Collaboration within an AI product team extends far beyond internal dynamics, encompassing a diverse spectrum of stakeholders. From engineers and developers to end-users, customers, and even regulatory bodies, this collaborative landscape shapes the trajectory and success of AI-driven products. Achieving seamless cohesion among these stakeholders is pivotal in navigating the complexities of development, ensuring alignment with market demands, regulatory compliance, and addressing user needs.
This week we had with us Paul Pilotte, Technical Marketing Manager at MathWorks. Paul Pilotte boasts over two decades of expertise in technical marketing and development across various sectors such as technical computing, security software, data communications, and test equipment markets. Presently serving as a Technical Marketing Manager at MathWorks, he concentrates on MATLAB toolboxes specializing in statistics, optimization, symbolic math, and computational finance. His educational background includes Bachelor’s and Master’s degrees in electrical engineering from MIT, complemented by an MBA attained from Babson College.
The interview with Paul Pilotte navigates his journey to MathWorks, exploring AI/ML product development and stakeholder collaboration nuances. It delves into leadership differences in AI/ML teams, aligning customer interests, the role of diverse stakeholders, strategies for teamwork, product lifespan extension, marketing approaches, and insights gained in this dynamic field.
AIM: How has your journey been until before we dive into the topic? How was your journey to MathWorks, definitely quite passionate about the world that you’re doing?
Paul Pilotte: I’ve been very fortunate in my career. I am trained as an electrical engineering and computer science major at MIT. I have a bachelor’s & master’s there. For the first ten years of my career, I worked as a design engineer and got some really good experience in doing hardware designs using MATLAB and using a bunch of different Technologies and then my career I found that the reasons why the companies I was working for both large and smaller startup companies and I found that the reasons for success were more about how close we were to customers. Given that I started pivoting my career to say if that’s what will make or break a company, I want to be closer to that. Gradually, my career started moving closer to customers, and I moved into other very technical roles, product major, and product marketing roles because they wanted to be closer to customers. I did that across a bunch of different industries. MathWorks serves many industries, including automotive and so on. I have worked for companies across many of those Industries. I had a mathematical bent and had to use a lot of algorithms and so on. Later in my career, I started working with more enterprise software companies and started getting a passion for satisfying software Enterprise customers. That led me to this role at MathWorks to pull it all together in terms of working with math and algorithms and focus on more enterprise customers across many really cool engineering disciplines and industries. And then the journey here at MathWorks. I started here about 15 years ago. It’s changed quite a bit. In the early days, I was focusing more on specific areas like symbolic math and optimization statistics, and gradually, these areas that I first started are the foundation of AI today. My journey here also extended into more statistics and math and graduated into more data science and analytics and eventually both predictive AI and, more recently, generative AI. I’ve been fortunate to have started more at the foundation level of or the underlying math concepts for AI and graduated almost as the industry and customers did as far as building more data science platforms and AI modelling techniques on top of that if you will. The other interesting thing to your audience may be sometimes it’s interesting, as I taught some customers, that if you want to model something and already know the equation for it, say Ohm’s Law, you should use that and not try to apply AI to that. I’ve been fortunate to have a strong foundation in electrical engineering and domain areas and mathematical Foundation areas. And it’s important for the audience to recognize that some of those older techniques may still be the best way to do things sometimes. In this world of AI and data-driven modelling, we should not lose sight of the fact that if you know the equation for certain things, that might be the best way to build a model.
AIM: What’s your experience building a product that supports AI/ML modeling for researchers and enterprises? Are there differences when leading a team for AI/ML products compared to those without these technologies, especially concerning stakeholders?
Paul Pilotte: The way I think about it, there are some things in this role of leading teams building software technology for users, some things that are common across all product development initiatives. I view my role as a product management leader as having a deep understanding of our customers and collaborating across many development, marketing, and field teams. Some of those things, I would say, are common rights regardless of whether you’re working on software or building products that have AI or not. The first thing is stay close to your customers. Part of my role as a leader in this area is to get close to customers to synthesize customer insights. What do customers care about? I do that by thinking of every opportunity I can get to be in front of visiting customers. I grab those opportunities. I also stay very close to our field teams because they’re at the pulse of helping customers. Another thing that is important is staying current on trends. I try to do a lot of proactive things around. I started reading the media about a year ago. I read the Wall Street Journal every day, I read it and found that it’s a good way to get coverage about how our companies and users think about the business world today. In my journey, I merged the technology background I have with more of a business background.
It’s important to stay current with trends. One way I think about it is to read what our customers read. Many of our customers or leaders read The Wall Street Journal or may read some journals. It’s important to stay current with what customers are looking at, and that’s true for all things. The other area, I would say, is cross-functional collaboration. It takes a team of people to work on products in AI space and Beyond. My role there is to be a good voice of the customer to our internal development teams and engage deeply with our development teams.
The last thing I’ll say is I see my role as being a visible external leader. I do that by working with many analysts and analyst relations and doing interviews like this to try to communicate to a broad audience. And then speaking at external events as well.
AIM: How important is it to also keep track of what customers are reading and their interests? Could you elaborate on that unique aspect?
Paul Pilotte: I see my role as not a technologist and knowing everything about AI. We have a lot of people here at MathWorks who are experts in that. They probably have Ph. D.s in those areas, but it’s more understandable what our customers need. I read things like IEEE Spectrum, which tends to be the reputable engineering Journal that many of our customers read; SAE is more of an industry journal for automotive. I try to immerse myself, and sometimes I imagine knowing my experience if I worked for one of our customers, what I would be reading and thinking about. Itleads you to an interesting thing: We have customers in the 5G Wireless area, and they’ve approached us for help with how you build or employ AI and 5G communication systems. So to do that, I need to know enough about 5G Communications, and I took some courses in college about that. But I feel like I have a good foundation in AI modeling, but I am more interested in understanding. What’s the trend around 5G Communications about automated driving or software-defined vehicles? I probably spend more of my time thinking about trends that our customers care about, and then my AI thinking is more, imagine if I were part of the customer’s team. What would I be doing in that environment? And that’s a good way to have empathy for your customers, which is the key thing that I’ve always thought about: empathize with your customers, trying to understand their problems and what they’re trying to do. And then what’s interesting about that too is that many of our customers are journeying from being customers who built systems like a 5G wireless system in a car. They’re looking to add more and more software to that and Automotive companies have had software and cars for decades but now there’s more of a push towards software divine vehicles around having the car be almost like your phone. You buy the car, but then some baseline things are included as part of it, like the transmission these days and the battery system. But other things like the infotainment system could be programmed, if you will, as a software option; some software things might be included as part of the system like a driver assist system or an automated Park assistance Software System. But then other things that automated Driving Systems could be optional add-ons, if you will. Many of our customers are also using AI but rethinking what it means to be a provider of engineered systems with many software components. And then part of that drives our customers to think about things from the requirements level up, if you will. So that’s another thing that I try to do. I try to imagine what our customers are going through, and many have AI as one of their strategic priorities. Some customers call these big bets, but you get me. What I do is I’ll often look at our customers’ investor relations pages and their strategy and again try to immerse myself into their thinking.
AIM: In the process of building a product enabling AI development, what crucial aspects and stakeholders, besides product management, play a pivotal role in ensuring its success? How do various elements like business development and other stakeholders contribute to this journey?
Paul Pilotte: Let’s start with customers. The end customers are one stakeholder, and then working back from that is our field engineering sales teams, who’re very close to those customers. Being close to them and because we have a global field team, it takes some time as a product manager to build relationships and get to know field Engineers across the globe. Still, it’s also very enlightening to understand whether the field Engineers supporting Automotive customers in Germany here are the same things that our teams in Japan or Michigan or in the valley with some of the newer EB companies. That starts with our customers and works back with the people serving our customers and then internally then becomes our software development teams, and then for us it’s both the teams that are building our AI technology and then for us we have other adjacent teams that are building other domain-specific tools like signal processing and image processing. We have the notion of a development program where we coordinate across multiple teams. Some teams are developing, for example, machine learning or deep learning modelling tools; others are providing more signal processing and image processing tools, while others are more dedicated to applications. Internally, there are a lot of stakeholders, who are in our product development teams, and there’s the product development leads in engineering, then the product management leads across other teams as well. Coordinating across all those teams becomes part of the interesting product management challenge and opportunity.
AIM: What strategies have proven most effective in fostering coordination and teamwork among these diverse stakeholders to align everyone’s objectives successfully?
Paul Pilotte: We call them programs, but it’s collaborating on planning your main goals, if you will. The only thing is that sometimes things change over time. An example is that five or six years ago, when we started focusing on AI, it was more about our customers trying to understand these tools, and, say, deep learning was taking off in those days. Early on, our customers just wanted to understand more about these tools and have some ways to get started. Our development team is focused on getting started materials around making it easy to do this with just a few lines of code or building apps to make it easy for engineers to begin learning AI tools and techniques. They’re both a lot of examples. In the early days, when customers would just get started and want to understand the technology and often had more of a technology focus, people wanted to know what deep learning is. How can it be used? In those early days, it was mostly about using it for images. But then, over time, I’ll bring this back to the customers, which was very successful for several years. We built a lot of on-wrapping materials for both commercial and academic users.
But then, over time, people knew enough about it that they moved towards wanting help for us to help them with real applications. And then we pivoted towards focusing more on applications of AI, and as a team, we said, what are the top applications? And we narrowed it down to several of those. Examples like victim maintenance are something that many of our customers building engineered systems need to ensure those systems stay operational. That was one big Focus. The other was moving with the market after people became familiar with this technology; then, it was more about how you apply and get value from it. Nowadays, we’re focused more on the applications that our customers care about. It’s also important as a leader to have some overarching themes to tell an inspirational story internally, and we use words like engineered systems. It might be easier to describe an engineered system for your audience. If you think about it, a wind turbine has several mechanical components, like the blades. They have a motor; they have a gearbox. It’s a system of different electronic and mechanical components that can be modelled with multi-domain modelling tools, which we have. But then, together, there’s a complete system whose role is pretty complex. It’s to harness wind energy to create electricity. We’ve thought about a lot of engineered systems. Then, that’s an overarching category, meaning our customers need to use AI with simulation.
An example is the maintenance of a wind turbine or automated Driving system. This is another good example where you have some software technology, some of which may be non-AI. Still, we build a lot of AI technology for automated driving systems for modelling and cool applications. From an engineering point of view, many of our automotive customers are moving towards designing and selling EVS. It’s critical to know the battery state of charge on your cell phone; you have the bars and know when your cell phone is charged. The same thing is true in an electric vehicle. It takes work to know an EV vehicle’s battery state of charge. It turns out, working with our customers, that if you take many of the measurements already available in a car, you cannot use a deep learning model to predict the state of charge in an EV vehicle actively. And one of our customers is using deep learning for that. But an interesting thought was that they took a very methodical engineering approach, saying they wanted to integrate this deep learning model into an existing multi-micro control unit on a car. They knew that memory footprint, and they had a 100 milliseconds total budget. It was an engineering problem to say I wanted to build a deep-learning algorithm to model the state of charge. But what are the fine requirements, like how much time can you allocate to making that computation and how much memory footprint can be allocated to that? It’s like using AI but making a more systematic design engineering approach. We see a lot of our customers doing that. As part of that, there are the design requirements upfront, building the AI model, integrating it into the overall battery system, and then simulating it and ensuring that all those requirements are met. This is the way of the future. Thinking of AI as just another design component for a design engineer to use to build an overall system.
AIM: In the context of AI-driven products that facilitate customer AI/ML model-building, how do you define and extend the product lifespan? What strategies are employed to enhance and prolong the relevance of these products in the market?
Paul Pilotte: Another story about that. When I got started, we had products focusing on statistics. It turned out that many of those products did machine learning; it just wasn’t called that. I found that a lot of products often have a life cycle where maybe they have capabilities that your end user needs and your customers need, but they may not be called the right thing. It’s hard for customers to understand it, or it’s missing other stuff. One thing we’ve done that could be useful for others out there on similar journeys is we had a product called a statistics toolbox. But because it had a lot of machine learning capabilities, we focused on making it easy for engineers to use by building apps to make it easy to select features and graphically pointwise build a machine learning model.
Then, we renamed the product and changed its personality to statistics and machine learning. Part of it is the marketing focus on your audience. You might call it repositioning, but it’s morphing a product to say it has all the foundations, but your customers are looking for something a little different. In the other part of that journey, We’ve also focused quite a bit on building reference examples for applications with machine learning. For example, we have another product that focuses on remaining useful life estimation.
But then we’ve built some reference examples for gearboxes or bearings. We’ve worked quite a bit to not just have the underlying capabilities but have complete examples that customers can sometimes use out of the box because it meets their needs or gives them a notion about how they might adapt it to their needs. Products have an evolution, but in some cases you might say that that product is at its end of life if you will, but what I’ve seen is it’s more that it has to morph into maybe something newer if you will or closer to what customers are looking for. Another example is that we have a neural network toolbox, which has been around for probably 20 years, even the AI days for many of our customers. We’re using their networks for embedded applications.
However, we then recognized that the modelling technique that became more promising was deep learning. We shifted our focus from classical Neural networks to deep learning. From then on, we started shifting our investment towards deep learning and supporting those algorithms for images first with CNNs and then LSTMs for more time series data. Responding to what you’re hearing from customers in many products and technologies can be shifted closer to what customers are looking for. That’s been my experience. That’s less about them not becoming useful. It’s more that it needs to change and more to be close to what customers are looking for at that particular point in the technology life cycle.
AIM: How do you create an effective marketing strategy for AI/ML products that reaches diverse customers, considering MathWorks’ experience working with various stakeholders like enterprises, researchers, and academics?
Paul Pilotte: Some things apply to all of your stakeholders and customer segments. When people want to learn about deep learning and AI, that is true across the board in both academic professors educating and commercial customers. Then we had a marketing strategy to get started on types of artifacts and tools that worked well across our industries. When things morph towards people applying this, it becomes different. There were some industry applications, and when we focused on applications, they were relevant for different Industries. Automated driving clearly for automotive and visual inspection is another focus that maps across many industry verticals. Then, some become more specific, like wireless communication, which is a big industry for us so doing something specific there made sense.
And academia became understanding of who your personas were. We have educators and researchers and then students, and we understand what each needs. Then we built a lot of onboarding materials for educators and curriculum materials for educators as well.
Our marketing strategy expands that as well. Some of our marketing is common across. Some things span across, for example, engineered systems that many educators care about. It’s a good way for students to know about it. You need to understand AI but know how to apply it in a particular area. Learning about automotive and mechatronics is important to get a job at an EV company. And the same message, tweaked a bit, also works for commercial customers. I talk about how the oil and gas industry can shift towards serving more clean energy by using AI for engineered systems. There are some commonalities, which I’d say first.
The other thing I appreciated as part of the Gen AI Trend is the importance of your Community Influencers. It turns on in Gen AI, and that’s also a focus for us. The first users who told us what they wanted and became very enthusiastic were some of our community leaders and someone in academia and commercial. And that’s a new way of marketing today around tapping into community influencers. They were the first ones, and part of my role as a leader in the marketing space was, again, to be close to customers. I often volunteered to talk to any of our users leaning into generative AI. It was ChatGPT in that case, and one professor who teaches physics used to in the older days; he would create many videos on using MATLAB to give the students the foundations they needed for math to be able to attend his physics class. And now we morphed and said he built a lot of cool stuff using MATLAB through check GPT, and now we shifted his thinking towards inviting students to learn the underlying math tools through chat GPT and come to his class ready to learn physics. He used some very interesting words, like he sees generative AI for his students to get to physics faster, which was very inspirational. Talking to customers was very interesting, and he is very passionate. I learned a lot from him, and then asked him about my time with them. This is great for you, but how about your colleagues, other professors, and the administration? What are they thinking? It became clear that he was a promising outlier leaning into this ahead of others. It helps you to also get a sense for there are some people who will be first to adopt a technology, but then they’re also the ones who can tell you a lot about what others care about and what it’s going to take for others to come along for the ride.
AIM: What key insights or lessons have you gained from your journey in AI product development and stakeholder collaboration? How have these insights shaped your approach in this dynamic field?
Paul Pilotte: Stay close to customers. Engage, Listen, learn, enable. In the AI space, I see two types of predictive AI using AI as a model and an engineered system, which is a systematic way to design requirements. Customers want to put those AI components into the system as embedded software. That’s another important thing. And then there’s Generative AI. Another key theme is about leadership; one is about people and collaboration.
Regarding my role as product management lead person, there are many other teams and people with whom I work. Part of that is inspiring, leading, recognizing, and encouraging because it takes a community and a team of folks. My other suggestion is to grab opportunities to get in front of customers as often as possible. And think about your customers; as I mentioned earlier, pretend you’re part of their teams and develop hypotheses. I often think about what I would do if I was one of our customers, and I develop a hypothesis that makes sense to me and then bounce it off of other people to refine my thinking about it. Part of that as a leader, is to develop stories and narratives about how customers will get value from AI. There’s a huge opportunity to collaborate with academia. It’s interesting. I gave a talk about a year ago.
One of our customers and the CTO lead the data science team, and I made a pitch to say they should reach out to their local universities for several reasons because many researchers want to work with industry. Students want to learn, and it’s a way to tap into students’ energy and find the best students to hire. A lot can be done for industry professionals to collaborate with academia.