December 18, 2012

Moneylax? Stats Meet Lacrosse with Krossover Video Analysis

by Matt Forman | LaxMagazine.com | Twitter

The Syracuse men's team is one of Krossover's lacrosse clients. The company expects to have between 400-500 clients at all levels of the sport for the 2013 spring season.
© Greg Wall

Baseball was forever changed by sabermetrics, the quantitative analysis of on-field performance initially popularized by Bill James used to better evaluate a player's production. The practice eventually achieved widespread fame in Michael Lewis' best-selling book "Moneyball." Shortly thereafter, basketball began gaining recognition for its similar statistical research. Former ESPN.com writer John Hollinger, who recently was hired as the Memphis Grizzlies' vice president of basketball operations, created many of these advanced metrics for hoops.

Is lacrosse next? If Vasu Kulkarni's firm Krossover is any indication, lacrosse is already on its way to embracing a data-driven approach. Krossover, a New York City-based sports technology company and online video analysis platform, could have 500 lacrosse clients for the 2013 spring season. Coaches are noticing: At the IMLCA convention, held Dec. 6-9 in Baltimore, nearly 150 coaches packed a conference room for Krossover's appearance on "Bringing Moneyball to Lacrosse."

Kulkarni, who played basketball at Penn, first developed the Krossover software for hoops. The company's efforts were featured in various national publications, The Wall Street Journal, Forbes and USA Today. Recently, Krossover has been focusing on its lacrosse application. Football is also on its way.

LaxMagazine.com last week caught up with Kulkarni, Krossover's founder and CEO, who also shared slides from VP of Analytics James Piette's presentation at the IMLCA convention.

Applying the "Four Factors of Basketball Success" to lacrosse, and framing success in terms of possessions — and efficiency and pace of them — Krossover contends the most important offensive metrics include: shooting (total shots), controlling possession (turnovers and failed clears), maintaining possession (offensive ground balls) and penalties (man-up scoring percentage). The most important defensive metrics: shooting (opponent total shots), limiting possession (opponent's turnovers and rides), regaining possession (defensive ground balls) and penalties (man-down scoring percentage).

The interview...

Tell us about your background? How did you get to where you are today, a 26-year-old entrepreneur, after learning to play basketball in India?
I was raised in India, as the biggest self-proclaimed basketball player and basketball fan in the world. I would have people record playoff games here and ship the VHS tapes to me, with relatives when they were traveling, so that I would get to watch games. There was really no other way to learn it. There are coaches in India, but they're 'coaches.' They really didn't know what a pick-and-roll is. It's getting better, but when I was growing up — which was not that long ago — it was still pretty poor. When I think back to what my coaches used to teach me as a high school player in India, it's unbelievable to see how little they knew about the actual X's and O's of the game. I grew up learning how to play watching videos that would be shipped to me from people here, and then we would play. By the time I was in high school, we had an asphalt court. There were still no indoor courts. I had never played on an indoor gym, on a wooden court, none of that stuff, until I got to Penn.

I came here thinking I was going to be playing Division I ball, because I had always been the best player on my high school team and I was the captain of my high school team. I assumed that if you're good there, then you'll be good here. Little did I know the disparity in size and skill between the two countries, so I was in for a rude awakening. But I did end up making the JV team at Penn, so I played at The Palestra. We had the NCAA official jerseys that were hand-me-downs from varsity from the year before. We got the shoes. We got to sit in the same locker room and come out and play on the same floor in the most historic gym in the world. For me, that was the greatest experience of my life.

I had always sort of been an entrepreneur, and I would always figure out ways to make money. I paid for college by running an eBay store out of my dorm room at Penn. I had always been resourceful, so coming out of my senior year I kind of figured that I wanted to do something in sports. I applied to all the jobs available at the NBA and Nike and whatnot. I don't even think I ever heard back from a single one of those jobs, which is funny because now I know everyone there and I could probably get a job pretty easily. But back then when I really wanted to work for some of these companies, no one even returned my calls.

I'm sitting there my senior year, I'm saying to myself, 'I don't have a single piece of footage that I could show my parents back home in India to show them that I played at The Palestra.' Or my kids 20 years from now to show them that a 5-foot-9, 135-pound Indian kid actually played for a D-I program. No one's ever going to believe that this happened to me. I thought about it, and it's not that there's a lack of film. Everyone films their games. It's just that it's such inefficient system and over time those tapes just get lost. You put them on a DVD, they get lost. You put them on a hard drive somewhere, you invariably run out of space and you delete games and that's why companies like Dropbox are able to charge people for archiving. It's because there's just such an inefficiency with dealing with large files and archives.

So I said, 'Great, let's build a business around archiving all of this old footage, because I think there's great sentimental value in it.' Being an engineer and a computer guy, I said, 'While that's fun and games, we can really make a difference in terms of building a data-driven approach in sports. I had actually interned one summer at a business intelligence company. That was sort of my intro to how data was being used in the enterprise level to make decisions. I double-majored in computer engineering and entrepreneurship. And my third major, I like to tell people, was basketball. I kind of put two and two together and I said, 'What if we could apply the same concepts to sports?' That's really how the whole thing started. I did a ton of research, saw what was out there and we got to where we are, but that was really how the whole thing started. I graduated in the end of 2008, so middle of 2009 is when we really started working on the company.

A slide from Krossover VP of Analytics James Piette's presentation at the IMCLA convention in Baltimore.
© Krossover

Do you feel like you've come along at the right time in sports, when people are most open to analytical and data-driven thinking?
As with anything in life, so much has to do with luck. Had I been five years older, five years younger, who knows what would have happened? I may have been too late to the scene, I may have been too early to the scene. From that standpoint, I do think we came out at a relatively good time. I probably had an extra year or two.

If I had started the company in 2010 instead of 2009 or 2008, I think we would have still been fine. And at the same time, if we had started in 2006, 2007, I think we would have been OK. I think a lot of it has to do not just with people's acceptance of data, but also with just infrastructure in general. Uploading 10 gigabyte HD files over the Internet even today is a pain. We have to spend so much of our technological resources solving the problem of how you get such a large file over the Internet. Currently we're at the point where we're breaking down 300 games a night. So when you put that into perspective, it's terabytes of data that you're trying to bring up from people that may not necessarily have a very good, fast internet connection or even know what they're doing. These are not the most technologically savvy people we're dealing with. Five years from now, hopefully that problem will have been solved in a greater way than we've been able to do.

When you're talking about the general acceptance of the data-driven approach, there are still a lot of old-school coaches out there right now who grew up doing things by hand, doing things manually, and a lot of them are still going to be around for the next 10, 20 years, maybe even. And those guys, some of them get it, a lot of them don't. A lot of them go on gut feel, they say, 'Hey, I've been doing this for 10, 15, 20 years. Who are you to come in and tell me that the motion offense set that I run is not the best play to run? I run it all day long, and we've been doing just fine.' If you're one of those people, I don't have an answer. I mean, I do have an answer, but I know that they're set in their ways.

What is encouraging, though, is the number of coaches that are coming in to their own at this point, or my generation — slightly older but they kind of came through this time where people are using computers, people are using iPads, iPhones. That's what I think is the most important. The general idea of using technology on the field over using just your gut feel and your eye. Every company — Nike, all these guys — have scouts who go out and try to find players for them. Colleges have scouts who go out, and a lot of times these guys just sit on the sidelines and by just watching someone for 20 minutes they're expected to make a decision that will completely change the company or their team for the next four years. They don't want to have to use any sort of data tool or technological tool, because this is their job and that's what they've gotten used to.

I think a lot of people see that as a threat. A lot of people are looking and going, 'Wow, if everything's going to be done online, my job as a scout may be diminished.' And to a lot of those people that I hear that from, I say, 'Listen, I don't think the job of a scout is ever going to go away.' There's still so many intangibles in sports that the data can't show you. There's so much anecdotal evidence of players that you have to be there to see and feel, that I think that their job is fine. I just think that your job can be done so much more efficiently if you're willing to embrace the data and embrace technology as opposed to sort of being stuck in your ways.

Offensive and defensive statistical concepts from basketball can translate to the lacrosse field.
© Krossover

How would you describe what Krossover does?
When it comes to lacrosse in particular, the market is slightly different. If you talk about basketball for a minute here, when you look at the highest level — the NBA — these guys are now using missile-tracking cameras inside the stadium, shooting at 60-frames-per-second and tracking the exact X-Y coordinates of each player on the floor and the ball, 60 times a second over the course of the entire game. At the end of that game, the NBA teams do a giant data dump, with which they literally know how fast each player was moving at any given time. They know every shot that Player X took, how close was the closest defender to him. The level of detail is just unbelievable.

The question arises: What do you do with all that data? How many teams actually have a data scientist who can sift through all of that information and come up with the two or three points, which can potentially be the difference between a win and a loss? I don't think any NBA team has figured that out yet. If they had, one of these small-market teams would be killing it. The closest that anyone has come is the Oklahoma City Thunder, and they did pretty well last year.

You look at that and then you come to a sport like lacrosse, where almost nothing is being done, for various reasons: A) It's a brand new sport; B) It's not a very big-budget sport; and C) There's no real professional league. The MLL isn't exactly making tons of money. When we look at what we do for a sport like lacrosse, we came in on the ground floor. No one has ever attempted to do anything like this for lacrosse. What we try to do is: A) Save the coach a ton of time — time that you'd otherwise be wasting and doing manual grunt work, which you shouldn't have to do. We're saying, 'Listen, let us do this so that instead you can just spend your time trying to make sense of the data that we're collecting for you. You don't collect the data. Let us do that part. Let us make a few inferences for you and then now you spend the rest of your time trying to figure out how that translates to your team and what you can do to get a better chance of winning.'

This latest talk one of our guys gave at the IMLCA Convention, he was talking about some of the lacrosse statistics that you can just look at a box score, [they] might be misleading. People look at certain stats and they go, 'That's one of the most important stats.' Faceoffs, for example. A lot of coaches look at, 'Oh, well, guys we have to win every faceoff." Or, 'If we win X percentage of faceoffs in a game, we're giving ourselves a better chance to win,' whereas if you look at data over a period of time you'll see that may not be nearly that important. It may be one of those stats that, regardless of what happens, doesn't really have that big of an impact at the end of the game.

Because no one's ever done this before for lacrosse, we're still early on in really trying to figure out what are those most important set of statistics that a coach or a player should be looking at on a nightly basis over the course of a season? And you can say, 'If X, Y, Z align, then we're giving ourselves a better chance of winning this game.' That's really our goal.

One of the most interesting things an investor once said to me — this was right when we were getting started and I showed him all the stuff we were planning to do and blah, blah, blah — and this guy looks and me and he goes, 'Come back to me when you can tell me how I can win a game. If you can come back and tell me the exact formula for a W, I'll give you as much money as you want.' I don't know if we will ever get there, and I don't think it's possible, because I think so much happens in a sport that is so dynamic and that's why we love it. The dream is to always get as close to being able to determine a win or a loss. That would be the Holy Grail. That was Seth Berger, the founder of And1. He started And1 out of Penn, and he sold the company quite recently. He's now a high school coach in Pennsylvania.

"Because no one's ever done this before for lacrosse, we're still early on in really trying to figure out what are those most important set of statistics that a coach or a player should be looking at."

— Krossover CEO Vasu Kulkarni

What does some of your early research suggest in lacrosse?
Similar to basketball, what we've said is there's something sort of like the four factors. In basketball we have this thing called the four factors. In lacrosse, we stress that shooting is important, that raw shooting percentage or total shooting attempts is often enough to look at and then also you want to look at your percentage of shots on goal. And the other things that we look at are things like controlling possession, so that's turnovers plus your failed clears, maintaining possession — winning your offensive ground ball battles — and of course penalties.

And then you also talk about defensive stuff. So things like opponent shooting, how often can you force a loss of possession and then regaining those possessions on which you've forced the other team to lose possession and then of course defending penalties. Those are some of the things that we look at closely for teams. Last season we had probably about 20 or 30 Division I programs, and our data guy was kind of spending some time looking at the D-I stats that we collected, and trying to look for trends and see if we can find anything interesting.

My favorite example is one I give for basketball — again, I'm a hoops guy, so even though we're talking lax, I always like to go back to hoops, and one of my favorite stats that I show people is we calculate for a point guard what percentage of his assists are coming near the rim, in the paint, in the mid-range and beyond the arc, so a lot of location-based stuff. And we're doing some of this with lacrosse, as well.

We generate a shot chart, and we generate an on-goal chart, to sort of say, 'Where are you getting shots from that are successful?' And then from a goalie's perspective, 'How do you score against a goalie?' And then for the goalie himself you want to look at, OK where are the opponents' shots coming from? Where are they coming toward the net? What do I need to do? Do I have any certain regions or certain quadrants in the net where I'm just letting balls through all the time that I need to work on?

Visualization on some of this stuff is important. Stats are just numbers. Sure, some people are numbers guys and they can make sense of it, but a lot of people — talk about your average high school coach, regardless of what sport — if you just threw a bunch of numbers at him, I don't think most of those guys are going to figure out what they're supposed to do with that stuff.

Our goal is really to start to visualize this stuff and give them things that really help them understand what the numbers mean. And to that effect, we've started to really think about using natural language processing to convert the data into things like scouting reports and recaps. Come next year, we'll have the computer essentially generate a scouting report on your team and the other team that literally reads as the way coach spent six hours getting their information and writing out sentences.

But that's the beauty of technology — you can use a natural language processor to write out sentences and talk about strengths and weaknesses of players and teams. What our coaches at Penn used to do before we'd play a team is they would have a seven-page booklet that they would have prepared on the other team, and they would have spent two days, four guys sitting in a room with PowerPoint trying to put together a scouting report. And now to think that you can do that with a computer writing sentences for you in two seconds is just, it's really cool.

Canadian-born Division I players have in general had "a significantly higher shooting percentage" than Americans, Krossover CEO Vasu Kulkarni said.
© Trevor Brown

Any other ongoing trends?
One of the interesting things my guys found is thatCanadian-born lacrosse players in Division I had a significantly higher shooting percentage. That's one of those interesting stats. Things like that are pretty interesting. If you ever read Malcolm Gladwell's piece on hockey players and why the ones born in Canada — because of the age cut-off they end up moving on to become professional hockey players moreso than people born later in the year — those sorts of interesting things that, whether or not they have a direct impact on the game, I don't know, but those are always interesting little tidbits to talk about.

It's just been fascinating to see how the game is growing. Three or four years ago, I had barely heard the word 'lacrosse.' We had a lacrosse team but I never went to a single lacrosse game, whereas now it's just becoming so mainstream and looking at the number of attendants at the conference, our talk was so well-received. We were worried that there might not even be 10 or 15 people showing up — there were three other talks going on at the same time. We had about 150 people show up for this. So just that one thing alone — the fact that there was that many coaches that picked our talk over two others that were going on at the same time — is a very, very heartening fact to me. That makes me feel very, very good, that these many coaches looked at the word 'Moneyball' and got it. They understood it and said, 'We'd better go see what this is about.' That's great.

What Division I programs have embraced Krossover?
Syracuse, Penn, Princeton, Brown, UMBC and Georgetown. What was interesting to us: Because we're Penn alums, we got an introduction to Penn coach Mike Murphy, probably two years ago, before we had even started building the lacrosse product. We went to him and coach was like, 'I love stats. I'm all about stats.' He showed us his giant booklet of things he was doing manually — shot charts that he was having his staff put together. He said, 'Guys, I'm really excited about this.' Fast-forward six months, we built something and we took it back to coach Murphy. He said, 'Guys, that looks really cool. But, you know, I don't know if I need the film to be broken down, we break down our own film. I'm happy to help you. but I don't know what I'm going to do with this.'

Fast-forward another six months, we come back to him with all these new stats, the shot charts. He looks at it and he goes, 'Holy crap.' Overnight, he decided that having his staff do this manually was pointless, when they could just pay us $1,000 and it's done for him overnight. He is a huge stats guy. He has a numbers thing that he does himself — those are private stats he keeps that he thinks are important, and those are obviously proprietary. He was like, 'These are the things I like to look at. Show me how I can get those numbers from the stuff you guys are doing.' We gave that to him.

The guys at Georgetown loved it as well. They were heavy users too. The other big thing, with the colleges, they use it not just to self-scout but also a lot them use it to scout out their opponents. Before every game that a team plays, they'll give us two or three games of their opponent that they would've gotten through their conference tape exchange policy. They'll give us that, and we'll go through and break it down. Most of the Division I programs find that to be even more valuable than seeing their own breakdowns, because they know what they're doing, they want to know what the opposing team is doing before they play them.

How many clients do you expect to have for the 2013 season?
It's pretty hard to say, I'll just be throwing a number at you. But we are shooting to have between 400-500 customers, between high school, MCLA, Division II and II, and Division I. We're at 860 or so basketball customers right now, and we're still selling them. The season already started and people are still buying, so we might hit 1,000 basketball teams this season. Our hope is to come close to half of that for lacrosse, which is fantastic given that there's one-tenth the number of lacrosse teams than basketball teams out there.

In that regard, the lacrosse community has embraced Krossover?
We've met with a number of lacrosse guys out there, like Body By Jake, who runs the MLL, and all these Division I guys who showed up at the IMLCA convention. That was very heartening to see. Between all of these things we've got going on. Sports is such a word-of-mouth industry. If two schools in the district are using it, all the sudden after waiting a year, the other five guys in the district have found about it, and they think, 'If we don't buy this, we're at a disadvantage.' That's the great part about coaches: They want to win, and they'll do anything to win. If that means they need to buy a $1,000 piece of software to even the playing field, they'll do it. That's what we rely on. And, so far, it has worked in basketball. I can't imagine lacrosse coaches being any less motivated than basketball coaches.

How many staff members do you employ?
We're 21 people, here in New York City. We have the people who actually do the logging, and there's a good amount of manual labor that goes into tagging the film. We have about 600-700 freelance people that just tag video for us. It's become a marketplace essentially. We kind of look at it like what Uber did for taxis: Being able to build a place where taxi-drivers, on command, can find a job, pick someone up, and then do things like search pricing. They can lower the price when there isn't as much demand. We kind of see ourselves doing the same thing with film breakdown. We have this giant pool of people that want to do this job. There's kids in colleges who, instead of doing their $7.25 library job, they want to be doing something fun. Their breaking down film, watching sports.

There have been so many success stories, guys like Miami Heat coach Erik Spoelstra, who started as a video coordinator and is now a head coach of an NBA team. Those are the stories that kind of help us, because now there's this whole group of kids who look at this job as being an entry point into the sports world. They want to do this, because now they can go work for an MLB team three or four years from now, and they can be the in-house film and stats guy. We've built this giant pool of people, and it keeps growing.

Literally, on a daily basis, we get between 40 and 60 resumes of kids who want to be part of this marketplace. We have two full-time staff members whose only job is to vet and train loggers, and make sure they're ready to go. Any given night, if we have 300 games coming in — which is what we have coming in now, at the peak of basketball season — we have two or three times that many people who are ready to break down those games. We know we can turn games around very, very quickly.

We can scale this model up as high as it goes too. When we get to the point where hopefully we have 5,000 to 10,000 customers, and we need to break down 1,000 games a night, it should be no problem. We should be able to have those people, and there's no overhead. We don't have any office infrastructure and absolutely everything can be done remotely. It's a sign of the times. If we had started this five years ago, people may not have had fast enough Internet to be doing this job from home, but today people can do it, and there's no problem.

The recruiting scene could change with a database of every high school player in the country, complete with a statistical analysis of each, Kulkarni said.
© Kevin P. Tucker

What's the future for Krossover?
There's three or four different ways this can go. There's the media angle of what we're doing. Because we're aggregating so much content, we can generate automatic highlight footage, or we can generate automatic recap stories. We have a shot chart, we have a box score, we have a play-by-play. We can essentially do what ESPN does for the 1 percent of sport, we're trying to do that for the other 99 percent. I see this being — once we get the critical mass — we have the potential to be the ESPN for everyone else, for high school kids, Division II and III kids, and even some Division I lacrosse programs.

How many stories are out there about Penn lacrosse? Sure, they get coverage every now and then. But on a nightly basis, nobody is giving all of these sports teams all the kind of coverage they want or deserve. We can do that, because we're doing it programmatically. We don't need to have 200 writers on staff. We need 0, because there are computers that will write all the stories for us. That's one of the places we could go with this.

My goal is that any media outlet in the country can buy a subscription, come in, and on any given night they can say: 'I want to write a story about Team X from last night.' Pick them, and it gives you the option of highlight reels, box scores, stories, and then on your site we do a revenue share. When you think about how much effort it takes to go from raw video footage to actually cutting up a video and making a highlight reel, it's absurd. That's why no kid does it. How many kids put up highlight reels on a nightly basis, or weekly basis? I bet the average kid puts up a highlight reel at the end of his high school season, because it takes him six hours to do it. But you could be so smart about it, picking out the most important two minutes of a game footage. Algorithmically, you can do that and generate two-minute highlight reels that can be shared on news sites. That's my vision for a media product, where it's a self-service, picking a team or player that you want a story about. Within 30 seconds, you have everything you need that you can embed on your site.

Recruiting is another place we could go. As much of a dirty name as recruiting gets, if we can make it efficient and cleaner, and essentially say to every college in the country: 'Listen, we have a database of every high school kid in the country. We have his height, weight, GPA, and every single statistic about him, as well as his video footage.' That way coaches can come in and say, 'I'm looking for a goalie whose save percentage was greater than 53 percent, and he had a GPA of better than 3.2, and he lives on the East Coast.' In 30 seconds, our system can find every single player who matches that criteria and hand it over. It reduces recruiting budgets, and hopefully creates a little more transparency, as everything happens electronically there's a trail of everything. There's always going to be dirty business in the recruiting world, but if you can change it by 1 percent, then fantastic.

We're working on some other things, for professional teams, to help with draft decisions, trying to figure out player IQ levels when it comes to a sport. We're working with some other companies on technology that people wear. You've seen so many of these devices, like the Nike FuelBand and Under Armour is working on clothing that's smart that has sensors on it. All of those things feed into what we're doing, where it's just another layer of the data.

If we know video, and we know what happened in the game, and we can add on another layer — the bios, the movements — now you've got something the world has never seen in a third dimension of data. Who knows? The next LeBron James may end up being an unbelievable player, not just because of his God-given ability, but because of his ability to make good decisions, and now you're fine-tuning the athlete. As great as LeBron is, if you could fine-tune LeBron and tell him exactly how high he needs to jump in order to have the highest possible shooting percentage, and how many steps he needs to take while going left versus right, and you can figure out how fatigued a player is at a particular time, and therefore know the exact time you should substitute him or bring him back in — those are the things that eventually add up to a win or a loss. Trying to get to that secret place where you know how you can make a team win, that's what we're shooting for.

Based on your analysis and ongoing trends in lacrosse, what expectations do you have for the 2013 season?
From our perspective, without making any crazy predictions, we don't have all the data. We're dealing primarily with a lot of the data we create, just because our data is far more robust than anything out there, so we look at that. We don't get to necessarily look at every single team in the country.

We were pretty surprised with how poorly Syracuse, as well as Penn, did last year. Penn did have a pretty tough schedule last year. But this year, in 2013, we're expecting both of those teams to do a lot better. Syracuse should be doing significantly better. And I don't just say that because they're our customers. I'm predicting we're going to see some big things from Penn and Syracuse this year.

I wish I could say we had data on Loyola from last year. Kentucky basketball was a customer of ours, and that worked out real nicely for us.


comments powered by Disqus