh, let's kick off. You ready to do this?

Speaker A: Absolutely.

Speaker B: First one, what would people who know you best say is your superpower?

Speaker A: I think professionally speaking. So it's definitely connecting the business with the technology solutions and so particularly with AI capabilities lately, either geni or machine learning. And so what I mean is that the understanding of the business context, what the problem is, um, what the workflow, what the exceptions are about, because they can really be a deal killer. And then pairing that with really the technology solutions with the overall operations and risk involved. And so because there's multiple elements in play there, pairing it with business, with technology, with risk and operations and making that all fit, fit together so that the company or the enterprise actually gets a workable solution that is going to really fit best for them.

Speaker B: Well, we've got the right person for today's podcast then all. Uh, right, next up, the most impactful technology of all time.

Speaker A: Oh gosh. So I won't go the easy route here and say the topic at hand, so I'll actually pick two and the first one that I'll select is the printing press. And so I think with that technology it really spread the capability to enable ideas and information sharing to a mass amount of population and people. And so obviously there is a parallel now with our AI solutions. I think second is probably electricity for me. And so I think electricity is that we sort of take it for granted now, particularly of us that live in modern societies and so forth. And so it's infrastructure that runs on the background that just enables and literally powers so many different things. And this is very close to me because I work for almost 13 years for a utility and energy company and so delivering that capability with people that need machines to live, that support their life. I think that is a particularly high bar as far as infrastructure wise. If we look at some of these newer capabilities of how they could possibly surpass that. But we'll see as uh, time goes by.

Speaker B: Well obviously great parallels to the idea of this um, utility in the background that uh, I suppose many of these data centers are trying to create for compute platforms for AI, but we'll get to that. Mhm. Next up, uh, we talk about AI automation. We're going to talk a bit about RPA and or ah, robotic process automation and whatnot in a minute. But let's forget about the enterprise for the time being. Get a little selfish. What would you like to see AI automate for you in your future?

Speaker A: So I think professionally in our companies is one topic. But really where I would like to see further automation is a lot of the manual tasks and labor that, that we do in our daily lives and at home. And so what I'm talking about is tasks from laundry to washing the dishes, cleaning. So a lot of this is really non value added work. Right. And so I would like to see further automation for those capabilities and tasks. And we had some technology developments there, we've had washing machines and dishwashers, et cetera. But where I want to see and why I want to see that automation is because if we can replace a lot of that manual, very time consuming work. Of course this is really impactful to people's lives. We can spend more time with our children, we can engage in our hobby, we can uh, speak with our parents or spend time with friends, or in my utopia, spend all day playing on the beach with golden retrievers. So

Speaker B: what a great answer. I love that. And uh, I should just say, um, that is probably one of the most, um, we haven't done the statistical analysis, but probably the most frequently, um, cited response to that question. And I hopefully there's a product manager out there somewhere working on a Roomba for my life because I couldn't agree with you more. Yep, that'd be fun. All right, last but not least, you've uh, been someone who's been working with data for decades. Um, what best practices, hints, tips, tricks, hacks have you figured out that help you?

Speaker A: I think definitely best practices, right. And so in my organization now, and to be clear, for the past three years I've been embedded as a Enterprise AI Transformation Lead for one of the top 10 corporate investment banks. And so obviously we deal with a variety of different data entities, terminal data, uh, information from many data providers and so forth. And so we continually deface these data questions, right? And so what I would advise and recommend to pay attention to, there is really some of the fundamental questions, what is the system of record? What is a source of truth? Where should we be getting this data? Perhaps from data source B that could have the same information, but even supplementary information. And then in regards to how we use this data for our AI enabled technologies and systems. And so there's other questions there, right? And they're business questions. How do we parse the data? What is the relevant data? How do we extract the most applicable information and how do we use it for AI empowered systems? And also what is the lineage and what is the explainability that we need to track there? And so that is becoming a really hot topic topic now, especially with all these additional regulations in the EU and the US in regards to data lineage, privacy, etc.

Speaker B: You know, we've been talking about lineage on and off, at least I have for the last sort of five plus years. And it was a new term to me when I I first encountered it, um, I'd say five years ago, maybe a little longer. And it does seem to be having its day. I think it's gone from being a bit of an obscure term of art to something I think even the boardroom are thinking about more regularly.

Speaker A: Yeah, absolutely. Yeah.

Speaker B: Well, look, let's jump into the topic. Um, I guess the winners versus the sinners when it comes to AI, enterprise AI programs. Um, you've. Let's just start with maybe a high level question and get a bit of your background because you've sort of touched on it parenthetically, but you have spent a big chunk of your career working on automation programs and enterprise AI programs. Uh, and you've somehow also managed to carve out a career lecturing, uh, teaching at several universities. How on earth did those two paths come together and how do you keep, how do you sustain it?

Speaker A: Well, yeah, so I would say first off it was really by happenstance. And so because of my professional experience, I was invited or given the opportunity to teach on the side. And so I thought it was a great opportunity. I love to teach. I really, really just appreciate instruction and not to sound corny, but helping others. And so I find it really rewarding. I think also that it gives a different perspective is because not only am I sharing my own professional experience and what I see, what's working, what's not working at many companies, et cetera, that I'm educating students and upskilling them, I'm also upskilling myself. Right. Is because if you want to know something, if you want to be an expert at something, teach it is because then you're on the hook to know it backwards and forwards.

Speaker B: That's such a great piece of advice. So let's talk a bit about, um, enterprise AI programs. I don't know, it shouldn't be any surprise to anyone, but billions, arguably possibly even a trillion dollars is being invested in building out AI data centers, both, um, uh, shared data centers as well as private capabilities, whether it's a model generation or active inference. But there's even more being spent not only on building the capability, but then on consuming the capability. I mentioned token maxing at the beginning. We've set, um, you know, a great example of Goodhart's law, right. When a measure becomes a target, it seems to become, it ceases to become a useful measure. We've got people trying to burn through tokens because it's some sort of proxy for modernity and productivity. So the media has gone from being, I think, treating AI as a Bit of a darling, um, three years ago to now, potentially asking, you know, maybe do we have buyers remorse in the enterprise community? I, I suppose help us separate fact from fiction. From your perspective, obviously one person's perspective. Is the buyer's remorse as bad as the media makes it uh, out, uh, is enterprise wildly more successful than we've been led to believe or is the truth somewhere in between?

Speaker A: Yeah, so I would say it's a little somewhere in between definitely. Of course from an enterprise or company wide perspective is that we are not getting the value that we quite expect. Right. And so I think this is apparent. There is a well known study by MIT that you know, there's practically 90% of AI use cases are not delivering any measurable value to the company and not some, but actually zero. So I think that's something to be aware of however in the past couple years is that we are seeing some particularly gen AI solutions at scale. Right. And so there are a number running in production and they are delivering real results. And so I think um, we're set it on the right path. Right. Of course this is a very expensive investment as you point out Paul. Right. That these tools and technologies that so much money is being spent here but then is the value matching? And so that's the big question. I uh, I do have an optimistic look that we are really just getting started and we just need to further iterate and optimize that. And so I think that is the possibilities of what's going to happen in the near future.

Speaker B: Yeah, it's an interesting conversation because part of this touches obviously on the idea of innovation as well. I was uh, listening to someone the other day talk about uh, the parallels to electricity going back to your alma mater. Mhm, mhm. And they were saying that when electricity was first made widely available, industry didn't quite know what to do with it. So they basically applied it in the same way they'd been previously applying um, you know, whether it was steam power or similar kinds of, you uh, know effectively similar kinds of automation power. And the old, the European term for it is paving over the cow path basically. Right. So in other words. Mhm. You're not building a new road, you're just basically taking an old dirt road and making it into an asphalt road. There's no innovation, it's just a little more convenient.

Speaker A: Yeah.

Speaker B: How much of it do you think is just people applying a new tool to an old problem versus looking for new problems to solve that suit the tool better?

Speaker A: I think that's a really good Point I think, um, that sort of reminds me about paving old pathways versus carving new ways of doing things. And so I think this is very interesting. I spent a lot of time in New York City, um, sometimes in Central park people will take shortcuts to go through a field and so on. And that will completely counteract the sidewalk. But an approach there is not repave the existing sidewalk but create a uh, new sidewalk where people are actually walking. Right. And so I do think that we need to consider that same sort of mindset and same sort of approach with these technologies as long as that design and approach makes sense for so many different reasons. And so for one of how that process or how that workflow flow is being handled, the sound design of that workflow, how it will interact with customers, how employees will interact with it, if it has a proper mechanism to handle many different situations and scenarios that can be emotionally driven from a customer experiencing job loss and they want some concession on their monthly payments, etc. And so these are really very important key considerations to consider as we continue to build these AI solutions and now agentic solutions. And so that will be now further complexity.

Speaker B: Yeah, absolutely. We'll talk about automation in just a second. So I'll ask the question. It's a bit of a complicated one.

Speaker A: Mhm.

Speaker B: Have you seen successful deployments of AI that you would look at and unequivocally say values being created? Part one of the question and part two of the question. Are uh, there some sort of defining characteristics of it that you would say our uh, listeners could take away as being paths to success?

Speaker A: Yes, I would say first off that I have certainly seen experience and delivered successful AI use cases and they're in many different areas and a common one can be a AI powered LLM powered chatbot answering HR related questions based upon an HR knowledge repository. There are credit assessment tools and so um, looking at a customer or company's financial information and aggregating and summary summarizing that data to present to a credit analyst. And so it's not the machine or AI that is making the decision, but it's actually a human. So really adding that sort of key human in the loop. And as far as the second part of your question there Paul, the best practices or advice here. And so there is really so many considerations to consider. So for one, it's obviously the proper definition of the problem, right? Any 90% of the solution is proper problem definition, aim what you're trying to fix, what you're trying to solve, trying to automate, properly define the scope, understand your requirements functionally and non functionally. Right. The performance and response of the AI tool and solution. Making sure your delivery is on par and that you're executing properly. And then also do not overlook training and change management. Right. I think it does not have a high enough visibility with AI projects or solutions. Right. Is that we're used to interacting with these tools in our daily life. We think it's simple and straightforward. I'll add a prompt into ChatGPT or into Claude, I'll get a response. But cascading that message this is what the tool was designed for, this is what the tool was not designed for. And make sure that you assuage uh some of that fear concern and you educate people on the best way and the benefits of using these AI and capable solutions.

Speaker B: All right, well let's um take then the counterpoint to that. What are some of the patterns or Well I guess you know the turn of art is sometimes referred to as an anti pattern. What are the, what are the, what are the patterns that will almost certainly lead to failure? Uh,

Speaker A: well so I think experimentation for, for AI sake and so we're going to add a AI tool. We want to do AI for company X or company Y that it seems like a cool capability. We'll add it and associate it to some of our internal data and we think this is a high value opportunity is that we're going to summarize financial reconciliations and that will be used for regulatory reporting. And so having a very unstructured approach from it from a assessment. So either lack of understanding of the problem, lack of value identification, lack of understanding of the process of all those different exceptions that could happen. And also completely glancing over the, the risk review. Right. What possible risk. Um, do we have hallucinations, do we have uh, customer sort of resilience or um, lack of receptance to this. Do we face organizational reputational risk? And so having that very unstructured approach can really be a recipe for failure.

Speaker B: One of the things I heard and maybe um, I not so much misunderstood but what might not have been something you said but it occurred to me is the sort of fetishization of AI for AI's sake. Uh versus in other words we need to be seen to be doing AI versus saying does the existence of AI allow us to approach a problem that previously was not able to be solved through traditional automation and IT tools in a new way? Is that. Have I heard you correctly?

Speaker A: Yeah, absolutely right. And I understand that there is a speed to market component of this Right. And so companies, if you're a B2C company, or even if you're a smaller enterprise, you're a startup, right? Is that you want to be first to market, you want to be the leading company in your industry. But, but at least establish some base paradigm for what your risk appetite is, what your delivery and your value assessment methodology is so that you make good decisions and at least you have a balanced approach with speed to market, with risk and delivery.

Speaker B: You know, I'm not even challenge you on that. So let's spar for a second.

Speaker A: Mhm.

Speaker B: The idea of first mover advantage I think is sometimes overstated. I mean, you know, you know what I'm going to say next. Right. Apple wasn't first to market with any of the products that it basically has gone to dominate the segments in. But even chatgpt, if you think about it, was arguably the, the name that people use when they talk about generative AI in the popular world. But it's fair to say, I think Claud has stomped all over them by being very focused as a sort of second or third mover, uh, on targeting the coding segment. And it's, it's, it's, if I understand correctly, it's actually even cash flow positive at the moment compared to the rest of its peers. So do you even need to be first?

Speaker A: Not necessarily. Right. I think um, the data as you mentioned supports that and I'm perhaps dating myself, but I remember the first MP3 player which was the Rio and I had a Rio. And of course Apple got into the game and completely revolutionized. Right, that, Right. And so definitely there is strong support to having a cautious approach, making sure that you build the right solutions in the right way. Not going after those high value pie in the sky opportunities, but to start with a foundation you must crawl before you can run. And so look at the low to medium value that are low to medium complexity and look for those opportunities. Opportunities, right. And then just don't do AI, build uh, AI machine learning solutions just for experimentation and that you, you know, have some sort of mandate or expectation that you should.

Speaker B: Okay, that's a great point. So let's talk a bit about experimentation though. So I think one of the challenges that I, and this is my hypothesis that AI faces at present is that sort of peak of inflated expectation, especially if your exposure to it is tangent is glancing. So you know, you're an executive, you picked up Harvard Business Review, you've read it, it says something about AI. You go back to your IT team and say what's this thing, what can we do with it? Oh, I've just been playing with ChatGPT. It just wrote a very interesting sounding document that looks like a lawyer wrote it. And you know, uh, I've been reading in the Wall Street Journal that everyone's getting laid off and being replaced with robots. We need to go do this. It seems like magic without having actually tried to deploy it themselves. So they've got, you know, and I'm talking even from my own personal experience, I had a conceptual understanding, 20, 30 years of IT background. I knew what it was about, but I'd never actually tried to deploy it and use it myself. I was looking at my team thinking and saying to them, you know, what can we do? Where can we go? And it really wasn't until I started to use it myself that I started to understand the strength and the limitations of it. But I think a lot of non tech people have kind of assumed that this stuff is basically flawless. It's a, you know, it's approximate to magic. Do you think expectations are too high versus reality?

Speaker A: I agree with your point. Right? What you're saying is resonating with me is because for me I, I interact with AI a lot. I use AI for coding and troubleshooting and I argue a lot with AI, right? Is because sometimes it just goes off in a completely different, different tangent. And I'm trying to solve a technical problem with AI, right? And then it's fixing perhaps the underlying issue. But I said no, the point is to fix the overall module. Like I don't need you to fix my reg X command, right? Is that we need to fix the entire function. And so I argue a lot with AI, right? But then to your point, a executive that they use it to plan their family vacation and it works great, right? And so why can't we just roll this out and replace L1 customer support? Well, there's a lot more to it.

Speaker B: I once um, uh, was speaking to a CIO of a major company in Australia. This is probably, I'm talking about dating yourself close to 20 years ago now. Uh, and he said that uh, one of the problems with the consumerization of technology is again this um, misunderstanding of the complexities of enterprise. And he specifically cited this, he said one day his CFO walked in and, and in a bit of a mood, not a good mood, M and just said why the heck are we spending X number of millions of dollars a year with. It was SAP. It really doesn't matter which company it is. I just installed QuickBooks at home myself and the CIO is sitting there looking at him going, I don't even know how to begin to process the reasons why that just not going to work for a multi billion dollar company. But, but again that sort of notion of inflated expectations, the idea that the barriers and the hurdles to deployment are uh, trivial to your point governance lineage, there's uh, all of these complexities, hallucinations, materiality of, of mistakes. It's not as easy as going well as you say, other than a, you know, relatively. And I'd even say Chatbot is a high risk situation if it starts to go off on a tangent and make suggestions that may cause legal problems for you, for example.

Speaker A: Yeah, yeah, certainly. And I was just having a similar conversation with a head of treasury for a very large global industrial services company. And so we were talking about employee AI enablement. And so it was about Copilot365. And so he knows Copilot365 is about $20 per user. And so he was asking, well, why can't we just spend $29 per user and roll out to this division and we'll turn it on for 10,000 people tomorrow? Like why can't we do that? Right. Well, there's a lot of reasons, Right. Of why you want to consider other things before doing that wide scale rollout even of a tool such as Copilot365.

Speaker B: Yeah, no, it's uh, this is, and I think to your point, the organizational change management, I think it's actually probably even more profound than uh, I suspect in 5 years time we'll look back and go, oh gosh, we made so many missteps on the way to figuring out uh, how to effectively apply this stuff?

Speaker A: Yeah, absolutely.

Speaker B: Do you think just related to that though, that because you used the word experiment a couple of times and something I try and encourage in the businesses I work with is that it's okay to run experiments provided obviously that you know the, the investment you're making, the, the board and everyone agrees that we're okay with a null result is a result. Right. It tells us it didn't work. Um, we're okay setting fire to that money and there is no expectation, absolute expectation of success. Uh, there's an expectation of rigor around the experiment so that the data it produces useful, but we're not expecting success. Do you think that there is too much pressure on people from a uh, change management standpoint to be seen to be successful whether it's successful or not?

Speaker A: I do. Right. And I think it has to do with a mindset and company culture. I think of course there is a push to get these AI solutions, right. I think there is definitely an expectation to deliver value to the company. Right. And so that there's definitely that expectation and to not fail. But I think failure is part of the equation is because Even experiments, all POCs, uh, are not going to work out and for just reason and so they probably should not work out. Right. And so that's why we do the experiments, that's why we try some of these tests, try some of these new functions in isolated environments. That's fine. But I would also say that building the culture and mindset and that it's okay for some of these investments not to work out and it's much better obviously to fail sooner than later.

Speaker B: Well said. So let's just say we uh, we, we're starting to see green shoots. Things are working in our organization. You talked about change management. One is obviously you know, around the customer, uh, around the people working with these tools. But there's another, I uh, suppose broader change management exercise which is when we think about governance models, the integration into the broader company operating model and cadence, how do we need to rethink governance and what change management program do we need outside of maybe rank and file it?

Speaker A: I would say that sort of governance, education, training, process redesign. And so the. There is a lot to unpack there, right. I think overall from a governance aspect that I think as I mentioned in the beginning it's rather foolish not to have a defined risk appetite or risk awareness. And how are you going to evaluate these AI solutions? And they can either be off the shelf solutions that have embedded AI capabilities and we see this more and more with enterprise enabled tools or if you're building custom tools yourself. But to have that evaluation, that risk governance, those risk subject matter advisors to be in the loop on this. Right. And so this is a bit different now with AI versus some um, historical automation capabilities. And for those of us that have worked in technology, right, like we've always had batch jobs, we could always produce a report, etc. That sort of expands with robotic process automation. But now it's different is because we have non deterministic solutions. And so that is a whole different evaluation paradigm with all these different subject matter M advisors for risk from cybersecurity, from compliance, from legal etc. And so that is something to take into consideration I think just kind of to finish off this point about process redesign, um, what makes sense, how do we capture the exceptions where do we need humans to review the outputs? Um, perhaps we do have some deterministic calls that we can make to a database that is not for human review. But what is the process design? What is the education here? What is the rollout plan collectively across all these AI solutions?

Speaker B: I've talked about this on the podcast once before. Uh, re engineering the corporation, you know, Mike Hammer, James Hampy. Sorry, Champy, sorry. I don't know if you remember, uh, that going back in the day, but business process, re engineering, uh, I feel like it's almost back to the future in terms of this is a business process re engineering problem. It's not an AI problem or a technical problem, certainly.

Speaker A: Right. And a lot of these fundamental business processes and uh, since I work in corporate investment banking, there is a very complex process for customer onboarding and your client, it is incredibly complex with multi, multiple, multiple functional touch points, etc. And so this process itself is being redesigned with AI and agentic AI. And so there's definitely a lot to consider in that application.

Speaker B: Let's talk then a bit about, um, uh, the future. You've spent time teaching AI to students, executives. What are some of the misconceptions that are showing up in the classroom and the boardroom at the moment?

Speaker A: I would say there's a couple. And so first off, I would say certainly focus on the technology. The technology is improving, right. We are seeing upgraded models. We started a few years ago with OpenAI 3.5 and now we're on 5x. Claude has enormous capabilities as well as the others, Grok, etc. Right. And so I think there is a question amongst my students and since I teach at the post grad grad level, this is managers and executives and so they have some expectation, well, why don't we just use the latest model, why don't we just use the latest tool and that will solve all the problems. Right. And it's the same question that comes up in the boardroom. Let's upgrade to the latest model, let's apply the latest model to all the use cases and we'll get the best results. So I think that's one sort of common theme. Um, that is not the right answer obviously. And there's many things to consider. And so for one, fitting the right model to the right use case. And we have Claude's Fable, uh, now, which is phenomenal. But if you have a straight email summarization task, you probably don't want to spend all your money using Fable. Right? And so I think you need to do a Proper fit there. I think the second question or misconception that is common amongst my students and executives is it goes back to the previous point. Isn't this easy? Right? Like what is complex here, and it goes back to what I was saying before is that you have people that are using these tools in their daily lives and you know, it's either um, asking OpenAI for different types of food and restaurant recommendations and rank them based upon proximity, et cetera. And so we use these tools, they are so functionally enabled. Why isn't this easier? And so I think it goes back to one of the themes of this conversation is that having a framework, having an approach, having some sort of discipline, balancing speed to market with governance and risk review, etc. Are just so important in this overall AI delivery landscape.

Speaker B: On that theme, uh, OpenClaw, it's fair to say, ignited and consumerized the idea of agentic automation for a lot of folks. I know many of my colleagues have implemented, you know, in their words, it's, you know, revolutionized their life. It's doing their emails for them and all sorts of stuff. What new skills? Agentic is a good example of it. Do you think organizations should be developing right now to get them ready for the next phase of AI?

Speaker A: More agentic AI solutions are coming, right? And so we have some scale for gen AI powered applications now with agentic AI we're going to have it actually doing things right, making those database queries, um, even making a phone call to book a reservation and then also going through that, planning that, optimizing the routes for continuous improvement, balance with value or cost or um, speed of output of the solution. Right? And so we're going to get more of these agentic AI solutions. The skills that I think we need is for one, at the business level, at the technology level. And so we need the business people that understand their business processes, understand their data, their customer, the exceptions, and to work with the technology people that also have the right skills around all the different services, the different models, uh, potentially pairing that with multimedia capabilities to ingest pictures, uh, video sound, unstructured information, etc. And then also going back to risk and change management. Right? And so we need those risk advisors who should be embedded from the beginning as part of any assessment of an AI solution, the education of cybersecurity, the legal impact of these AI service related contracts, the upskilling of compliance people and third party risk management together with change management. Right? How do we assuage concerns? What is the rollout? What do we need to Train people on what is the impact of the redesign to the organization, what do people's jobs look like of uh, what they're doing now versus what they're not going to need to do in the future and the different value added work that they can add and then they can focus on. Since we're going to be working with the machine.

Speaker B: Last thought on this topic is, uh, you know, let's just give you a hypothetical. You um, uh, bump into a chief executive officer or board member, um, at a social on the beach with your um, golden retrievers. And um, and they uh, they shake their heads and uh, they say, look Mark, you seem to know about this, this AI stuff. I feel like my, my organization doesn't get it. You know, I want to get started with it, but I don't know what to do. Everyone's giving me different advice. Where do I begin, what should I do? How do I get this? Right? How would you summarize it?

Speaker A: First off, I would say it depends, right? It depends upon where you are. So this is one of my favorite managers. I would always come to her uh, with a problem and I would say, should we do A or B? And she would say, well it depends, right? And so she would ask me lots, um, of follow up questions. And so I, I would say it depends, it depends upon where they are, um, what their capabilities are from a technology perspective, from a resource perspective. It depends upon what they've done already, that maybe they've experimented, maybe they have one solution, maybe they have no solutions that are running in production. Right? But then I think coming back to the point is that it starts with a balanced approach, risk speed to market controls, a proper intake and assessment framework paired to value and your organizational priorities, making sure that you have a really firm delivery mechanism. And then sort of kind of last around this out, your education, your change management, your AI enablement. And then how are you going to monitor these solutions, right? How are you going to govern them when they are running in production and they don't bring your business down or critical function down as we've seen. And so this does not happen to your companies like it has happened to others.

Speaker B: I mean, at the risk of maybe trivializing your answer, but I'm not, I'm probably saying the opposite, which is it's, it's called work for a reason, right? It's just plain old hard work. There's no shortcuts, there's no get out of jail free card with this, you've just got to do the hard yards.

Speaker A: Exactly. Yes. It is work and it is diligent and structured work. And the work is different now. Right. The application of AI and having an AI roadmap is not necessarily the same as having a technology roadmap. The fundamentals might be there, connecting it to strategic goals and alignment, looking at your capabilities. But now we have determinist, non deterministic, deterministic solutions. We have monitoring, we have governance right there. There's more really to consider there.

Speaker B: Such a great point. So many more questions for you, Mark, but, um, um, conscious of your time, it's the end of your day. If people are interested in learning more about, uh, either your work or want some thoughts on how to implement enterprise AI, where can they go to learn more?

Speaker A: So you can find me on LinkedIn. And so again, my name is Mark Rizzi. Um, my AI and automation firm is called Leighton Bridge. Um, I particularly speak on these topics through, um, some other networks. So feel free to connect or find me on LinkedIn.

Speaker B: Thank you very much for joining us today. It's been an absolute pleasure having you on.

Speaker A: Thank you so much.

Speaker B: Brilliant. Well, that's a wrap everyone. If you do have any questions from Mark or just comment about today's episode, we'd love to hear from you. Uh, as always, drop us a line in the comment section. We read them all. Uh, or hit UP Cloudera on LinkedIn using the handle Laudera with your thoughts. As always, we've got a great array of guests coming up, so if you have enjoyed today's episode, and I hope you have, uh, make sure to like, share and subscribe so you don't miss another drop. Thanks again to Cloudera for hosting another episode of the AI Forecast. And thanks again for listening in. We will see you all next week.

Speaker A: Sam.