Why CPMAI Matters in AI Projects — with Mike Hyzy

[MUSIC] The AI Today podcast, powered by the Project Management Institute, cuts through the AI Hive to deliver you proven strategies, real-world examples, and actionable insights that drive AI project success. Tune in to learn from PMI hosts and guest experts. Hello and welcome to the AI Today podcast. I'm your host, Kathleen Mulch. And if this is your first time listening to AI Today Then Welcome, I'm so happy that you found the podcast. And if you've been here before, then welcome back. We are now well into season nine of AI Today. So this has been quite an adventure. And it's been a lot of fun to be in this AI space now for just about a decade. So you'll notice that AI Today is now part of Project Management Institute. And our podcast is on PMI.org, officially, which is super exciting. As we continue to integrate into PMI, we'll be interviewing and engaging with various thought leaders, AI practitioners and project professionals to highlight both the AI transformation of project management and the project management of AI transformations. I always love when I get the opportunity to interview guests and folks that are actually applying AI and CPMAI in the real world. So I'm so excited to be joined today by Michael Heisey, who is Vice President of AI Strategy and Product Development at CGI, where he leads AI Products Studio and oversees strategic AI initiatives for Fortune 500 organizations. So welcome, Mike, and thanks so much for joining me today. Thanks, Kathleen. I'm glad to be here. So just a little bit about my background. I've been in Product Development for about 16 years. Eight of those had been focused on AI and machine learning across a variety of industries, finance, healthcare, insurance, QSR, distribution, aviation. I've also written a book on Product Development in 2023. I built PM practices for consulting firms. I stood up Innovation Labs for clients and I've developed AI adoption and Product Development methods now deployed globally across CGI. So here's what actually qualifies me to have this conversation today. I've failed at this badly. Early in my AI career, I ran a machine learning initiative. I exactly the way I'd run every other successful product launch before it. And I watched it fall apart before it even got started. So that failure forced me to rethink everything I thought I knew about managing complex technical work. Besides that, I've also called found it something called the Chicago Future Slan. It's an organization where we bring corporate executives and academic researchers together and we use surrealist and futurist techniques to explore where technology's heading. The reason I did that is because I kept seeing the same thing in the market today. Organizations are trying to solve today's problems with yesterday's methods. And by the time they launch a solution, 18 months later, the problems have changed. So in the age of AI, we always need to be thinking and strategizing ahead. And that's that core tension between how AI projects actually behave and how we've been trained to manage them. And that's what I'm excited to talk to you with you about today. Yeah, you know, I like how you say that. I feel like even the tools of today aren't going to be the tools of 18 months from now, right? Things change so fast when it comes to AI and just life in general these days. So you really have to stay at the forefront of it. And I know that you have been a, I call you the OG CPMAI certified, right? So it's been for many years well before the acquisition of Cognolidica to PMI. So for our listeners out there, you know, and this was also a few years ago, right? When I say Gen A, before Gen A, I made AI really hot and popular, right? So what motivated you to pursue the CPMAI certification? Maybe talk a little bit about how you found it as well. I mean, for our folks that don't know, it is a methodology for running and managing AI projects that was developed in 2018. And now is an official gold standard certification from PMI and so you can find additional details about Michonneotes if you're interested. But what motivated you specifically, Mike, to pursue the CPMAI certification? And how has it changed your perspective on managing AI projects? So I will give you the short answer and the long answer. Short answer is failure motivated me. So long answer was it was 2018. I was leading my first enterprise machine learning initiative. Had a strong team executive sponsorship directly from the CIO on the client side. There's a clear business problem. Had an adequate budget on paper. We had everything that we needed. By the end of the first quarter, we were six weeks behind on an 18-month project. So we're actually building multiple machine learning products about seven, I think, in 18 months. And we'd build a model, we'd chase evaluation metrics, we'd improve our numbers and discover like the underlying data had quality issues. We hadn't caught or there was new data. We had to clean it, retrain it, chase metrics again, find more problems, circle back. So we have data scientists building 80 hours a week. We're spending money, we're generating activity, but we weren't making real progress. And this is my first project with a data scientist. So when I came in, I'm like, designer, engineer, product, check, check, check, who are you two or data scientists? Oh, that's interesting. What do you guys do? And the more that I got to know them, and I was very lucky, the first two data scientists I ever managed, became friends of mine, but they're like, we don't like agile, it doesn't work for us. Like here, let's pull up the hood. This is how we operate. And that's the methodology I knew. I knew agile scrum and then I figured, like I figured out, and by talking to them, it wasn't designed for it. It's excellent for iterating on features, but it's not built for iterating on hypotheses or your raw materials incomplete. Massey data and your success criteria shifts as you learn. It's like the famous line about the unknown unknowns, right? Like we don't know what we don't know. So I sought out a better approach. I started, I think they pointed me in the direction of crisp DM. And through I found that, I said, all right, like this is good, but I was like looking for something more. And I found catalytic cut through Google search and then validated by a friend that I had in Washington, DC, and vouched for it. I dove in and then I found this like six phase methodology and I'll never forget the image. And I found it and I'm like, oh, this is how it works. Like, all right, I could work with this. This is good. So like the emphasis on data understanding and data prep before chasing model metrics, it was like the discipline that we were missing. And it was good enough that like the data science just were on board with it. But then I have to go convince like my stakeholders, the design and the engineering team, like, hey, we're going to work in kind of a different operation. We're going to blend this methodology into what we're already doing. And I reframed the work as scientific inquiry applied to enterprise delivery. So you're testing your hypothesis against reality before you commit to building something. And that was my first experience. Yeah, I love that story. And it, you know, back in 2018, right? So it's the world looked a whole lot different. And it is built upon crisp DM, which was really around data mining, right? And it was developed in 1999. So if you think about that, you go, wow, the Agile Manifesto was published in 2001, which all of this feels like forever ago. But so crisp DM wasn't Agile in it or tips. So you can bring in those best practices, but not be, you know, a true and very rigid in the way that you do things. So one thing that I know our listeners love is use cases, right? Real-world examples. So you touched upon this a little, but can you walk us through a real-world use case of how your organization used CPMAI methodology when running and managing these projects? And also, you know, talk about that measurable business results and the impact that you had on your project. Perfect. So I'll keep going with the first use case and then kind of go to where we are now. So once we adopt that CPMAI, the decision was counterintuitive. We stopped. We pulled back from model development entirely and went through data understanding properly. We found what was uncomfortable. We've been chasing evaluation metrics on a foundation that couldn't support what were built in, no one had validated or assumptions about data quality. So we kind of went back and said, "Hey, this is a new structure." And like CPMAI gave us the framework to spend the time and data prep that maybe we sure can't, maybe not, but a little bit more structure. And then we could say, we're in phase three. Phase three takes as long as it takes instead of that we're blocked again. And, you know, it was kind of like running, you know, one set with the data scientists and then still like having like architecture design on different tracks, but really treating the data park different. So we were a couple, I think we were two months into it before I started implementing CPMAI. We were down six weeks. We're in the red. And I think it was at least a month after that I implemented CPMAI and kind of took control and had some structure to the project that we recovered those six weeks that we'd lost in the first quarter. So we got back into the green. We were delivering functional models. It solved the actual business problems that we'd set out to address. You know, the, the, it was comforting for the team, right? Instead of like spending like unlimited amount of hours, you know, billing, billing 80 hour weeks distressed, tired, like we're like, all right, you know, we have a plan now. Now, you know, I will say, and for a different conversation other podcasts, adoption of the client to these new tools was a different challenge, right? Because they're like machine learning. What's that data science? We don't understand this. But, you know, internally we've extended the methodology and it's not just model development, but now it's AI governments. So initially CPI, I went toward like, hey, I need something to help manage data science projects in a consultant aspect. Because even as agile ever I want to be, I'm a consultant. I've a start date and an end date, right? It's just project based by nature. As I've continued to grow in my career and at places now like CGI, we've a partnership with, with PMI and we're really promoting CPMAI because they have the AI governance built in too. And that's incredibly important. We're looking at big enterprise projects, you know, building and bias testing, compliance checkpoints, ethical review from the beginning, rather than retrofitting it when legal race is concerns at the end. That's not the time that you want to start doing this. So the same phase discipline network for when I started this is now like a governance framework that's built in. And again, it's that discipline. It's having that structure to make sure that we are, we are covering all of our bases. And we're not getting to the end. We're building networks and having to go backwards. And especially like the industries that I talked about that we work in, finance, insurance, aviation. Like, these are all heavily regulated industries and you have to have logs of why decisions were made and you know, what's the accuracy of the models and different things for different situations. So if we're working for an e-commerce thing and our model's like 60% accurate about what hat you want to wear this summer, that's okay. But if we're working in medical and we're 60% accurate on, you know, a diagnosis, that's not okay. So it's really nice to have a framework that, you know, have the best practices in development, but also the governance and all these other tools that help you do the right thing. Yeah, and I like that you bring that up too, right? I mean, yes, for heavily regulated industries, it's so incredibly important. But then for non-heavily regulated industries, these aren't nice to have. It continues to make people feel trust and confidence in using this, right? And that you move forward. And that, because you know, I know, when you said this, I'm like, you're speaking from experience if it's at the end and you have to go back and kind of retrofit and say, oh, this is all the governance steps that we put in place. Or let's kind of put it on as a bandaid on top that never works well ever. So it's nice that you do it throughout the whole project, right? And that it is short iterative sprints and you can continue to iterate on this. So just because you had an 18-month window, didn't mean that you took 18 months to get through one phase of CPMAI, right? You're never going to move the needle forward. I like as well, you know, you have a very international lens on things. CGI is a global organization and you get to talk with many different groups. So AI projects often come with very high expectations and also sometimes big misconceptions as well about what AI can and cannot do. So how do you coach teams and stakeholders to manage these perceptions and set realistic expectations for what AI can and can't do? That is a great question. So I will let you know what I'm hearing right now. Decative meetings, senior leadership meetings across many not just different clients that I work with, but different conversations that I'm hearing. I'll go out to AI meetups in Chicago and abroad and keep something that has been trending recently that's a little concerning. And it's like, all right, well, here an executive or a leader like we have an operational problem. Let's go a gentick and then like, well, what does that mean to you? Silence or you know, we're just going to give all our employees access to co pilot so they could build their own agents. It's like, well, did you think through the security implementations of that and and hear all the other factors or we'll hear that we need to move faster. Let's automate the entire SDLC. Okay, well, which parts all of it timeline few weeks and what's what's a few to you? And then, you know, usually that's where the conversation starts. You know, AI has changed a lot of things, but what AI has not changed is like the basics of strategy. Right? Are you trying like what is the problem that you're trying to solve? So a methodology that I put together that's kind of guided me over the last couple of years is looking at the great strategist as our time. Like, you know, Peter Campo, Richard Romley, Martin, playing to win. So like all these great strays, you then look at what is consistent across them. So gold rad, a theory of constraints. So, you know, every organization has constraints. So identifying what's your aspiration? Where do you want to go? What's the bottleneck that's keeping you from getting there? Identifying the bottleneck? What's the root cause of the bottleneck? How are you going to solve those? So what tradeoffs are you going to make and what you're comfortable with? What are the fitness criteria for those tradeoffs that will guide your decision and then committing to an action holding people responsible? There's some other things that's very high level, but I see a lot of organizations just jump to you know, we're going to fix it with a agenda. We're going to fix it with AI. Well, what problem are we solving? And how do we know we solve that? And now you start to hear that a lot, too. It's like, well, we got to figure out the problem that we're going to solve. Are you solving a mosquito bite problem? Are you solving a migraine problem? Those are two completely different things, right? Mosquito problem. It's a niche. It's an itch. It's annoying. You're probably going to live with it and you're going to go on with your life. But if you've ever had a really bad migraine, you know it stops you in your tracks exactly where you are and you will do anything and probably pay anything to fix that problem. So again, when you have the power of AI, which type of problems you're going to go after and solve. It's not that you're just solving problems, but the validity of the problem and the quality. These are basic questions. Every PM in your audience probably asks them on, has been on a traditional project. But when AI gets attached to an initiative, basic discipline often disappears. You know, executives get excited, timelines get compressed and someone ends up holding a bag of unmet expectations. Well, I heard, right? There's a story that AI makes engineers work 30% faster. So that nine month project that you have is now six months. Okay. Well, yeah. Can you code faster? But then you have the debug time increases. Like there's a lot of different variables. But we're getting there and I truly believe like we're going to have this the speed to market what engineering is going to happen, but we're still working out some kinks. So the coaching I provide is direct like AI file is a scientific method. You have a hypothesis. You test it. Sometimes the data tells you you're wrong. And that's not failure. That's just learning that saved you from building the wrong thing. And it's okay. And if stakeholders expect a linear march from kickoff to deployment, they're going to be disappointed because AI doesn't work like that. So when AI is non deterministic, you can ask a large language model the same thing in five different instances. You're getting five different responses. We're also going to have to be really comfortable with that going forward because that's how AI works. It is non linear work. So when stakeholders see a methodology with six phases and built in iteration loops, I think that's also helpful with CPMI. I begins to calibrate their expectations to reality rather than than vendor marketing. Yeah. At PMI, we have a vision, the M O R E vision for the profession. And so the M is to manage perceptions. And I think that's really important specifically with what you were saying where perception is reality. So you have to meet people where they're at. I also find just like you said, AI is this shiny object. Whatever that looks like to people, right? Because there's no commonly accepted definition. They kind of make it what it is. Some people confuse automation and intelligence. And when we talk about agentec because that is the flavor of the day. So everybody goes, well, just we're going to do agentec AI. And it's like, do you even know what you're saying when you're saying we're going to do agentec AI? Is it a retrieval agent? Is it an, you know, agentec workflow? I mean, these are very different. And the skill set involved is very different too. Yeah, 100%. So I like that you bring that up. And again, that goes back to what problem are we trying to solve? So I like that is a mosquito bite or is it a migraine? If it's a mosquito bite, probably don't tackle that at least not first, right? Tackle those really big. I say, what is it? A real big business problem because it has to be a pain that you're going to want to invest all this money time and resources in. So I think that, you know, that's usually how I say scope it. But I really like that idea because I think everybody can understand. Is it a little mosquito bite? Is it a little pain or is it something that really is going to like knock me out for a day and I'm laying in bed and I can't do anything. And this flows well into my next question, which talks about mindset, right? Because there is a mindset shift that takes place when we're thinking about running and managing AI projects. And especially for project managers, maybe who haven't run one of these AI projects yet. So what mindset shifts do you believe that project managers need to make when working with AI initiatives, especially around things like collaboration with different team members or iteration cycles and risk management? Yeah, another great question. And for the PMs listening, you already have most of the skills that you need. The mindset shifts are about applying them differently. We were having a conversation a couple months ago and I went back and I I read the the CPMAI training because I could through work as we're a partner. It was just nice to go back and it's like the the having the first quarter of the training is just aligning around artificial intelligence and machine learning terms and definitions and the different parts of it. I think that's the first step is understanding the language that you as a project manager are going to coordinate between different teams because collaboration becomes more complex because you're working across tribes that generally don't speak the same language. Data scientists, ML engineers, business stakeholders, clients, legal ethics, marketing, right? Like your job is to be the translator in the room. So when the data scientist says we improved F1 score by six points, you translate that into what it means for the business outcome. Actually that everyone cares about and vice versa. So iteration cycles, they feel different and traditional project management going backwards feels like failure. Your instinct is to push forward show progress and AI returning to an earlier phases often the right move. So CPMAI builds in this explicitly. You might finish the data prep and then start model development and discover something that sends you back to data understanding. That's the methodology working correctly. Your job is to help stakeholders understand that this is the process, not a deviation from it. So I think risk management expands drastically what I was talking about earlier. You're not just tracking schedule and budget. You're managing model risk data quality risk bias risk regulatory risk. The EU AI act is real when we talk about thinking globally industry specific requirements are multiplying governance is an overhead anymore. It's essential and that's where PMs can add enormous value because you understand how to build controls into process rather than both in the amount at the end. And yeah AI tools are making some of this easier, which means the discipline that you bring becomes the differentiator. Yeah, like how you said that to following a methodology right it gives people permission to go back where that's not always intuitive and you just want to move forward with the project and that's actually the farther you move along when you're not ready. The farther you get into trouble. And so I like that you know you bring that up and so for listeners here right it does give you that permission because going a step back actually isn't setting you back in your project. It may be setting you up to succeed. Organizations that want AI results, not just experimentation, right? There's a lot of news that's been out in the market, which I know. The news always likes to, they never really highlight the great stuff because those are boring stories. They always highlight the failures. But if you really want those true AI results and not, you know, the AI failures or the pilots that go into pilot purgatory, right? That's a term that I've heard where people can't actually move forward beyond that. What role do you see frameworks like CPMA AI playing to help make that happen? That is a great point. I love that term pilot purgatory. And here's the pattern. An organization runs a pilot, builds a proof of concept, generates executive excitement, and then nothing. Model never reaches production. The initiative quietly gets shelved and everyone moves on to the next experiment. So experimentation without structure is expensive tinkering. And what a framework provides is the bridge from interesting demo to system in production, delivering business value. It forces you to think beyond the model. What does deployment look like? How do we monitor performance? How do we handle model drift? What's the feedback loop? So CPMA AI includes operationalization as an explicit phase. And that matters because it creates accountability for outcomes, not outputs. And that is just something else that we deal with on a daily basis, an AI, right? Accountability for outcomes, not outputs. It's very easy to get excited about an output, but we want to drive towards outcomes. And here's what I'll say to the audience listening, like this is where your skills become essential. Building something impressive is engineering. Getting something impressive deployed, adopted, governed, sustained, that's project management. That's the gap most organizations cannot cross. And it's the gap a rigorous methodology helps you bridge. Yeah, I love it. This has been such an incredible conversation. I could go on forever. That's why I want to thank you so much for joining. But before I wrap up, I always ask my guests the same final question, have been doing it for nine years. And I love it because everybody I have never gotten the same answer twice, right? Because everybody takes their own personal experience and their own personal ideas around this question. So as a final note, what do you believe the future of AI is in general and its application to organizations and beyond? Well, as I alluded to in the opening, I spent a lot of time thinking about this. And that's why I co-founded the Chicago Future Salon. And it sounds unconventional, but it really breaks people out of when you're thinking about where technology is heading. That's the reason we use these artist techniques of the surrealist and the situationist from the 1930s to the 1960s to really use these creative provocations to get people thinking differently about where the future is heading. Because almost the one thing that is guaranteed for the future is not going to end up like we think it is, right? And that status quo moment. So how do we break out our thinking to look at different scenarios? So right now, I think that we're in the tool adoption phase. Organization our deploying AI's tools, co-pilot's assistance, productivity enhancers, humans remain firmly in control, and we're using AI to augment specific tasks. This is where most enterprises are today. The next phase, and it's arriving faster than most leaders realize, is agent adoption. So we're moving from AI's tools to AI's agents that handle entire workflows autonomously. Kinsey research suggests that AI systems could potentially complete four days of work without supervision by 2027. Think about that timeline. That is next year. Well, I'm going to be on your podcast hopefully. Maybe next January. We could do a year from now. Let's see where we're at. And it's like, yeah, 90% of my job is automated through AI. But what does this mean practically? You have agents reporting to you like people. A Jensen Huyang recently said that IT will become the HR of AI agents. And that's not hyperbole. It's the organizational reality that we're heading towards. Or charts will include both humans and AI agents. HR and IT will need to collaborate on managing this hybrid workforce. Now if you want the optimistic long-term view, Derio Amote, the CON Thropic, published an essay called Machines of Loving Grace, that I think everyone in AI should read. His thesis is that if we get AI development right, we could compress 50 to 100 years of scientific progress into five to 10 years. He's talking about curing most diseases, dramatically extending human lifespan, and solving problems we've considered impossible. Is that optimistic, of course? Is it possible? It might be. The trajectory of capacity improvements in technology suggests that it might be how fast the technology acceleration curve is happening. But here's what I tell executives. Whether Amote's vision takes five years or 50, the direction's clear. AI is not a technology wave that you could set out. The organizations that build the muscle memory now through structured methods, thoughtful governance through learning to lead hybrid human AI teams are going to be the ones positioned to capture that value. So the question isn't whether AI will transform your organization. The question is whether you're going to be leading that transformation or reacting to it. I love it. I love how you talk about what the roadmap looks like. What a year from now will look like. It's really hard to predict what a year will look like. But if 2025 was the year of AI agents, right, is it possible? How is it proving value? I say that 2026 is the year of agent governance. What does that look like? And where you said, okay, if we are going to have agents out there who are doing things on our behalf, how do we monitor that? And I earlier when you talked about how this is iterative, right? Operationalization is a phase, an intentional phase in CPMAI. And you have data drift and model drift. You can't just build these agents and set it and forget it, right? If we say AI isn't set it and forget it, agents are not set it and forget it. And that is something that organizations really need to grapple with. That if you are going to have this look at your workflow, it doesn't mean you're not going to have humans because you still need a human to look at the workflow. Make sure it's performing as expected. And then one thing that we didn't even get into is what does workflow redesign look like? If we want something automated, do you automate it? I mean, this is like RPA of 2018, 2019 era where people would just say, oh, let's just automate that. And so now it's like, let's just build an agent for that. And it's like, but maybe you shouldn't even be doing that step at all. Yeah. And when we get back to it, there's accountability. When we talk about AI agents and that HR thing, it's like, yeah, you deployed a bunch of agents, but it's like, if one of your employees goes and deletes the mainframe, right? Like you are responsible for that. That is your employee. The agents, you have to treat them as your employees that that responsibility bubbles up. And then also our double tripled down on your RPA statement. That's a really good comparison, right? Because it's like, oh, we're just going to automate this process. Well, again, did you think through the second and third level of faxes if you automate this? Did you think through every step? What about the news cases? There's so much to think about. And then that goes also goes back to like prioritization. There's a million problems that you can solve. Are you going to try to solve 100,000 problems? Are you going to focus on maybe one or two problems and you're going to go after those deep and really think through it and spend the time that you need to build out the AI architecture, the way that it needs to be built to actually solve your problem. Yeah. Well, this was wonderful. And you know what? I'm going to take you up in a year. I'm going to re-emploit you on the podcast and we'll talk about what happened and did all these things come true. New Groundhog Day tradition. Right. I love it. I just love it. All right, Mike, this has been such a wonderful podcast. Again, thank you so much for joining me for your wonderful insights and for a really great conversation. Thank you, Kathleen. Thanks for listening to this episode of AI Today. Please make sure to rate this podcast on Apple Podcasts, Spotify or your favorite podcast platform. And if you want to lead both the AI transformation of project management and the project management of AI transformations, check out resources from the project management institute, including the CPMAI certification, our various eLearning courses, and the PMI by AI blog. Until you join us next time, keep innovating.