Finding, Evaluating and Prioritizing AI Use Cases with Jacques McGregor (VP Digital Transformation, Marathon Petroleum Corporation)

The change keeps changing. Transformation is hard. The leaders really need to change first. Change management is not done by consultants. Often businesses bury their head in the sand because it's too difficult. Here is the biggest culprit. Everything is changing and evolving. Change change change change change change change change is the only constant in business today. I'm Nliwar habt and founder and CEO of change activation platform Tiger Hall and chair of the executive council for leading change. I've made it my mission to find practical approaches to change that actually work. This podcast is a chance for me to sit down with some of the change and transformation leaders real impact in their organizations and then share those learnings with all of you. In each episode, we'll uncover strategies and insights that you can apply right away in your own transformation journey. Welcome to the only constant. Are you wondering why so many companies still approach AI like it's a shining utoi or a productivity aid instead of a tool that actually drives real business results? Today we're going to cut through the noise and all that fluff together with Jacques Mcregor as VP of digital transformation at marathon petroleum corporation. He has fun time driving AI change where it actually matters on the balance sheet and in the boardroom. Jacques has a great framework for finding and prioritizing high impact AI use cases. So we're going to dive into that and we're also diving into why a Microsoft co-pilot rollout may not be one of those use cases. And a quick note, the views expressed in this podcast are those of the guests and do not necessarily reflect the views of marathon petroleum corporation. This is my conversation with Jacques Mcregor. Many organizations start by asking where can we use AI instead of what problem are we trying to solve? Why do you think that that mindset is still so pervasive? I think it's a great question and I really feel strongly that most companies are approaching this the wrong way is we shouldn't be looking at trying to find a problem to solve with AI as opposed to what business problems are going to move the needle the most for the company if we solve them and then figuring out if AI even has a role to play. But I think the reason why most companies are approaching it this way is you got a conferences or the board is hearing this from other boards and so on that AI is the next, it's the technology to sure and we have to just start using AI everywhere and again I just firmly believe it's the wrong approach. Because one business process that AI may solve really well doesn't mean that that's the best process process at every company to solve. Right, you should really be focusing on what problem is most meaningful to solve at your company and then seeing if AI can solve it. In your review, what do you think the number of percentage of business cases that can actually be solved by AI? Because now it seems like everyone thinks it's 100% of problems can be solved by AI. Do you think those are the 10 or 50 or? I think it's really good to break AI down into three categories and see based on the flavor of AI, what kind of business problems do you have that AI could play a role in? For example, AI could be broken down into classical AI, which is AIML or AI machine learning and that's really good at predicting things. The next flavor is generative AI and that flavor of AI is really good at generating content. And then the third flavor of AI is a genetic AI, which is really streamlining and automating processes. So it's difficult to just blank and say what percentage of processes can AI solve without getting a little deeper and saying what type of AI are you talking about? And certainly when you're talking about back office processes, more accounting, supply chain and so on, agentech AI is a really good opportunity for that. When you talk about generative AI, that's going to be really more in the creative side, the knowledge side, the marketing side, sales side and so on. And when you think about AIML, that's really good for predicting failures, for example, in the asset maintenance space or predicting events or magnitude of certain things like risks and so on. So again, I think it really depends on the type of AI you're talking about. And with AIML or machine learning, it's going to really be well suited towards scenarios where you're doing number crunching and looking at data and trying to find models to predict outcomes. And that works really well in places like maintaining assets if you're in a manufacturing environment. Whereas if you look at generative AI with generating content and streamlining content creation, that's going to work well in legal. It's going to work well in the creative side, marketing, advertising, sales and so on. And then if you look at agentech AI, that's going to really work well in areas where you're trying to stream my highly transactional processes, like accounting, finance potentially call centers and so on and streamline those processes. And some of the scenarios you will blend them together, like generative and agentech, maybe in the IT operation space. So it's really tough to give a percentage of what coverage do you have in all your business capabilities that AI can play a role. I would say it's much easier to break down AI into one of those three categories and then see where it could play a role in your company. But again, I would still say trying to determine first the problems that you're trying to solve that are most meaningful to the company. And then see if AI can play a role to solve those. And the entire conversation has been so focused on just generative AI as well, right, for the last two to three years, for obvious reasons. And then people tend to forget the predictive part and the agentech part. I mean, now it's a little bit more agentech coming into the conversation as well. But I think that's a good point to break it down into those three and not just silo in on generative. Absolutely. I think a lot of people don't realize that AIML has been around for many years since the 50s. Whereas generative AI mainstream has been around for the last two to three years, energetic AI is the new kid on the block right. And so most people are tending towards the generative and it's energetic and often lose sight of the opportunities that classic AI can solve. And I think the business case to be honest, the business cases are easier to craft and hold more weight from a value perspective for the classic AI, the AIML. Whereas the generative might be a little more difficult to justify, the agentech I think there's going to be some clear business cases for that because it's streamlining transactions, streamlining processes helping to automate some of that. And depending on how much a company believes in human in the loop, could really stream on that significantly. So yeah, because the agentech is what addresses the labor market at the end of the day rate. And that could be a big business case in replacing labor speeding up processes and so on. Yeah, it does, but so does the generative right the generative if you think about an attorney, for example. And summarizing contracts attorneys that are using generative AI and their workflow are going to have dramatically higher output than attorneys that don't. And for whatever field you think about, you can extrapolate into those as well. So I think that agentech and generative both have a role to play in streamlining operational costs, labor costs and so on. But I think that agentech is probably going to have a stronger role to play. Whereas most people when they think about generative AI, they're thinking about using Microsoft co-pilot and you can create a PowerPoint, quicker or a Word doc and quicker. But if that's not what you do day in day out, it's going to have a small impact. Yeah, and people have come to equate AI with co-pilot and judge you be T and those are the two. And those use cases, yeah, those use cases are beneficial, but they're not frequent in most people's data and activities. Whereas the scenarios that agentech AI plays are all in are day in day out, highly transactional, highly repeatable. And so I think that the business cases are going to really focus on AIML and agentech. But generative, they're definitely really good use cases, but the frequency is what's going to result in a higher business case. Yeah, that's a good point. And you brought up also that CEOs and boards are usually the ones that go in broad bays, implement everything and just do everything in AI, right? And I'm wondering, what do you think transformation leaders can do to reframe that conversation a bit, to be more a business problem first rather than just AI technology first type of conversations? Yeah, I think that you need to have some structure and discipline around that approach and the approach that I like is looking at the business capabilities and stratifying them between what's going to move the needle the most and result in a marked improvement in your competitiveness versus capabilities that are based business and as long as it's working. It doesn't make sense to invest in. And then looking at for those capabilities that can move the needle the most are they capabilities that have a lot of friction in them. What's the maturity of those do they have a lot of risk associated with those and using those four lenses to identify what's going to move the needle the most and then quantifying that for the board for the ELT and so on to say this is really where we should focus our attention. It's tempting to look at the easy scenarios where AI plays a role but does that business case move the needle the most for the company because certainly all of them can have value but it's what's going to have the most value for the company. Yeah, and what are those four lenses could you double click a bit on the. Absolutely. Yes, yes, if you look at the business capabilities for a company you could strike them into three categories the first is based business so based business might be payroll invoicing things like that and if you invest in those you're not going to get more market. share, you might get a lower cost to operate, but it's not dramatic and it's not going to have a marked improvement in your competitiveness in the market, right? The second is value added where you'll get a temporary bump in competitive nature, but it's not long-lasting. And the third is strategic where if you invest in those capabilities, it will stand you apart from your competition. The second lens is around friction. Is there a high degree of friction in these capabilities? And what you're looking for there is are there a lot of workarounds. Now people working around the system because the process doesn't work very well. The third lens is around maturity and there's many models out there online and so on, but let's just take a model of one through five. Is the capability, low in maturity, it's a one, which means you're running your business on Excel and you heavily dependent on one or two people and if they leave the company, you've got a big problem. Or if you acquire or divest assets, it's a major disruption to the process. Or is the process really mature and it's auditable. Is really good analytics around it. It can respond well to market disruptions or acquisitions or divestitures and so on. And sometimes people feel that the second and third lens are related around maturity and friction. And they are similar but distinctly different because you may have a pretty mature process, but there's still friction and people are working around it because there's still some gaps in the process. And the fourth lens is around risk. And this could be reputation risk, safety risk, process risk, financial risk, whatever it may be. Is does that capability introduce risk or is it introducing risk and you should really address it or if you change it or introduce risk and you really want to look at all four of those lenses to help navigate you to the areas that are going to move the needle the most from the company. And I like this approach because it's a disciplined approach. It's very structured. It's not subjective. And you can also then see the line, the connection of those dots between addressing those capabilities and your PNL or your market share or whatever the metric is that you're driving your optics. Yeah. I find that very interesting. Could you give us an example of like take a process or a company like how would this work in practice? True. Let's take an example of asset management, to asset maintenance rather. And you know, so that'll appeal to most users to call us this who are in a manufacturing environment where it's a pretty, pretty common capability. And on the surface that may not seem like investing in your asset maintenance capabilities might be a big improvement for the company, but if you look at reliability, reliably, it is really important or throughput of product is really important. And that obviously leads to margin capture and that's obviously really important for the bottom line. So there's definitely those connections there. So I would say that asset maintenance is going to be on the higher end of that pyramid, if you like, of if it's a base business value added or strategic capability. And again, it depends on the different companies and different industries as if it's value added or strategic, but I don't think it's going to be base business. Whereas there might be other activities in your manufacturing environment that I just base business. Business is usual table stakes. If you improve in that, your customers might be, yeah, that's not really a big deal for them. But if you can be more reliable as a supplier or reduce your operating costs and that helps you to be more competitive. And what I mean by reducing your operating costs is reduce your maintenance expenditures to maintain your assets more efficiently and effectively than in doing so result in higher reliability that makes you more competitive. So I'd say that that's what moves that up that pyramid of whether or not that's an area to focus on. It's that's important for the company. But then you want to look at is the maintenance process currently a mature process or immature process? Are you running your maintenance on Excel or paper? What do you have systems that are tracking reliability? You mean time to repair? Mean time between failure? Do you have a really good hand on the performance of those assets? And are you using, for example, AIML to predict when equipment may fail? Which then certainly moves that maturity up up the curve? Is there friction involved in the maintenance process? Are people working around the process? Are they not capturing failure codes in the system because you know, you have the capability certain mature process? But people aren't using it because it's not workable. There's a low degree of user experience or adoption and so on. And then risk. You know, you could certainly type personnel risk or process risk to the process if it's not running well. And so if you look at those four lenses, again, depending on the company, it might be a really important area to focus on. So hopefully that helps to set stage for applying a disciplined, structured approach to really isolating the areas of the company that are going to move the needle the most with the company, whereas other processes may not be as meaningful to invest in. Because then to the day, we all have limited resources, both money and time. And we want to make sure we invest in the money and that people's time on the things that are most meaningful for the company. Yeah. I love those four lenses. I think you're spot on. And I think about where does the value actually happen, right? And not just this broad based application of AI. I'm curious about your thoughts. Like if you take these four lenses to a co-pilot rollout, which is what most people can now equate with AI adoption to rolling out Microsoft co-pilot. How do these four lenses show up in a co-pilot assessment? Here, it's interesting because I think you have to also isolate the use cases for co-pilot and not just broad strokes to say everyone's going to have access to co-pilot and have added. I think that's a pretty lazy approach to applying AI. It's also an expensive approach because you're not going to get the return on investment. So I think you need to identify the personas that are going to work in using co-pilot. And what's the value proposition for each of those? But I would also say that if you apply the four lenses, it doesn't really relate as well because you're not isolating a specific capability. It's like saying who gets to have email and who doesn't have email, right? It just becomes part of the way of working. But I will say that co-pilot is a little unique in the sense that for most people, unless it goes back to the comment about frequency, if the frequency is low, but you still believe there's benefit there to use co-pilot. People may make sense to deploy co-pilot, but I think there's a bigger opportunity is that it gets people familiar with AI, whether it's not understanding the principles of AI, so that when you roll out business process specific AI scenarios, people are more familiar with what AI can do for them. So I think there's value in deploying co-pilot regardless, but it being structured around which personas you focus on, which you use to have as you focus on. But it's also an opportunity for your employees to dip their toes in the water a bit and get familiar with AI. So this is really more of a broad strokes approach to AI, before you have more of a vertical approach to AI where you're saying, right, within asset maintenance or within finance or within hiring, talent acquisition, we're going to implement AI that's use case, focus use scenario focused where that's much deeper. And if people have a general understanding of what AI can do for them, I think they're going to adopt it quicker when you start getting into the use case specific deployments of AI. So I do think while maybe the business case is a little tougher, depending on the audience and the personas, I think there's additional benefit by employing a broad view of AI first through co-pilot, has a lead in to the vertical use case specific AI deployments. It's almost like the education layer that comes with awareness stage. Yeah, yeah, here we have a phrase, you've got to pay tuition first, right? So it's a little bit like tuition to get ready for that. Pretty expensive tuition. Yeah, yeah. But it is interesting that I don't want a short change co-pilot because I think there are really some good scenarios for co-pilot. It's like, you know, the legal group and so on or marketing, developing content, they do that day and day out. And this dramatically boosts their productivity, but also the depth of the richness of their content improves as well. So if you're going to use co-pilot once a week to build a PowerPoint side, you're probably not the best example of a use case for co-pilot. But if you're using it regularly, I really believe this value there. Yeah, no, thanks a lot to you, I think you're right. And I'm curious about the lens 2 and lens 3 with friction and maturity. The maturity. Can you assess and measure friction and maturity before you start looking at a process? Yeah, so there's a lot of models online around maturity. Again, the one that I've referenced is 1 through 5 or there's another model 1 through 4. And I think you really need to engage with the users who use the process. And then also sometimes the users upstream and downstream of that process to get their take on how well the process runs. And there's pretty defined criteria on, you know, level 1, highly manual or it's very excel based, it's not in a transactional system and so on. And then the other end of the spectrum, it's highly repeatable, highly audible, lots of analytics around the performance of the cycle time or the defect rate or whatever it may be. So I think that the maturity of a process is pretty easy to measure, but it's going to take interviews and discussions with users. The friction, that's a little more difficult because you basically ask some people, what do they do outside the system outside the process? And often times they don't want to share that information, right? But it's going to take people who do something around who have an appreciation for what's really being done as opposed to what should be done in the system or in the process. And again, it's also usually a Linux sigma term, got to go to Gamber, speak to the people out in the field who are using the process and gather that information and find out how often are people building their own databases or power automate solutions outside the system because the system, the process is not working well or they're running models on Excel because the system doesn't work and so on. And just understanding how often that's happening and that points to friction because the process is not supporting the full needs of the business. We need some transparency there like when talking to people that they actually tell you you do. You do. Yeah, you want to get them to face that for a bit. Yeah. Interesting. And out of these four, like if you could pick one of these four lenses that most companies get wrong, they're going to be like, "Oh, I'm going to get it." when they choose where to apply AI, which one would that be? - For me, for sure, it's the first lens. I think people are naturally drawn towards where's the pain? And if there's a high degree of pain, it's a squeaky wheel gets the grease right. So where's the highest degree of pain? Or friction, let's solve that. Without seeing, if that actually makes a difference for the company, it certainly makes a difference for the person in that process. They're happy that you solve their problem. But was it meaningful for the company? And I think people often skip the first one. What's the most meaningful errors of the company to focus on? And then see if there's pain. Then see if it's this friction and low maturity. Because you really want to find the strategic capabilities that have low maturity, high friction, and there's risk involved. - Yeah. And what is a good way for people to find that out? Like the CEO obviously would know what moves the needle and what doesn't. But like for someone who might not have that or a wide overview and want to come up with use cases for AI, what's a good way to educate themselves on that first lens? - Yeah, I think the first thing is to start with building out a capability hierarchy for the company. And you may start small, you may say, within a business unit, maybe it's within sales. Let's build up the processes and the capabilities in that hierarchy. And then you engage with the business unit leader and say, if we invest in call center, will that move the needle? If we invest in sales or customer relationship management or pricing, which one's going to move the needle the most? And the answer's different for every company based on their maturity, their industry, are they in a commodity market or a non-commodity market? You know, is brand management their area to focus on advertising? And you figure that out and see what's, and so I think it's really engaging with the business unit leaders and their key leaders, their leadership team and understanding what's the most meaningful. Sometimes they'll have a read on the friction and the maturity, but I think oftentimes you have to go a level or two below the leadership team and speak to the people who are actually using the process. And sometimes the process may be run really well in one region but not another. That points to a different issue, right? Or maybe you've grown through acquisition and some systems are being used in one area but different systems in another area. Maybe it's a system issue. The answer could be different. But I think on the first lens, understanding which capabilities are the most meaningful to the company, it's talking to the business unit leaders and their leadership team. The second lens, the third lens, the friction and maturity, I think you've got to go several layers down. On the risk, I think you've got to go back up to the top of the house again and speak to the business unit leadership and the leadership team. - No, it sounds like that you bring a competitive nest as well because I think many are forgetting that at the end of the day, the company's role is to compete in the market and win the market. - Absolutely. - Yeah. - And work customers, right? But I feel like when many are talking about AI, it's about how do we make it easier, saving time, saving this, saving that, but not necessarily about tying it back to how do we become more competitive in the marketplace. - Exactly. - It reminds me of the companies in Silicon Valley many years ago that were competing by having bean bags and table tennis and free lunches. And now many of those companies have stopped doing that because should add improved the employee experience, but it didn't make them more competitive at the end of the day. You've got to survive financially. If you want to add perks for employees, that's great, but that's not what brings in money at the end of the day from customers. - Yeah, exactly. We've touched a little bit on this horizontal versus vertical and there is this big debate right about where the value comes from vertical or horizontal. And I like how you frame the horizontal copilot like the education, initial, like tuition fee to dip your toes in the water. - What do you think is the, from an ROI perspective, is that all in vertical? Or is there any ROI in horizontal? - No, no, I definitely think there's ROI in the horizontal as well. If you are going to apply copilot across the board, but I really think it's having the discipline to identify users and personas that are going to have a higher degree of frequency to use the product, or to use the tools of the capability. Again, if you purchase a whole bunch of copilot licenses and you give it to everybody and you're not really disciplined about how they're using it and so on, I think the ROI is going to be low because the frequency could be low. But if you are identifying users that way, because again, the value proposition for copilot is generative AI, which is generating content. And if you are targeting users who don't generate content very often, you're not going to get the ROI. But if you target users and functions and departments in the company, where they generate a lot of content, where they are summarizing content, creating content, modifying content, reviewing content regularly, then I really believe there's strong ROI there. - It seems like many tie the ROI to like, especially with knowledge workers, right? 'Cause we generate emails and PowerPoint decks and like too many PowerPoint decks and as word and Excel and all of these things. So a big part of it is generative and then of course summarizing, meeting notes and stuff like that. But do you think that people are over indexing on productivity metrics when it comes to the ROI calculation of horizontal AI? - I do think so. Because at the end of the day, it's like when you're trying to sell it, you've got to sell this to the CFOs and you've got to capture their hearts, minds and wallets. And what they care about is, are you reducing cost or increasing margin or increasing market share or something like that? And to say you've increased productivity is great, but in support of what? If you just think, well, I've now made someone's job go from eight hours to seven and a half. Okay, what are they doing at that extra half hour? If they're not doing anything, then you haven't helped the company. I'm sure you've made it a bit of experience for the employee, but the company's not getting more done as a result of that. So it's meant to drive value for the company. And so you have to articulate, you know, to increase productivity in support of X. And what is that X? What is the outcome? If you don't have that answer, then I think it's not a very compelling value proposition. - Nothing more needs to be spent on thinking about what that X is that you're doing instead, right? And I think that comes back to resistance as well to AI because many employees are just being told, like you're gonna save time, you're gonna do less, but no one really tells them what they should do instead when all of that time is saved. - Well, exactly. We always say, - Not gonna give you work like balance, right? So, yeah, and it will, work like balance is important, but oftentimes we say, well, we're going to automate some processes or streamline things to help you to work on more higher value add activities. That's great. What are they? What are those higher value add activities? And so if you've got processes that are really tedious, then you can improve them through agentic AI, for example. That's great. People can go and start doing more analytics. And what you often find is people weren't doing this analytics because they're consumed with a tedious process. So there really is a connection there. But if you just say, well, we're gonna streamline this process and now you don't have to do as much of the tedium, but you don't have an answer for what's next, that's an incomplete conversation. And back to the generative AI, for example, if you've got a legal group and they review X number of contracts today and so on, but you use a lot of outside counsel because you can't process all that content internally. And through copilot, you can increase your throughput and you have less reliance on outside counsel. That's a clear value proposition. You're gonna spend less on outside counsel and you're getting more done internally. And you're also improving the experience for your lawyers internally because they're not having to manually shift through contracts. They're managing together in a quicker by using technology smarter. You're also increasing or improving your cycle time for your internal constituents who need legal to review contracts. So there are definitely examples where copilot has a big role to play and it's tangible value for the company. But you just have to be disciplined about identifying those use cases. - That's so true. - So for someone who's listening to this to wrap up, we wanna go away now and implement and find good use cases, where do you recommend them to start? What other things they can do in the next 24 hours to double down on those? - You know, because the scenario is different for each company, it's tough to answer that. But what I would say is make sure that you clear on the value proposition for each type of AI. There's so much to push to use AI and just find a home for AI. Be more disciplined. Understand that AI typically has three flavors. And understand that AI ML, it's gonna help with predictions. So a lot of data, number crunching, using strong statistical and regression modeling and whatnot. What scenarios fit that well? Separately from generative AI, which is generating content or summarizing content, whether it's video images, text and so on, what scenarios suit that well. And then the third one is agentic AI, which is gonna streamline processes, automate processes and shorten cycle time for processes and then which scenarios fit that well. And what you may find is scenarios may use two or three, but don't look at a specific step in a process, look at an end-to-end process and see how you can make improvements in that end-to-end process with one or more of those three if that process is important to the company. (upbeat music) - AI is not a strategy and everyone else is doing it is not a business case. Jot challenged us to flip the script to start with the already existing business problem and work backward to the solution. And if that means more tough conversations, more discipline and more willingness to walk away from the shiny object, then it's worth it. It just won't serve your organization's corgals anyway. Thanks for listening and see you again next week.