Data Brew by Databricks

Data Brew Season 4 Episode 2: NBA Analytics

March 09, 2022 Databricks Season 4 Episode 2
Data Brew by Databricks
Data Brew Season 4 Episode 2: NBA Analytics
Show Notes Transcript

For our fourth season, we focus on connected health and how data & AI augment and improve our daily health. While we’re at it, we’ll be enjoying our morning brew.

Alexander Powell chronicles the evolution of sports analytics and how professional sports teams use data as a competitive advantage. 

See more at

Denny Lee (00:06):
Welcome to Data Brew by Databricks with Denny and Brooke. This series allows us to explore various topics in the data and AI community, whether we’re talking about data engineering or data science. We will interview subject matter experts that are deeper into these topics. In this season, we’re going to focus on connected health and how data and AI augment and improve our daily health. And while we’re at it, we’re going to enjoy our morning brew. My name is Denny Lee, I’m a developer advocate at Databricks, so one half of Data Brew.

Brooke Wenig (00:36):
And hello everyone. My name is Brooke Wenig, machine learning practice lead at Databricks and the other half of Data Brew. And today, I’m thrilled to introduce Alexander Powell, who is head of quantitative analysis and development of the NBA team, the Charlotte Hornets. Welcome Alexander.

Alexander Powell (00:49):
Thanks for having me guys.

Brooke Wenig (00:50):
So to kick it off, I would love to learn more about how did you get into the field of sports analytics?

Alexander Powell (00:55):
Yeah. So I was always into sports, played basketball at Kenyon College, a small Division 3 school with a good academic program. And I found there that I had this passion for math and numbers and stats, but it never occurred to me that I could combine the two. Luckily, I had a professor who showed me the way there and that led me to exploring this quantitative field within sports that’s been really growing over the last decade or two. So I got my start in college just messing around with what I was playing in and then expanding that towards the NBA and other sports that just interested me.

Brooke Wenig (01:36):
And so obviously you’re focused on analytics for basketball, but in terms of the evolution of sport analytics, what fields are most advanced? I think many of us have heard of Moneyball and analytics for baseball, but can you talk a little bit more about the evolution of analytics for professional sports teams in general?

Alexander Powell (01:51):
Yeah. So baseball was the first noteworthy use case, I guess. Essentially the word analytics is more just problem solving through numbers away. And that’s been done obviously since sports started, but from a much larger use it’s in this century, it started with baseball reason being it’s this very discreet sport. So it’s a lot easier to assign value to different events within the game. Where a sport like basketball or soccer, it’s very difficult because there’s so much movement. And so new technologies have allowed the shift to go from baseball to other sports, being able to increase it. And also the cultures of the sports are so different that it becomes easier in some sports than others to communicate those ideas to the people who’ve been doing it long before there were advanced computing and machine learning within sports.

Denny Lee (02:46):
Yeah. Actually, I’d like to dive a little bit deeper into what you just said here, which is like, what were some of, in terms of the cultural differences between, let’s just say baseball and basketball that made it either harder or easier to help explain that concept of analytics to, into your case basketball?

Alexander Powell (03:02):
Yeah, I can’t speak… Never worked in professional baseball. I have some friends who do, but I think that one of the… They just had a head start with the [inaudible 00:03:12]. The simplest answer was because of how the sport was set up. And basically most of the game was already collected from a box score standpoint. If you look at an NBA box score, there’s so much missing, you’re left like, “This isn’t as helpful as it could be.” Where in a baseball box score, balls, strikes, all of the events are recorded within there. As well as I think there’s so much more… The game is such slower paced, those conversations can happen within a game. So you can make those real time in-game adjustments based on data where in a sport like basketball, we’re not going to stop the game just talk to a player while he’s dribbling up the court, because by then the shock clock will go off. So we have to find other ways to communicate.

Alexander Powell (03:54):
Usually it’s pre, post game. We’re slowly starting to get where we can do some things somewhat real time. We’re almost more waiting for the data to speed up. But then even a sport soccer, there’s no way you’re going to be able to communicate to a midfielder on the other side of the pitch because that communication barrier is just so difficult that you have to almost infuse it more into your scouting and your training, than baseball can actually adjust things in the game.

Denny Lee (04:21):
So you’re telling me I can’t do Morse code or waving flags as a way to communicate that for [crosstalk 00:04:26].

Alexander Powell (04:25):
You can try. And it’s interesting to see how everyone communicates, but 24 second shot clock can be quite difficult to communicate exactly what you might want.

Denny Lee (04:37):
And so, and last time I checked, I’m not allowed 50 timeouts in basketball. Got it. That’s fair. Fair enough. That actually leads generally into that idea of well, then because they’re so different and it involves so many non-technical people, How do you share that insight? How do you get that information out to… What type of insight are you even sharing? And then, because you did mention the fact that it’s often pre or post, for example, are you sharing insight like you’re recommending a player needs rest. I’m just curious from that context.

Alexander Powell (05:07):
So my focus is mostly on our front office side, but within our group a few of each of those components. So we do have someone who will be with the coaching staff. And in essence, he’s an assistant coach and most teams have a position this now, which is the growth of the sport in the last five or so years is having someone who has that influence with game day decision making. Sometimes it might be within game and a timeout, but a lot of times that’s pre-game prep. We have a sport scientist who they’re looking at data from a player rest or player load standpoint, as well as how can we get these guys performing better? And then more, my team’s focus is how can we optimize our team, our roster construction, be it from scouting from the draft, the trades, free agency to make our team better from that standpoint. So there’s internal focus with the sports science. You have your day to day focus with your coaching staff group. And then our front office group is looking much more long term projections of players.

Brooke Wenig (06:08):
And so I know that the Charlotte Hornets are a small market team. Are there any challenges that are unique to small market teams that larger market teams wouldn’t face when it comes to analytics?

Alexander Powell (06:17):
That’s a good question. There’s definitely, I think our analysis has a little slightly different focus. Now, obviously never worked in a larger market, but from our perspective costs become much more of a question because we don’t have the same revenue streams that a larger market team may have. So, I particularly try to infuse in all of analysis on players or teams, the financial component. Which I think a lot of times your armchair fan thinks it’s like, “We’ll just go get this player,” but there’s so many different nuances in terms of how you can get that player. One player may be better than another player, but the cost is different that you’re almost looking at a value per cost of each player. Not that larger market teams don’t do that. They certainly do.

Alexander Powell (07:01):
But I think for us, it becomes even larger importance. And I think also we have less tools because of that financial component of the small versus large market, we have less tools to improve our teams. So we have to find different nuggets, which I think is where analytics can really help organ small market organizations to… It’s like the As in Moneyball where that was their advantage is, we may not be able to pay players more, but we could maybe invest more in this area that will increase our team’s value.

Brooke Wenig (07:31):
Exactly. And I think a lot of those concepts also translate back over into the tech field, of people aren’t always about getting the highest salary. It’s where do they have the most to for growth or the way my previous boss had always framed it is, “Everybody has a different form of currency.” And so for some folks would be getting more play time, other folks could be living close to home. So, that’s a great point you make, it’s not always about just paying the most amount of money to get a player. And so given that the Charlotte Hornets are part of the NBA, what exactly is their relationship like across all of the analytics within the MBA. Do you leverage any common data sources or data providers? What does that collaboration look like?

Alexander Powell (08:06):
Yeah, unlike a few sports, we don’t collaborate heavily. We all know each other. For the most part, we see each other at different events, but it’s less collaborative in terms of data sharing. Now, there are a few things like second spectrum, which is every arena there’s 12 cameras that sit in the rafters, they take 25 pictures a second of all the players and can generate these coordinates of the players on the ball, and we’re able to use those. And so the league actually forms a deal with them. And so we all have access to that same data, basically we’ve all said, we agreed with the cameras in our arena. We can all get data from all 1,230 NBA games. So that’s for the most part, the extent of the sharing, other than things the combine where the whole league puts together. But other than that, it’s an arms race who can use the data better.

Denny Lee (09:00):
Got it. Well then I can that naturally segues into… Well, I think I already know the answer and then the answer’s no. But by the same token, I did want to ask the question anyways. So is there an implication that you’re trying to stream the data and you’re trying to, as the pictures coming in, you’re trying to process make decisions there? Or is it less of that type of scenario, more of a batch analysis. The high level question I’m trying to ask is like, is there an advantage of streaming data within the context of basketball? I’m just curious.

Alexander Powell (09:26):
Certainly. Not as much as other sports, just because of those communication difficulties we talked about. But we’ve really pushed over the last part two years to get our data in real time for games. That way, if our coaches do decide that there’s something useful and we have found that there’s value particularly, you only have 15 minutes to talk at halftime to your players, but if you can find a few nuggets that are communicable to your coaches or your team. And even having that immediately post game, the problem is the data latency from being to get data from cameras, ingest it to second spectrum, then ingest it to us, clean it up and get it in a usable fashion is… The latency’s slowly getting to the speed where we could maybe use it in a timeout or between quarters.

Alexander Powell (10:13):
But yes, we’re slowly starting to real time. We’ve been doing real time, all play by play box score stuff for years, but now getting this really granular level data. So that maybe there’s one or two insights that can be communicated in these really short conversations that are happening between coaches it’s it breaks. But, I think that’s somewhat rare in the NBA. I think most teams are pretty comfortable just like, Hey, we’ll look at it the next morning, or we’ll look at it later tonight after the game. But I think slowly as the latency improves with the data, then it’s becoming more advantageous to have on the bench.

Denny Lee (10:52):
Well, that’s really cool. So, that good to know that actually even streaming is working in these scenarios. Well, then I guess that natural segues to my other question, which is, what is right now, at least common for all these teams that… Basically what they want to know but they’re not actually really currently getting from their data, whether it’s due to latency or due to analytics for that matter?

Alexander Powell (11:11):
I think the two big things that we’ve all been asking for. And when I say asking it’s we know it’s probably not going to be done in-house, at least not at scale. That there’s data providers or league providers that we’re hoping that can solve these problems, is getting that same granular level data we have in the NBA in other leagues around the world, be it college or international leagues, so that when we’re making these predictions, we can really have scale of, “Hey, this is what a guy did when he was 18. And here’s what he is doing when he is 25. Can we extrapolate those?” Right now the quality of basically prospect data is limited. I think we’ve done pretty well with it to date, but there’s so many variables when you’re evaluating a 16, 17, 18 year old that having that more granular data that’s consistent across every league in the world would be helpful. As well as I think the next step, particularly this focus a little on the sports science and the player development realm, those body pose information.

Alexander Powell (12:15):

So rather than having the XY coordinate for every player, could we have the locations of their arms or even a 3D map of their body to be able to know, “Hey, not only are you doing this with the ball, you’re moving in this direction, your hands are this way.” I think one of the smallest use cases is coaches like to say, “Well, you were close to the shooter, but you didn’t have your hand up. You didn’t close out.” From the data, I could say, “Oh, well, he contested it well. And since that he was X number of feet away from the shot.” But if we could say, “Well, yeah, he had his hand up and he blocked his vision this much,” or, or something to that effect we can measure player development a little bit better as well as we could start understanding the biomechanics of players a little bit more in terms of how they move pre or post injury and help the performance aspect of the game.

Brooke Wenig (13:07):
And how far out do you think that is?

Alexander Powell (13:09):
There’s a couple companies trying it now. We’ve tried a few things in-house actually. I think it’s a year or two away. We’ve been told that, that part at least a year or two away. I think the getting the collegiate level data is probably a little farther, but we’ll make do. Every team has their own little tricks to get some of that signal out of the noisy data we do have. And we’ll continue doing that.

Brooke Wenig (13:34):
And speaking of collegiate data, how are the analytics for college sports different than pro sports?

Alexander Powell (13:39):
They’re pretty different. Mostly just from a budget standpoint, these teams don’t have full-time people me on staff. It’s becoming a little more common for them to have people around the program. So some places it’s a volunteer, some places it’s a student, a graduate assistant who also took stats as an undergrad or something. And there’s a few companies in the space trying to help out and assist with this problem that college coaches have, which is we’re now finally for the first time ever have data, but we don’t have the skills or the manpower to handle it. We’ve actually partnered with Davidson on a few projects, but they have a great student led program where they help the coaching staff with scouting and practice adjustments based on data collection.

Alexander Powell (14:32):
And it’s a bunch of stats and computer science majors who have basically said, “We like basketball. We want to assist and help.” And a lot of times those are great grounds for us to hire from, or learn from even. And I think the other big difficulty is we play 82 games, they play probably 30 usually. So, the sample sizes are a lot smaller. So you have a slightly different aspect in terms of how you’re going to look at and use that data.

Brooke Wenig (14:57):
And with those games, how often does the data team travel with the team so they can provide that insight at halftime or in between timeouts, do you have to call up the coach? Do people actually travel for away games?

Alexander Powell (15:09):
On the NBA level? Yeah, so we have one guy who travels with the team full time. I’ve been on a few trips, but it’s more from an assistance perspective, not sitting on the bench. But we have a guy whose job is to sit on the bench. We have another guy who’s back house sports science. And so he helps with… We have chips the players will wear pre-game and in practice to measure load and movements. So he handles that aspect of player performance. And I’m usually back in Charlotte helping with the analysis and the management of the data.

Brooke Wenig (15:40):
Can you talk a little bit more about the chips that they wear?

Alexander Powell (15:42):
Yeah. So each player has… We usually have them wear it on their shorts, just because that’s center of mask. There’s a little pocket in their shorts where they can wear something on their waist band. But basically it’s just an accelerometer. You can buy an accelerometer for 10 bucks on, on Amazon, but it’s a little bit high power than that, but it can communicate to a laptop that we’ll have sitting court side. But basically we measure all of those movements of a player. So you can start to measure their loads, how many times were they jumping. With a practice standpoint, are you grabbing a ton of information out of it? Probably not just because you practice so little in the NBA, but when you start to do that every single day of the year, and we’re able to use… We’re not allowed to wear them in games, due the CBA, but with the cameras, we’re able to estimate that information.

Alexander Powell (16:32):
But basically if you can have an hour, a long sample of that, you can start to make assumptions in terms of when a player’s tired. Should we work him more today or less today. So, a lot of times now if when there’s good communication, the coaches can say, “Hey, I’d like to do a few more drills, but do you think our guys have the energy for it?” And it really helps particular health and performance staff who do a great job at managing both the art of player performance and health with the data and the quantitative part to really be able to measure what’s happening on the court, as well as the weight room in games, et cetera.

Denny Lee (17:10):
I’m curious, does this potentially help with also when the player gets injured to monitor them to see how well they’re healing themselves up?

Alexander Powell (17:19):
For sure. So, we have a couple of tools. So force plates are one thing that a lot of teams use where the players jump on these plates, they measure the force. So you can start to see their explosiveness, their deceleration. And so we’ll based on their prior history, we can set these benchmarks for them to return to play, and be able to establish how healthy they are. And this is pretty common across sports. Actually, I would say soccer is one of the ones that is really good at that aspect of player health. Because you can start to say, “Hey, sure we know what all players are at, but how can we personalize this information to you and your body to get you back when you’re a hundred percent healthy?” Not just when you might look healthy or say you feel healthy, but how’s your body responding?

Denny Lee (18:06):
Well, then actually that really naturally segues to this idea that as they’re getting older, by the very definition, should their physical health be not as good, not as optimal. So are you seeing, as you’re recording over time, basically be able to adjust those thresholds or those benchmarks accordingly so that you can in essence, try to predict that, “Okay, you’re two years older now. You’re going to need to actually take a few more breaks or you can’t do as many free throws or you can’t do as many three pointers because of that.” I’m just curious if it’s gotten to that point yet, just curious.

Alexander Powell (18:40):
Yeah. But I think a lot of it’s more… This is where there’s a lot of the art to the data science, is having the people who can start to communicate as a guy gets… When he’s 18, maybe he doesn’t need ice bath every day, but as he gets to be 25, now you need to, just because you’ve had all of these impacts on your body, you’ve played a couple hundred games. But being able to have a lot of people on staff who can have that communication and that relationship with players with coaches, I think that helps to ease that curve and that that impact on the players.

Brooke Wenig (19:15):
How receptive are they to getting these piece of advice or suggestions like, “Hey, you should really take a risk because our data says that you’re overworked.” Do the coaches typically listen, do the players have pushback?

Alexander Powell (19:26):
Yeah. Now, luckily I’m not usually the one having to communicate that, but I think the sport’s grown a lot to where there’s really open dialogue. And we do a good job of sitting down throughout the year to make sure that there’s those conversations so that when it gets to a tough moment, that, that conversation’s a lot easier. I think players understand because it’s their health. That you have a lot of guys who they just want to play, but if you can show them why things are important for them and for us. We don’t really pull guys out of games, but it’s more of, “Hey, can you maybe get a little more sleep tonight,” or, “Hey, maybe we’re going to cut you off and practice today,” so that we can manage their information. And honestly, this is probably the least explored area of data science and sports right now. So we’re still learning, honestly.

Alexander Powell (20:21):
We’re still collecting a lot of data in this area. And hopefully within the next few years, it’ll grow and players are starting to be even more accustomed to wearing a chip and doing all of these things that maybe 10 years ago a player would’ve scoffed that because they never seen such a thing.

Denny Lee (20:44):
Cool. Well then now let’s switch gears a little bit, because you did mention the fact that you had reached out to college folks who potentially they’d be coming into the field of sports analytics. Let me just pose that question to you. If you’re interested in getting to the field field of sports analytics, what’s the hiring process like?

Alexander Powell (20:58):
It varies by team and situation, but typically during the year, it’s really hard for teams to hire. So a lot of those hirings happen in the summer, post-season. It’s summer league is a big opportunity for teams because this season’s ended, you hit that down period where you can start to focus internally on what you need or who’s left, or how you’re growing. The Sloan Conference, every March is a really good opportunity for teams or at least for students to… Or people looking to move into the NBA to share themselves. It’s basically all the nerds in the NBA go to MIT Sloan, and there’s a big conference every year. I think the biggest advice I usually give is finding… Anybody can say they know the math, the stats, the basketball, because it’s a rare…

Alexander Powell (21:50):
Data science is a rare field where you not only have to know the science, but you also need to know the industry you’re working in really well. And being able to show that you can do that, be it some portfolio of work, be it some blog posts that are a way to show what you can do, or just a portfolio you can share with teams and employers really can help you not just say that you can do the work, but show a team that, “Hey, this person knows how to do what they say they do.”

Brooke Wenig (22:18):
And so do you typically see people go into this field of sports analytics directly from college? Do you see them go in from other industries? Like they worked as an engineer, data scientist elsewhere, and then they transition what’s typically the career path like?

Alexander Powell (22:30):
I think it really varies. I know people who work in analytics and the NBA who were in investment banking, who were in data science jobs previously. Some who learned the analytics aspects while in the NBA. They saw it as a growing field and “Hey, this is something I’d like to do.” So I think it really varies. There’s a lot of people who enter as interns in those internships role into regular jobs. The NBA uses interns pretty heavy, not as a go grab me coffee, but just particularly the summertime, there’s a lot to be done. And a lot of untapped things that someone who’s an intern or working in college can do. And so we worked with college students even during their school year to say, “Hey, you have this project for school. Your professors recommended you, and we think it would add a value for us as well, as well as get to know you.”

Brooke Wenig (23:23):
And I’ve got to ask, I know with baseball, they use the term sabermetrics. Is there any equivalent term for using day data for the NBA or for basketball?

Alexander Powell (23:33):
Not really. I think people have tried to coin a few terms and just none of them stuck. They didn’t form actual words, which was the… They were more acronyms. But typically we say analytics, which I think is almost a misnomer because it’s not you have to be able to program and do all this stuff to analyze something. But it’s basically, we’ve called it analytics, even if someone’s a developer or if they’re an actual analyst, it’s all grouped into analyst. And a lot of times it’s, if you are in an analytics group, you’re the person who knows technology the best. So you end up also doing a little of everything, but it gives us good opportunities.

Denny Lee (24:18):
Got it. So you’re the IT shop as well, are you saying?

Alexander Powell (24:22):
Not fully. I’m glad I’m not. We have a good IT person, but you do end up assisting in a lot of ways that you didn’t expect to. But it sometimes gets your foot in a door. You didn’t think you would be in.d

Denny Lee (24:33):
No, I definitely say it in gist. But I’m actually curious then from that standpoint, because of the requirement for that type of domain knowledge, is there an implication that for sake of argument, now that I’m an expert “in basketball” that it can’t, or can, or cannot translate easily to another sport? Or do you feel that because some of the at least primitives are close enough just as long you know the sport that you can actually basically jump from one sport to the other, if need be?

Alexander Powell (25:02):
I think there’s definitely people who can jump from sport to sport. And I’ve done a few things in other sports and I’ve seen some people flip sports. But I do think it’s difficult because mostly because there’s a level of communication that has to happen with a bunch of non-technical people. Most coaches have either been coaches their whole life or they played, and then they became a coach and the same with front office people. So you have to really find a way to, can you speak a scout would speak? Where you information isn’t derived from what you saw with your eyes, but it’s derived from what you saw with numbers. Like, for example, I rarely give our GM or assistant GM a number. I would never say, “This guy’s doing XYZ,” from a number standpoint, it’s more of I’m saying it exactly the same way the scout next to me would say it, but my information’s backed by numbers because of that communication.

Alexander Powell (26:00):
If you were to throw me in another sport, I think I might be able to do that, but it would take a… There’s a lot of a big learning curve there. So it helped that I had played basketball so I could easily come in and fit in from a communication standpoint. And I tell the people who start in the NBA… We have a guy who started for us about two years ago. And the biggest thing I said was, “In meetings, just close your laptop and listen to our Scouts. Go sit with them at games, figure out how they speak so that when you go to communicate, it can be received.” And I think that a lot of people who are sports junkies can do that across sports, but maybe others cannot.

Brooke Wenig (26:36):
I think that’s some really great advice. It’s something that we commonly see throughout data science, which is focusing on the business problem. It’s not about coming up with the coolest model, it’s about what’s actually going to be used and convincing the stakeholders that they should trust your analytics and your predictions. And Denny, I think you’re asking about human transfer learning.

Denny Lee (26:54):
Yeah, absolutely. That’s what it boils down to because, what I find interesting is that number one, exactly to what Brooke called out. Super sage advice. Brooke’s talked about from the standpoint of machine learning, I’ll apply the same thing for BI, back in the day. Which is basically we were trying to show these cool BI models. Nobody understood what we’re talking about. So finally we explained what the output was and then over time they finally understood that. And so it’s exactly this human transfer knowledge where, which basically, as you’re learning it, you’re able communicate and then apply that over and over again. And provided that you actually know what you’re talking about if you decide to go to soccer, for example.

Alexander Powell (27:39):
Right. For sure. I think one of the things that I’ve tried to do here is, if we ask our Scouts to rate a player on a certain scale, all of our machine learning models will rate them on the same scale. There might predict something in one scale and we adjust it so that it… But that way it’s communicable. And we’re not saying, “Our Scouts are writing something on a qualitative scale, and then I’m over here saying on a zero to 10 scale.” Being able to give something that is similar, I think helps our executive making a decision to really digest it and use it rather than, “Here’s some neat numbers.”

Denny Lee (28:16):
Awesome. Well, I guess the only other thing I wanted to ask is any other advice that you’d to give, especially from the perspective of sports analytics that may or may not be the same as any other analytics, especially since you did call out there is the non analytics components of analytics in your arena, at least?

Alexander Powell (28:35):
I think the thing that’s just interests me the most is, I think even across sports or across data science industries that the things I’ve found that have had the most impact or the most value or even just the most interesting is the things that I can apply from. “Here’s something from quantitative finance or here’s something that they’re doing in soccer. Here’s something they’re doing in another sport,” because a lot of times we can just be in this bubble of basketball analytics. And we’re stuck with the same toolkit when a lot of times, if you know… Like we talked about, knowing another sport well enough. If you know it well enough to say, “All right, here’s analysis they’re doing. And it’s so common in that sport, but we’ve never thought how to use it in ours.” We can find a lot of those similarities. And I think sometimes we view it as, “Oh, we’re doing data science and basketball and that’s awesome.” But it’s sometimes not as dissimilar from these other places as it may seem, but it’s a little more fun.

Brooke Wenig (29:35):
Definitely more fun. But even the field of data science, it’s just taking concepts from other fields like math, electrical engineering, computer science. So I think having that pollination from other or sports can really give you unique insights into basketball and then vice versa. So I’m going to go ahead and close out the session. Thank you so much, Alexander, for taking the time to chat with us about basketball, sports analytics, some of the open questions and what we’re going to expect to see in the next one to two years. Hopefully, next time there will be a coin term for the Sabermetrics of basketball. But thank you again so much for taking the time, a chat with us today on Data Brew.

Alexander Powell (30:08):
Awesome. Thanks guys.