Today on How AI Happens, we are joined by Bob Bress, the Vice President and Head of Data Science at FreeWheel, where he leads a team of data science experts in developing and implementing complex analytical solutions for leading-edge technology and advertising programs
Bob highlights the importance of building interdepartmental relationships and growing a talented team of problem solvers, as well as the key role of continuous education. He also offers some insight into the technical and not-so-technical skills of a “data science champion,” tips for building adaptable data infrastructures, and the best career advice he has ever received, plus so much more. For an insider’s look at the data science operation at FreeWheel and valuable advice from an analytics leader with more than two decades of experience, be sure to tune in today!
Key Points From This Episode:
Tweetables:
“As a data science team, it’s not enough to be able to solve quantitative problems. You have to establish connections to the company in a way that uncovers those problems to begin with.” — @Bob_Bress [0:06:42]
“The more we can do to educate folks – on the type of work that the [data science] team does, the better the position we are in to tackle more interesting problems and innovate around new ideas and concepts.” — @Bob_Bress [0:09:49]
“There are so many interactions and dependencies across any project of sufficient complexity that it’s only through [collaboration] across teams that you’re going to be able to hone in on the right answer.” — @Bob_Bress [0:17:34]
“There is always more you can do to enhance the work you’re doing, other questions you can ask, other ways you can go beyond just checking a box.” — @Bob_Bress [0:23:31]
Links Mentioned in Today’s Episode:
Bob Bress 0:00
In some ways, right data scientists come in, and they want to be sort of that center of excellence, right? The expert that folks go to. And really what we want is the opposite, right? Really what we want is everybody to be educated within the space, because that's what's going to uncover more opportunity for data scientists and people who are trained in the space to do more interesting things.
Rob Stevenson 0:24
Welcome to how AI happens, a podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists, and machine learning engineers, as they get technical about the most exciting developments in their field and the challenges they're facing along the way. I'm your host, Rob Stevenson. And we're about to learn how AI happens. Here with me today on how AI happens is the vice president of data science over at freewheel part of the Comcast family Bob Bress. Bob, welcome to the show. How are you today?
Bob Bress 1:01
Great, great. Thanks for having me. Excited to be here.
Rob Stevenson 1:03
I am thrilled to have you because lots to go into for both of us. Maybe we can we start a little bit about what freewheel does just set some context, we'll give like the elevator pitch and then we'll get into some of the nitty gritty of the data science over there.
Bob Bress 1:16
Sure. Well, freewheel is a global technology platform that powers the TV advertising image industry. And if you think about TV advertising today, it's really anywhere that video advertising is happening. So that could be streaming platforms, traditional television, or any any digital platform where you're seeing video based ads. So freewheel really is the ad tech behind how those ads are getting from the advertisers to the suppliers who are hosting that that advertising. And we've seen over the past couple of years, as I mentioned, there's been a big shift to streaming platforms, right, as more and more viewers are watching less traditional television, and moving more to the various streaming platforms. You're also seeing advertising shift to many of those platforms. And so freewheel powers, a lot of the tech that's behind that.
Rob Stevenson 2:06
Got it. So that's freewheel. What about the Bob Bress? Fidel, let's get to know you a little bit. Would you mind sharing about your background? And how you wound up in your current role?
Bob Bress 2:13
Yeah, yeah. Well, you know, I've been in the analytic space my entire career, which is 20 plus years now. And I've been in a number of different industries over that time, the common link, there is really, analytics and data science, data science over that period wasn't always the hot term that it is today. Right? Right. When I started, it was more Statistics and Operations Research. And suddenly, as data began to drive, really everything in every industry and analytical techniques became more sophisticated. And computing power became more enhanced, right? Data science really took off, right. And so about 12 years ago, I took a role in a company called visible world, which was sort of early in, in addressable advertising for television, which is really think about targeted advertising, like you see online, but on television, and I could tell that that was going to be a big part of the future of how we see advertising on TV. And so was excited to join that space, then that company was acquired by Comcast a number of years ago, and now rolled into what's known as freewheel, which is really goes beyond just traditional TV advertising. But like I said, the whole digital ecosystem. And since I've been there, I've been leading our data science group, again, wasn't always known as a data science group. But today, more commonly is. And it really incorporates not just statistical analysis, operations research, but machine learning AI, and really any application of advanced analytics that we look to build into our products.
Rob Stevenson 3:49
Gotcha. So because of your role, it strikes me that you probably sit between a couple of different departments, there's much interplay, that is important between the data science team, between the teams developing various ML and AI technologies, deploying AI and ML approaches, perhaps even the product team. So how do you kind of characterize that relationship? When you think about representing the data science team to the business writ large? What is your approach?
Bob Bress 4:12
Yeah, it's interesting, and each company takes it a little bit differently. freewheel is a very technology and product focused company, right? So we're building advertising technology, platforms that that power the ecosystem. And as you can imagine, those systems are very data centric, right? Data is what's powering an automated decisioning is what's powering those platforms, right? So data sciences has a pretty big role in how we build those things out. And so at least that freewheel data science is embedded within our product group because it's so important to incorporate advanced analytics, latest in ML and AI into our product suite. Right. It's important for the technology but also just to be competitive right within the space because I think every major player in the space is investing heavily In this area today. And so for us, our main focus is really how do we incorporate these concepts within our products going forward? But it even goes beyond that, right? It's also how do we work with our revenue and client facing teams to solve client problems using data in innovative ways? And how do we connect with other parts of the organization to make sure we're using the right data, uncovering sort of new opportunities in the space that we might not otherwise be exposed to? So it really does cover sort of the breadth of of the company, while still having that sort of very product focused primary goal.
Rob Stevenson 5:38
Gotcha. I'm also curious what indicators you look for, to ensure that data science is I want to say taken seriously that makes it seem to to like fringe. And as you say, data science is like this sexy output. Basically, like every company needs data science team, data is everywhere, we're creating more data than ever before. I don't want to say taken seriously. But how do you ensure that your company is a data driven company? How do you know that your team is not just a fashionable arm of the business? And it is actually contributing?
Bob Bress 6:06
Yeah, well, it's a great question, because I think one of the challenges, right data science people faces, particularly for folks in the organization that are not trained in the area, right, it seems like a little bit of a black box, we know those folks are doing something sort of fancy and quantitative, we don't totally get what it is or how it applies, but seems kind of interesting. That's really not what we want, right? What we really want is the entire organization to have a level of education around what data science does, so that everybody plays a part in identifying areas where data science can help solve some of the business problems, right? If we're always the ones having to uncover the problems, you get a lot less sort of demand for the skill sets that the team has, then you would have people all over the company can uncover them themselves. Right? And so your question is important, because that means that as a data science team, it's not enough to just be able to solve quantitative problems, right, you have to establish connections to the company in a way that uncovers those problems to begin with, right. And so I would say there's three, three things I would say I see important for that. One is, and we spend a lot of time doing this, establishing relationships across the company, right? That means with teams, not just product teams, but revenue teams, support teams, any teams that are dealing with sort of data and problems, where we could use advanced analytics to either automate or support or provide better decision making capability, we want to have connections to those teams, which might mean regular check ins might mean more or less formal conversations where we talk or even just presentations, like here's what we're working on. Now, we'd love to see if there's any commonality to what you guys are seeing. So one is establishing those relationships, I'd say another one is, and this is really to your point is becoming an establishing data science champions, right? You need people in the organization who are going to highlight the great work of the team, right? Who are going to take opportunities to promote wins that the teams has, and present the business results that they've driven. And when those are made more public, at least within the company, it drives more people to think, hey, well, I have a problem. It's very similar to something they've solved. Or, hey, if they could do that, maybe here's something else we can have a discussion with. So I think having champions sort of inside and outside the team who can promote that are important. And then the other one, I'd say is having a talented team and in and of itself, right. So I think when people sort of recognize that some of the folks on the team are uniquely positioned to solve these data centric problems, that I think they're more prone to rely on those folks to come to those folks as they see some of the challenges. But yeah, that's one of the biggest things we work on is really establishing sort of that credibility around the rest of the company.
Rob Stevenson 8:58
It's so important to do that almost internal PR, because particularly within data science, also within our listeners will be more familiar with it like AI and ML aside, probably, but you're doing something that most of the company probably doesn't intimately understand. And you do because you're a technical person, and it's your job, and you understand why it's important. But there's this weird part of your role where you also need to be explaining and bringing people into the conversation is not just enough to do something awesome and shipped off to GitHub, like you also have to make sure that the rest of the company is able to receive it. And so like for you like you have had this background as an individual contributor, and now in your leadership position. Are you finding yourself having to do more of that piece of it?
Bob Bress 9:40
Yeah, for sure. It's interesting. It's almost opposite what some people might think, right? And in some ways, right? Data scientists come in and they want to be sort of that center of excellence, right? The expert that folks go to, and really what we want is the opposite, right? Really what we want is everybody to be educated within the space, because that's what's going to income uncover more opportunity for data scientists and people who are trained in the space to do more interesting things, right. And so the more we can do to educate folks within and even clients right outside the company on the type of work that the team does, the better position we are in to tackle more interesting problems, right and innovate around new ideas and concepts.
Rob Stevenson 10:23
Yeah, that makes sense. It's tricky, because it's like, it's maybe not the skill set that got you into the position. You maybe got into this role, because you were really, really technically sound and had this capability. And now there's like this marketing side that you're having to learn.
Bob Bress 10:39
Yeah, yeah. And it's interesting, because if you think about folks that typically join, like a data science team, they join it, because they're interested in the nature of the work, right, the quantitative work, a lot of individual contributors who just really want to work on these interesting problems, and haven't really maybe thought about that part of it as much. But the reality is, in order to maximize sort of that work, right, you have to make an effort on the other side, too, which is sort of helping others have an understanding of what it means to do data science work, what the capabilities are, what type of business problem that uncovers,
Rob Stevenson 11:15
yeah, it's not enough for you to understand why what you're doing is useful to the business, the business misunderstand.
Bob Bress 11:21
Yeah, yeah. And then the other side of it, too, is it's rare, right? Somebody comes into the job, and you just hand them, oh, here's the problem. Here's the data, build a model and give me the answer. Have at it, right. It's more like, here's a very vague business problem. Is there anything we can do to sort of explore analytically how we would go about solving this right? And then even beyond that, how can we do that at scale? Right? How can we do that leveraging other teams around the company to sort of scale that solution up? So there's a lot more that goes into just build a model analyze data?
Rob Stevenson 11:56
Yeah, of course. Bob, I want to ask you to opine a little bit on the role of data science, maybe outside of freewill even. What do you think is the interplay between a mature like rigorous data science approach, and its connection to developing AI technology?
Bob Bress 12:12
It's interesting, I think one of the things when we build a data science team, number one we're looking for people from we take people from a variety of different backgrounds, right? Certainly, there's sort of a quantitative technical capability that folks have to have. But in terms of skill sets, I think one of the things we look for is they're really a problem solver, and somebody who can learn quickly. And I think like what we've seen more recently, like with AI advances, generative AI, and then the like, is the field is changing quickly and advancing quickly. And that is going to continue to accelerate, right. And so as we bring in folks with different sets of data science skills, I think the most important thing is that they can adapt, right as this field continues to accelerate and still be a leader in the space, right? So we could hire a generative AI expert tomorrow, but in three years, it's going to be something else, right? Or it's going to be even more advanced, certainly than than what it is today. And so I think the big thing, in terms of connecting with the latest and AI advancements and ml advancements is building a team of problem solvers, and lifelong learners, who can quickly pick up some of this new technology and connect it to the business, right, who's staying up with the industry and looking at how it applies to our business itself. So certainly, there is a bridge between the basic skill sets that they're bringing in here. But the advancements that are happening now is not going to be what they learned in school. So I think we spend a lot of time on continuous education there and, and do our best to keep up with it.
Rob Stevenson 13:52
It's such good career advice. And I think it's applicable anywhere, but especially in this sector, where things move so fast that the TensorFlow verse pytorch debate is one I've had on this podcast. And it's like, wherever you come down on that, I'll use it. This one provides better visualization. Fine. What about in you know, a couple of years when there's a third one that everyone agrees is better than TensorFlow or Pytorch, as they weren't gonna care how good you were using Pytorch? If that type of technology does your is different at that point. And then to your point, it's like, the person who was adaptable the person who was like, oh, yeah, TensorFlow was good for this Petros are good for this. They had their place, but I'm not an I'm not a pytorch expert. I'm not like branding myself. That way. I'm not indexing on that one skill. That is the person who is going to be more more dangerous, right?
Bob Bress 14:34
Yeah, for sure. I feel like this is one of the things that's probably changed over the past 10 plus years is Whereas you might on job openings, right, you would see must have these pieces of software technology. Whereas now I think there's a lot more openness to it because the reality is, first of all, some of them are so new, you're not going to find somebody with five years experience and the other is like you said it may not mean that be the tech does your next year. So I think you want to be able to take advantage, though of the latest that is available. You know, you're seeing things now like automated coding support, right? and things of that nature. We always look at stuff like that, right? How can we take advantage of what's out there to make our jobs more efficient and help us do our jobs better and faster and quicker? And I think we will continue to do that and look for people who who do the same?
Rob Stevenson 15:25
Yeah, it makes sense. Of course, there is some baseline of technical ability and savvy someone has to have. And that's like a decision you have to make. What do you look for the technical side? Obviously, the curiosity, the ability to problem solve, but like, I like to think I'm a curious person, Bob, I have, I can solve problems. But I don't think I'm a good fit for your data science team. What? What is it that you look for what were like the actual technical things that are important to you?
Bob Bress 15:47
Yeah. Well, around the curiosity, right, someone who's curious has a lot of questions, but they're also asking the right questions. So I mentioned before, it's rare that you're sort of just handed a nicely packaged problem and you go solve it and submit the answer back. Usually, to get to what it is you have to solve, you have to ask a number of the right questions, right. And a lot of times in the data science space, it's around the nature of the data, right? Where does the data come from? What does it represent? How was it process? Does it need to, you know, was there other processing that needs to be done? What do I get out of doing exploratory data analysis that maybe uncovers anomalies in the data that I need to be aware of that, that maybe weren't highlighted to me? So it's somebody who thinks around all the sort of adjacencies of the problem versus just, hey, let me go build a model, right? It's very rare that just building a model is going to work with unless you've done sort of the pre work to truly understand the full context of everything you're working with, right? And that even includes, like client expectations, and does this solve the client problem? And and this is where a lot of that communication comes into, right? So I think asking the right questions, and a lot of that will just come from that curiosity and experience taking a project from beginning to end, right from the point where you have data that is not ready to the point where you can produce a result that you're confident has a business impact. So somebody who can show they they've taken a project through that course, that they they know what questions to ask, and sort of what pitfalls to be aware of.
Rob Stevenson 17:24
It's interesting that the idea of like the siloed, code monkey, who was will everywhere in tech for a long time, it's even still is fantastically well compensated. It seems like that doesn't exist in this side of the business, like the problems are too nuanced. And there's too much Interplay for someone to just be like, I don't know, I just opened my computer and I write the code. And I close my computer like,
Bob Bress 17:48
yeah, yeah, very rarely works. Well, that way. Unfortunately, I think there's so many sort of interactions and dependencies across any project of sufficient complexity that it's only through that collaboration across teams that you're going to be able to sort of hone in on on the right answer, right, or even take the right approach. So there's just a lot that goes into it. And the reality is, there's also a lot of people that you can lean on to make your job more efficient or willing to help you. If you take the initiative to ask right. And so the people who take that initiative, I think you're gonna have an easier time than those who are just looking to sort of do it all themselves.
Rob Stevenson 18:28
Who are some of those like allies and stakeholders in the business for you?
Bob Bress 18:32
Well, the key Frost has been as we work with different teams across the company, like, if we look within product management, as you established product managers, where you develop sort of an exciting use case, right, that's really proved itself out a lot of times, then that relationship continues, that product manager now now has a greater awareness of what can be accomplished and will sort of come back to engage the team on other projects, right. And the same would be sort of on the revenue side. So for some of the teams that are engaged directly with clients, right, as they face different client problems involved with sort of advanced data applications, they start to get an understanding of what can we come to the team with, and this is where that sort of championing comes into place. Because once you have a successful use case, and then those folks can take that to their colleagues, right, you can promote it to their colleagues, and then you have more people who are coming to you, as opposed to you trying to say, hey, we have a couple ideas. Obviously, we're not in the day to day with your client, which may or may not pan out, right. So you really want projects coming from both ways, right? You want the business to bring them forward. You want to educate them in a way that they can do that successfully. And you want to be close enough to the business that you can identify areas where data science can support. So I think we try to work from both angles.
Rob Stevenson 19:53
Yeah, that makes sense. I want to get a little bit into the the nitty gritty of your operation over there, Bob, and I'm going to ask Good question, I think is a big one. And it's going to be hamstrung by its brevity. But let's start with this. What is an adaptable data infrastructure?
Bob Bress 20:08
Yeah, to me and adaptable data infrastructure is a few things. One, it allows you to run different data driven applications off of that architecture, right without needing sort of a separate architecture with basically the same data. Number two, I think it provides a way to adapt as sort of business and client needs change. So what I mean by that is, if we want to solve a slightly different problem with the same data, we have to re architect that solution. Right. And I think one of the things I'd say is certainly cloud computing, technology has enabled us to be more adaptable in a lot of ways, right? I think the capabilities that we find on cloud computing platforms allows us to move a lot more quickly than if everything were saying on premises, right? Or if we had to redesign how we store or leverage data, I think that becomes much easier. But when I think of adaptability, you know, those are some of the things I think about.
Rob Stevenson 21:10
Gotcha. So how do you go about building that adaptable approach?
Rob Stevenson 21:13
Yeah, number one, I think, like I said, leveraging cloud computing technologies and platforms, you know, at free will, we'll use platforms like AWS, snowflake, data, bricks, anything that can support how we build and enable data pipelines in a way that supports a lot of the analytical use cases that we have. And I think that requires an investment not only in the tech, but in certainly making sure folks are trained in that space. The other thing I would say is, if we look from a data science perspective, right, if you, what we also don't want is hiring data scientists with expertise in modeling and analytics to spend most of their time building data structures and pipelines and 5% of the time on sort of the modeling aspects. And so some of these platforms certainly create a lot more efficiency there. We also invest a lot in data engineers, right? So data engineers who have expertise and using these platforms, that just really have great skills in how they build them in a way that allow for a variety of advanced data applications on the back end. So those are some of the big things we do.
Rob Stevenson 22:24
Gotcha. That's helpful. Thanks for sharing that while we are creeping up on optimal podcast length here. But before I let you go, I want to ask you to share with the folks out there in podcast land, what is the best career advice you ever received?
Bob Bress 22:39
Yeah, well, I would say my earlier days, sort of in an attack, we would do everything we could to sort of innovate new at tech products and capabilities using data. And we were often faced with tight deadlines, we're looking to present sort of new concepts and capabilities to clients. And I remember one of the leaders in the company at the time, I remember one time, we were basically over our deadline. And I at least felt we did everything we possibly could to do what we can. And we just had to go with what we had. And despite the fact that we were sort of over the deadline, it was after hours, the leader would still ask, Well, what more could we do? Or couldn't we do X, Y, and Z else? And left me thinking? Well, yeah, I mean, I guess we could we don't have a lot of time. But yeah, there's more we could do. And we would keep working at it. And sure enough, we would uncover some other sort of enhancements to what we had built. And sure enough, what we ended up presenting the client often was way better than where I probably would have landed, had we not asked that question on my own. And so I think for me, the takeaway is to look at, well, what more could we do, there's always more you can do to sort of enhance the work, you're doing other questions, you can ask other ways you can go beyond just sort of checking a box, right? A client may have asked for x, but how can we give them x plus, right? How can we give them How can we not just satisfy their conditions, but delight them in a way where they're surprised and compelled to work with us? Right. And so I think one of the takeaways I had there was just always ask that question, what can we do to go beyond? What can we do to provide an enhanced experience and product to those who are working with beyond what they've even asked for? Because we're the ones in the midst of it, right? We're the ones with the data on sort of the know how, and we're in a position to do more. So that's one of the things I would say. I take away.
Rob Stevenson 24:40
That is great advice is easy to be good enough but it is better to ship the best thing you can I think we should all be striving for that about this has been fantastic chatting with you today. Thank you for walking me through the data science operation over there and your background. I've loved learning from you today.
Bob Bress 24:54
Thanks for having me. It was good to be here talking.
Rob Stevenson 24:58
How AI happens is brought To buy sama. Sama provides accurate data for ambitious AI specializing in image video and sensor data annotation and validation for machine learning algorithms in industries such as transportation, retail, ecommerce, media, med tech, robotics and agriculture. More information, head to sama.com