In The Arena by TechArena - Fantasy Football Supremacy with WalterPicks' Sam Factor

Episode Date: December 1, 2022

TechArena host Allyson Klein chats with WalterPicks' co-founder Sam Factor about how AI helps deliver 17% superior recommendations to fantasy football lineups vs. the major recommendation sites.

Welcome to the Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome to the Tech Arena. My name is Alison Klein, and today I'm delighted to be joined by Sam Factor, co-founder of Walter Picks. Sam, welcome to the program. Thank you. Thank you so much for having me. So Sam, I invited you on today because I think you and your co-founder, Dylan, have a really fascinating story with what you're doing with Walter Picks. I bet a lot of folks in the audience have not heard about your company. So why don't we just start with an introduction of Walter Picks and what you're delivering to the world? Yeah. So the shortest way to put it is we build tools to help people win at fantasy sports. So really our main offering has to do with season-long fantasy leagues. Those are like the traditional fantasy football leagues you play with your friends. But the tools that we have built are powered by machine learning, and they can help you in your season-long fantasy league. They can help you in daily fantasy contests. That would be on platforms like DraftKings and FanDuel. Or if you are more into the sports betting side of things, we also have tools for player props as well. So really our entire platform is powered by player projections,

and that is what powers the tools that helps us make decisions in fantasy, in daily fantasy, and with player props. Now, this is something that came out of some work that you did actually when you were in college. And it's a little bit different than a lot of the fantasy products on the marketplace because it's rooted in artificial intelligence. So tell me a little bit about how you and Dylan created it. And what was the impetus for looking at artificial intelligence as a tool. Yeah, absolutely. So first off, I've loved fantasy sports for about as long as I can remember at this point. Actually, my longest running fantasy league has been going on for about 15 years. I'm 25. So basically, since I was 10 years old, I've been in love with football, specifically fantasy football, but really all sports. I also played sports

in college. So I've always been very interested in sports and fantasy sports. And I was a math major in college. I have a passion for math. I actually taught high school math for two years full-time after graduating when we were just starting Walter Picks. I was actually part-time CEO, full-time high school math teacher when we first started this. But really, math and sports were always two of my favorite things. And so I was fortunate enough to be at a school that allowed me to really combine my two passions. So Dylan, when I met Dylan, he was a computer science major at Ithaca College. He graduated the same year as me. And we took a machine learning class together our senior year. I was actually the only non-computer science major in that class. So we actually made a really good

team. I understood a lot of the math side of machine learning, which was covered in the class. Dylan was a really, really great coder. So he understood a lot of the coding, which I was very much getting up to speed on back then. So we made a good team in that sense. But Dylan had actually never really watched much football. He had never played fantasy football. So he was still sort of learning the ins and downs of that. Now he's a very passionate fantasy football fan. He actually won one of my fantasy leagues last year. But that's sort of where my two passions collided is that machine learning class where I was allowed to build a machine learning model with Dylan as our final project that kind of went on throughout

the whole course. At that time, it was actually just a model to predict running back fantasy football points. So very, very narrowed down, obviously, now that we have our app and our entire product that powers all player projections for all positions and all teams, and actually multiple sports now as well. But we started very small. And once we figured out how to do it well for that specific category, we sort of expanded out from there with a similar process. Let's go into the covers a little bit. What are the parameters that you identified for that particular challenge, running back points? And what kind of data size did you use to train your model?

Yeah. So the data size, really the thing I like to really focus on the most, or at least start out with when talking about machine learning, because I think you can get lost in a lot of the headlines, is the data is by far the most important thing. So we actually, we started with just a basic linear regression model, which you can actually sort of build without machine learning. Now it's a much more complex sort of mix between a random forest model and a linear regression model. But the data is always going to be the most important thing. If there is bad data in any model, no matter if it's the most complex machine learning model you can find, it's

not going to output useful things. So that was the bulk of the project, was figuring out what is the best data to try to run this model with. That was the bulk of the project, was figuring out what is the best data to try to run this model with. That was the bulk of the work I did in the class. And ultimately, in sports, and definitely in football, it's very high variance, right? There's a lot of random things that can happen. If you're watching a football game, even just last week, there was one catch that was ruled a catch and a touchdown that was worth eight fantasy points. It was reviewed and overturned, called not a touchdown. Very controversial.

Called the change of the outcome of the game, too. So maybe change the outcome of whether you won or lost your sports bet, whether you won or lost your fantasy game, and very arbitrary type of decision. There's a lot of those types of moments across sports games. So it's a very high variance thing to try to project, which is part of the reason I found it so exciting. Because when we were in a machine learning class, a lot of what you do, you project things really, really accurately. Like you can teach computers how to read, you know, if the iPhone, you can like highlight text

from photos. Now that's machine learning. But in sports, you know, maybe you have 25%, 30%, 35% accuracy, and that's really good. So one of the interesting things for us was we were trying to compare ourselves to other projections out there as a good benchmark rather than just the usual R squared metric you might use in a stats class. But really the stats that matter the most for football is how much you're touching the ball. And that sort of intuitively makes sense. The guys that touch the ball more have a better chance to score more points. And the guys that touch the ball the least have a chance to, they never really have a chance to score points.

And so we're ultimately trying to predict volume first, just the sheer opportunity someone has, and then efficiency after that. But the opportunity is the most important thing. And the interesting piece about that is most players don't touch the ball, right? There's hundreds of players in the NFL. So if you just dumped all of the game logs from every game, which is the very first thing we did, we took basically all of the, every single stat for every single player that offense, defense punters i mean there's players that are on nfl teams that never actually step on an nfl field there's actually a lot of those players on the practice squad and things so if you took the whole data set i think that started with going back to around 2000 and this was back in 2018 or 2019 so about about 19 years worth of game logs for every single game, every single player is

hundreds of millions of data points. But maybe 90 million of those were worthless, right? They're all sort of just zeros because most players, again, don't touch the ball. Most players also aren't relevant for fantasy football specifically. There's even some players that maybe they touch the ball a couple of times a game, but they aren't in anyone's fantasy lineups. And so we're trying to build relevant tools, relevant projections for really like the top 100, 150 players. So narrowing down and building rule sets of basically filtering down the players that matter, but also building projections for today's players relevant to players that look similar to them in the past is really where we found is the best way to build out the projections. But yeah, most of the time that went into making our projections

quality was actually just focusing on the data and less so on the actual dynamics of the machine learning models. No, I know that you did this for a class and then football season came around and you actually tested the model and it performed really well. Tell me what that was like and how you knew that you might have something that was marketable. Yeah. So yeah, that's exactly right. So we built this in the spring, right before we graduated in the spring of 2019. We both graduated, went our separate ways. I mentioned earlier, I became a high school math teacher. Like I already mentioned too, I was very into fantasy football and I had spent all this time building this model.

I wanted to use it for my own leagues and I wanted to see how it was doing in comparison to some of the other big projections out there, ESPN, Yahoo. There's endless projections you could find on the internet, but those are sort of the big ones a lot of people rely on because those are the most popular platforms to play fantasy on. So we were tracking our projections

against theirs in that first season and they did really well in comparison. And so that was really exciting. I think from, you know, we have something here that can be valuable for a lot of people. Also, we built something ourselves that we can build on

and make even better. I think we were also really confident there was room to improve because we had only spent about six months on this originally, and it was really focused on running backs at the time. And after that first season, we had all sorts of ideas of how to make it even better. So we were really just excited overall that we'd built something that worked, and it was working better than a lot of the big companies out there that had much bigger teams than just us two. And so that really gave us the motivation to turn it into a product and try to help people make the best fantasy decisions they can make. to ask you, as you started with running backs, as you looked at tackling the entire lineup, did the parameters or did the thought process of how you would actually make recommendations change by position or was it pretty consistent?

Yeah, it definitely changed. Each position was a little bit different than the next. Wide receivers, tight ends were sort of the most similar to each other but really each each position was its own problem but as we solved each problem the next one got a little easier because we had the framework from the previous one and it was really just looking at the data a little bit differently each time and it's different data points by passing attempts versus rushing attempts versus targets all sort of different different stats that come from similar places. But overall, each time we solved the problem, the next one got a little bit easier.

The other thing that I think is worth mentioning that's a little bit different than the machine learning stuff, but is still data analytics. I think part of the way we've differentiated ourselves is the way that we display information within our tools. We've tried to show the variance. So we try to show that there's a range of outcomes that's upcoming for each player in each game, which really helps in the decision-making process of should I start or sit this player in my season long league or should I play them in daily fantasy? What's the best case scenario for this player? What's the worst case scenario for this player? And that type of data visualization can be really helpful for decision making.

Now, let's fast forward to 2022. We're entering fantasy playoffs, which is one of the reasons why we're doing this episode today. Everybody's focused on their fantasy teams right now. And fantasy is growing. It's going to be over a $45 billion industry in the next five years. So this is not just a niche thing. This is a huge business. Tell me about the response that you've received in the marketplace. You mentioned that you started with football. Are you doing other sports now? And which sports are you doing? And what's the outlook for Walter Picks as you move forward? Yeah, so it's been really exciting and really interesting for me, you know, going from sort of just this person who really loves playing fantasy

like a lot of other people to sort of getting a very close view of what the industry looks like. I've gotten to meet a lot of people I never thought I would have gotten to meet. And I really, in building, you know, this business, I've seen just how big the fantasy community is. And it really is a very massive industry, but it's also filled with people that are very, very intensely into fantasy. So it's not just, I forget what the number is, around 60 million or so fantasy players. A big chunk of those people, they're not just setting their lineups once a week and that's it. They're looking at this every single day. They want to consume content around this. And there's a lot of data out there showing that fantasy sports players

are the most active sports watchers too. So they're spending a lot of time consuming and watching sports. It's really more of a hobby and a passion than just like a pastime for a lot of people. And I experienced that myself growing up as well. So I think that plays into why the industry is so big. And it also has these really crazy network effects. Because if you're in a fantasy league, there's probably eight to 11 other people in the fantasy league with you. So it can also grow, The leagues can grow really, really quickly like that. We've expanded to the NBA and March Madness so far. We are really focused still mostly on NFL.

I think that'll always be our biggest audience. I think the NFL and fantasy football will always be the biggest fantasy sport. Part of the reason I think that is just the weekly cadence of the NFL works very well for people where the NBA is every single night. Most sports are basically every night. So you have to set your lineups basically every night in other fantasy sports. But we are definitely looking

forward to expanding to more sports. Sort of what I was saying a little bit earlier, as we solve the problem for one sport, the next one gets a little bit easier, even though each one is definitely a little bit different. So I think we will continue to expand to more sports, but we will never lose sight of improving what we've already built for the NFL and NBA as well. And sort of where the future is for us, we're going to keep building tools to help people with their fantasy decisions. We have so many different things that are already on a roadmap to build. But really, even outside of our current product, long term, we're just going to keep building products that help our audience. We have a really fast growing audience on the content side too. We haven't touched much on this, but I actually was a

writer in the fantasy industry before my junior year of college. Actually, I began writing for a couple of different sites. I eventually got paid for my writings though. I put paid in air quotes usually because it was paid less than, it was a per article rate. So I think it was actually less than like $3 an hour that I was getting paid. But I loved fantasy sports. I liked writing about it. And I still create content about sports. And that's been a big success of our business as well. We have almost 1 million total unique followers across our different social media accounts today. And we started from zero followers across all of our social media accounts about two and a half years ago. So that's been a big success for us as well and definitely will continue to be a focal point for us. One of the biggest things in this industry that people are struggling with, especially when you get out to sports betting, is retention, retaining people on your platforms and on your products. The sports betting industry specifically, they're spending

way too much money to acquire users. Most of them are losing money every single month in hopes of being profitable sometime in the distant future. They just want people on their platforms. But part of that working is retaining the people that you're acquiring. And a big piece of that is actually content and having an actual relationship with your audience. And that's something that we've done really well. And so because we have this big audience, we'll continue to build new platforms, new products for them as well. That's fantastic. Sam, one final question for you.

Where can I send people that are listening to get on Walter Picks and start utilizing your tools for their fantasy leagues? Yeah, so we're a mobile app. We're on the iOS and Android app store. We've been very focused on mobile. I feel that could be its own podcast in itself from a tech standpoint of sort of where the sports industry, fantasy industry was all online, all on websites.

It originally started on paper and pen, and now it's almost all on your phone, especially for the younger generation. So yeah, you can find us on the iOS and Android app store, just WalterPix, completely free to download our app. You can use most of the tools for 100% free. There are some premium subscriptions in the app that is how we monetize, but those are totally optional and they do come with a seven-day free trial as well. Fantastic. Thank you so much for being with us today, Sam. Thank you, Allison. Thanks for joining The Tech Arena. Subscribe and engage at our website, thetecharena.net.

All content is copyright by The Tech Arena.