How AI is Transforming Project Management with Alan Mosca, nPlan

Welcome to the B2B Agility Podcast, where we look at the factors that drive success in B2B marketing, with a focus on the people, processes, data and platforms that make B2B brands stand out and thrive in a competitive marketplace. I'm your host, Greg Kilstrom, advising Fortune 1000 brands on MarTech, marketing operations and CX, bestselling author and speaker. Now let's get onto the show. If you could eliminate one of the biggest roadblocks to successful project delivery in your organization, what would it be and why? Agility requires not only the ability to adapt to change, but also the foresight to anticipate it. This means embracing data-driven insights and leveraging technology to navigate the complexities of modern project management. Today, we're going to talk about how AI is transforming project management and enabling true agility in industries that haven't traditionally been known for their rapid pace of change. To help me discuss this topic, I'd like to welcome Alan Mosca, co-founder and CTO at Enplan. Alan, welcome to the show. Hey, Greg. Thanks so much for having me. It's a pleasure to be here. Yeah. Looking forward to talking about this with you. Before we dive in though, why don't you give a little background on yourself and your role at Enplan? Yeah. As you said, I'm co-founder and CTO, so we've been here from day zero. Enplan is, in terms of startups, maybe relatively old now where we've been going eight years, which in construction terms is still a baby, but in startup terms, we were pre-GPTs and all of that. We were founded as an AI company or at the time as a machine learning company. My personal background is I used to work as a quant, so I used to do financial modeling at trading firms. While I did that, I did a part-time PhD in machine learning theory because it was just really starting to beep on everybody's radar and it became so interesting that I got hooked. I got to the end of that and co-founded Enplan with my co-founder, who told me about some of these problems that are happening in construction, and I was like, "This seems like a problem worth solving," and so I ditched my career in finance, which surprisingly actually correlates very well with what we're doing at Enplan in terms of quantifying risk and thinking about different scenarios both on individual projects and portfolios. I did manage to transfer over a lot of my learnings from that, which I wasn't expecting at the time. Great. Yeah, definitely construction. I don't know a ton about it, but a little experience in the large-scale space, lots of moving pieces, and I think that's what we'll talk about a bit here, and want to start by talking about how AI is really transforming those large-scale projects and project management. We certainly talk about AI a lot on this show. I think everybody talks about AI all the time, everywhere, it seems like. But can you share a no-pun intended concrete example of how AI is actually improving project planning, scheduling, or risk management in a large-scale construction project? Yeah. Yeah. We have a ton of use cases. One of my favorites, because we started working with them quite a few years ago, is the TransPennine Root Upgrade, usually shortened to TRU, which is a 10 billion pound, so it's like trades depending on currency billets, call it between $12 and $15 billion, of upgrade works on a route in the north of England. So there's electrification, and there's a lot of complexity to it, because it is also done without shutting down the line, so it's 10 years of work, give or take, that needs to happen. It's very expensive. It's very complex. There's new stations being added, and all of this whilst you keep the trains moving. We started working with the TransPennine Root Upgrade four years ago, and one of the stories that I actually love talking about, because it's very simple to understand, is the TransPennine Root Upgrade has this reporting structure, so construction of rail is kind of centralized. There's a department for transport in the UK that is ultimately in control via multiple delegate structures, but they have a governance reporting system. Every four weeks, the project team at TRU would produce a report, both for governance, risk reporting, and general project controls, like these are decisions, these are the problems that are coming up. This is what we're going to do about it. Making that report took them six weeks, every four weeks. So you end up in this paradox where anybody who's reading information from that report, even if they read it on the day that it's finalized, is reading information that is 10 weeks out of date, which means you can't really change things in the future. By the time you're reading about a potential problem, it might have actually already happened. So we put in a lot of automations through our agent, but also just the nature of the fact that we were able to do a lot of forecasting for them quickly at speed to look at emerging problems, that got down to roughly two weeks now. Wow. Wow. And it's still a very intense two weeks. The project controls team at TRU is, I want to say, hundreds of people just working on this type of stuff and the scheduling and the planning. So we've actually helped them all of a sudden, instead of having two weeks every month where they're working on two reports at the same time, they're going to two weeks every month where they're free and can actually do their job, because the reporting isn't even their job. It is something that I have to do on top. So I felt that is a really good success story and they have their own estimates of how much they managed to save, not so much by saving time, but being able to think about things that they wouldn't have thought about before. There is this thing called possession of track, which actually most circles in the United States is called an occupation, I think. Either way, it basically means that the track is closed for that period of time. Usually it's overnight or on a holiday or on a weekend because you don't want to disrupt commuters. And all the work in planning is preparation to be ready for the moment you start the possession. Because the track possession is planned very meticulously down to the single minute. The rail operator has fines for, no, the rail owner, which is network rail in the UK, has fines that are ridiculous amounts of money. I think it's like 35,000 pounds per minute if you go over an occupation, because you're delaying trains at that point. So everything is planned super meticulously, there's a lot of margin, but you want everything to be ready for that point. And so they had this exercise that they were doing for an occupation over the Easter period. And we started working with them on that, and they figured out that they weren't going to be ready. And these were all the things that they needed to do. So in the end, they were ready for the occupation. But if they weren't, it would have probably cost in the eight to nine digit order of magnitude. Wow. Wow. Yeah. But what you're saying, I mean, in conversations about AI, the efficiency, let's do things more quickly. That's often the topic of conversation, but you bring up a few other really good points of, again, just the free mental real estate to be able to think about things, but also the time to plan and mitigate, either mitigate risk or mitigate, in this case, tens of millions of pounds of potential fines and other costs and things like that. And so, again, it is efficiency, but it can be so much more, right? Yeah. So I like, in general terms, when I talk about AI, a lot of people are talking about like, oh yeah, automation, let's do this thing, but quicker by pressing just a button. And that is cool if you want to charge $20 a month, right, and buy a Chad GPT, but it's more about like, okay, now that you can do this, what is that enabling you to scale up, right? So if I'm taking a forecasting procedure that used to take six months and I can do it in 10 minutes, which is our core value proposition, what does that mean? Well, it means that I can do hundreds of them every day, which means I'm now all of a sudden entertaining this entire space of scenarios for different plans and you're kind of like looking a little bit like low key thinking about multiverses and trying to pick what is the next thing, and I'm sorry if you're not a Marvel fan, but then that's the value. To me, that's how you create the value, rather than I'm going to save 50 hours a month of manual work. I'm like, sure, that's cool, but organizations that spend billions of dollars, 50 hours a month is not registering on any scale. Yeah, it's a blip. What do you think, is that kind of a misconception there? What do you think the biggest misconception is that you encounter when you start talking about introducing AI into project management and how do you address that? Well, the biggest misconception that I find now is that everybody just thinks that when you mention AI, you actually mean charge of ET. I see this a lot in the thought leadership circles within project controls, which is the wider domain that we operate in, where people are giving webinars, this is how you use AI, and it's just a series of problems. That's like 0.1% of the things that you can do. We've developed an army of different machine learning algorithms, plus our own LLM agents, plus our own LLM models and generative models for plans and everything else. Those are the things that, to me, are exciting, rather than how here's a couple of cool problems. The problem is that we're still very anchored on AI equals LLM right now, rather than one of the things that you can really do, thinking like, what's AI in 2027? Yeah. Yeah. I think as consumers start using things like chat, GBT and everything, it solidifies that further. Obviously, AI's been around for decades, but it's as if it was invented in 2022 or something like that. Yeah. I mean, depending on who you ask, how do you get beyond, is it showing use cases? How do you get beyond that initial mental block? Yeah. You need to get them to see it, however that is. We do, obviously, our commercial team does an enormous amount of demos, King, whatever else, webinars, but also we have this MO in our commercial team. We think about our B2B structure, in a sense. We sell to very large enterprises that operate on decade timescales. That sales price in itself is a century basically, but also we set it up so that a pilot for us is never less than six months. So you could say, I'm going to pilot chat GBT for a month or two and see what I can get out of it, and that's enough to make a decision whether I want to pay even the $200. For us, it's more like, okay, you need to run through these loops of planning, forecasting, a couple of times until you start actually flexing a new muscle that helps you think in what if terms about the future of your project, rather than, oh, we're behind on this. And that takes a little bit of time, and there's so many different stakeholders you can imagine. Every time we onboard a new customer, that's about 50 to 100 users, they all have different job roles. So there's a lot of managing that complexity as well. So any pilot less than six months, we're just not going to get to that point where our customers see a return within the pilot. So we'll just refuse point blank to do a three-month pilot, and we'll always push for at least six months. In most cases, it's the year, so it's the first year of usage of the product is called a pilot, and you get to the end of the year, and we can actually then say, these are things that we told you at the beginning of the year, and the ones that you said weren't going to happen, these are the things that happened. So you even end up with a, I told you so kind of moment, which is not fun, right? Because you have to go to someone and say, hey, you ignored me, I was right, right? Clients love that. Yeah. Yeah, clients love being told that their plan isn't good enough, there's an enormous amount of risk, they're not going to hit their planned date, and then I told you so. This is the best ways to close a deal quickly, right? And so you've put all of this together, plus the fact that humans are not built for thinking in probabilistic terms about the future at scale. So the primary point of data that we operate with is construction schedules. So a schedule for a mega project is effectively a Gantt chart that will have between five and a hundred thousand activities, right? If I put this in front of any human being, they're going to at best pick 50 things that are worth focusing on. And so you've left like 99,950 things unobserved that might blow up in your face. And because it's a Gantt chart for a project, it's like a very long sequential schedule where everything's connected together, as soon as one thing goes wrong, everything else pushes to the right. So it kind of ends up in this self-fulfilling prophecy, and that's what we did at the beginning with studying this data, is we found that that's the reason why projects are late is because we have this situation where it's impossible to mitigate everything, it's impossible to know what's going to go wrong, and we're doing our best efforts manually, but you really need like a scale of computation and thinking that can't be done even by a hundred humans most times. So, in that scenario, and so I definitely don't work in construction, I work primarily in marketing, and there's plenty of challenges there, but different time scales and complexities for sure. But one of the challenges that's often run into when trying to introduce automation or other types of AI and is just access to the data that helps feed into those models. So I wonder, what does that look like in large scale construction? You know what you need to be looking at, but is the data readily available? How does that work from the customer's standpoint? The answer is a very unsatisfactory, it depends. So you have a very wide scale of organizations that have everything meticulously organized. Here's our SharePoint, here's our Schedule repository, and you can have access to them so they get loaded into the AI. There will be organizations where everything is on Dave's laptop, and Dave quit ten years ago. Totally. So, you have like those two extremes, and so in one case it's very easy, obviously. In the other case, not so much, but there'll always be something, right? And because most of the time we're working with the individual project, the individual project has its own data setup. So you have this added layer where you have the central organization that's doing lots of projects, but also you have each individual project is basically its own company most of the time, right? If you think about, I don't know, an energy company, Vattenfall that are building wind farms, each different wind farm is a different company that has a different project director and a different team and a different setup, and they decide the tools that they use. And so it gets like very fractally broken up very quickly, and we rely a lot on the users uploading what they have available for the questions that they have that they need solving. There is one step usually that we do push for all the time, which is fine-tuning our models on an organization's schedule data. So if you have been doing projects in the past, you have your schedules, so you have the schedule you started with, and then there's like the monthly updates until you finish, then when you finish, you have a version that tells you this is what actually happened, right? And you can imagine the difference between those two schedules is enormous most of the time, but that allows us to learn that and that's how we do the forecasting. So by using that as a fine-tuning data set for the specific customer, we learn how that specific customer, like how does Meta build data centers compared to Amazon, compared to Google Cloud, compared to Microsoft, right? And they do it in different ways, right? I mean, without even looking at it, it's a very easy guess to make. They do it in very different ways. And so we fine-tune our model with the data that they give us for that so that it performs better for their specificities. Yeah. Yeah. And it is generally very messy. Yeah. So from your perspective and looking at the, I would imagine, if a customer comes to you and they have all the stuff, it's not on Dave's laptop and it's in a centralized place and it's organized and all that, that takes some leadership and some organization on there and before they even come to the table, from your perspective, what for leaders listening out there that are thinking about how to do this, whether they're in construction or not, what kind of leadership does it take to have the kind of culture that is ready for this kind of AI adoption and automation and everything like that? How should leaders be thinking? Yeah. I mean, it's probably not going to be a surprise, right? But we work with very risk-averse companies. Sure. Yeah. And so the more risk-averse you are, you said that most of the work you do is in marketing, right? Marketing is way less risk-averse than construction, which is if you then push it to the extreme, you've got infrastructure, energy projects, nuclear power plants, defense, there's a lot of risk-aversion there if it comes to sharing data, right? So we've had to put a lot of stuff in place like certifications, security clearances, but also the leadership needs to want to explore opportunities and there needs to be like a certain amount. I use the word hunger, maybe a little bit too liberally, but leadership needs to be hungry for opportunities to improve their organization. And that's usually the spark that ignites everything else. Most of our GTM involves marketing creates awareness with the potential users and the users are the people that are not quite on the ground, but they're in the project management team, but our buyer is three, four, six layers above, right? Many times it's the CEO or the CFO that makes a decision to buy. And they only see the results on the balance sheet effectively, right? Or sometimes in some reports that they get and through, you know, third, fourth order of indirect information. And when those decisions happen, that's when we have a successful deployment, when we have the leadership involved wanting to do this. Conversely, you have the opposite, and chief data officers are usually very good, but every once in a while there'll be a chief data officer that appears out of the blue in the middle of an engagement and says, why are you using our data? Our data is worth $2 trillion. So you have to pay us or you're not getting it. And usually then we walk away when that happens and we have done so many times, but that's because there isn't that appetite for, yeah, maybe it's a misuse, it's an abused word, but innovation, right? This isn't that appetite. Look at what can we do better and how far can we actually go? And so those organizations usually are the ones that then show up in the laggard part of the curve when things start going really, really well, right? So like now they're adopting SharePoint and they'll look at ChatGBT in a couple of years once everybody else has figured out what to do, right? Yeah, yeah. So then going a few rungs down the org chart, so to speak, you know, what should leaders be making sure that their project managers and the people, you know, whether it's on the ground or, you know, just managing projects and more mid-level, like what skills should they be having to be ready? Because I would imagine this is also, even if it makes things a lot easier and in some cases quicker, it's still a mindset shift to be thinking about some of this stuff. Like, how should they be preparing for this kind of shift? Yeah, I love this question because a lot of the time we hear from leadership, a version of this question, which is like, "Oh, but how can I change the organization so that the people that work in my group or in my organization are prepared to use AI in the right way?" And actually what I'm seeing is kind of the opposite. And what I mean by this is that, I mean, you have kind of like 50/50 breakdown. You have like the fingers in the ears that I want to hear about it, AI is going to take my job and so therefore I'm going to be anti-AI for the rest of my life, or fully embrace it. It's starting to get quite polarized. There's not a lot of people sitting in the middle. But most projects have both types of people. And so really all that leadership needs to do is give those type A, type B, I don't know, not in a personality sense, but give those people space and that's really all you need to do. Just let them come up with innovation ideas, try some experiments, let them sometimes make some mistakes as long as you put safeguards around them like, "Yeah, the mistake's not being too costly." And good stuff will come out of it because those people are naturally attracted to doing these things. And we get job applications from people that work with our customers all the time. So we'll show up at a client and with our customer success team and our relationship management. And typically out of three, four engagements, there'll be one or two people that send us a CV within six months. And those are the people that are excited. You need to nurture them. That's the only thing that needs to happen. It's not like you need to introduce mandatory training, et cetera. I think that's maybe later, but that's compliance-y sort of stuff. As soon as you make things homework, people are going to just reject that. Good point. Good point. Yeah. I love it. Well, Alan, thanks so much for joining today. One last question for you. I'd like to ask everybody, what do you do to stay agile in your role and how do you find a way to do it consistently? Yeah, I do a lot of things, but I'll try and keep it small and usable and referenceable. So we have had, from day zero, a lot of people now call it an ambidextrous organization. So we have a research team. It's called NERD. Proudest moment of my life was naming it because it's M Plans Experimental Research Department. And they publish papers. They run a thing called Machine Learning Paper Club, which everybody is open to join anyone in the world, where we talk about a new Machine Learning Paper every week. So obviously, I'm a part of that. And then, so I spend a lot of time with them. I spend a lot of time with other founders learning about what new things are coming up. What is everybody doing? I may be weirdly, but I like making myself uncomfortable. We have every six months an event called AI Day, where we try and announce something new and something big. Usually LSAT, for me and the organization, are relatively impossible tasks. So that if we get 50% of the way, that is something that we can announce. And that keeps me, I guess, afloat. Because AI is the fastest-moving discipline, definitely currently, but I think in the last couple of years, probably in the history of humanity, there's something along the lines of a million papers a year being written, so I believe it's... Don't quote me on this number because I haven't done any research on finding the actual number. But the two largest conferences of the year each have about 50,000 to 100,000 submissions. Just those conferences, right? And this is just the research. Think about all the products that are coming out and the code that is being written and all these things. So AI is very easy to get, way overwhelmed. So the other bit is try and stay sane and try and filter out the things that you shouldn't be looking at very aggressively. Well, again, I'd like to thank Alan Mosca, co-founder and CTO at Endplan for joining the show. You can learn more about Alan and Endplan by following the links in the show notes. Thanks so much, Greg. Thank you. Thanks again for listening to the B2B Agility podcast. If you enjoyed the show, please take a minute to subscribe and leave us a rating so that others can find the show more easily. You can access more episodes of the show at www.b2beagility.com. While you're there, check out my series of bestselling Agile brand guides covering a wide variety of marketing technology topics, or you can search for Greg Kilstrom on Amazon. Until next time, stay focused and stay Agile. you