Could Big Data Have Helped the Montreal Canadiens Win?
In a conversation with the CEO and co-founder of Sportlogiq, Craig Buntin, I learned what his computers see and how they might have been able to change the outcome of the Canadiens’ playoff run.
Every Habs fan is in mourning today, and we are all wondering what the Canadiens could have done differently. Could we have cheered harder, grown the playoff beard earlier, been more religious about attending games during the regular season? Did not the coaches do everything they could? A friend and hockey buddy, Jake Chadwick, recently turned me onto a Montreal startup that has me pretty excited for the future of my city’s team, as well as the entire sport, hell just the future in general.
Sportlogiq is working out of TandemLaunch in Montreal to give computers the ability of sight, also known as “computer vision”. Sportlogiq is focusing the power of computer vision at one thing: hockey. In a game of inches, Sportlogiq aims to give coaches, scouts and players the power of Big Data insights to find that elusive edge for the win.
Computer vision is a broad, open landscape of possibilities. Giving computers sight is one of the major precursors to artificial intelligence, which is still a ways off, but will have massive rippling effects. You probably already use it without realizing it. Facebook has automatic tagging, Google will find you similar-looking images through visual recognition cues and Flickr 4.0 will automatically organize all your photos by style or subject. Computer vision is the cornerstone to future things like Google’s self-driving car.
But back to hockey!
What Sportlogiq does is it can take the main video feed from a game and convert this into a massive data capture of what every player did on the ice. It can tell us how fast Subban was skating, how hard Prust checked the goalie and even how effective Gallagher’s gritty play actually was. At this point Sportlogiq can capture 93 different possible events that can happen in a hockey game (a hit, a deke, a wrist shot, stretch pass, dump-ins, etc.). It then maps these out to provide a dashboard of metrics and patterns that can be analyzed by coaches and staff.
Sportlogiq takes the main video feed from a game and converts this into a massive data capture of what every player did on the ice.
Why focus on hockey?
We focused on Hockey because we’re Canadian. It’s what we do, we know hockey and we know we can totally change the sport.
Hockey analytics are so far behind other sports like baseball. In baseball you have all these discrete events: a pitch, a hit, a run, a throw. Hockey is just so dynamic that to try and put a simple algorithm on it and find your one stat that matters is really difficult, if not impossible.
By focusing computer vision on hockey we’re going to do analytics better than anyone else in the world. Nobody other than us has the amount of data that we have on hockey. So, the people who will find those unseen metrics that matter will likely be us.
So could ‘moneypuck’ be right around the corner?
If you use ‘moneyball’ as an example, the Oakland A’s invented it and when they deployed it they were the top team in the league. As soon as most teams started doing it, basically you had these pockets of teams that weren’t and were just simply at a disadvantage and playing catch-up just to stay with the pack. That’s what happens, the first teams that jump on it are going to have a real advantage and once you’ve got half the league using it, everybody else has to use it otherwise they’ll just be at a disadvantage.
Has hockey embraced big data?
Last summer was known as the ‘summer of analytics’ for hockey. Most teams staffed up with data scientists and you even saw a lot of the fan sites being shut down because the teams were buying the data and then hiring the people.
This is good news for Sportlogiq because now we have data experts in the hockey clubs who appreciate the value of our solution. But the first team we talked to just flat out said ‘we don’t believe your system is accurate enough.’ So, they took their analyst team and data-mapped 3 games by hand themselves. We took those same games, ran them through our system and then we compared notes. We were 4–8% off. We were really disappointed until we manually went back through everything our system had found and realized our numbers were correct. The error was human on their side and the team’s data scientists saw that Sportlogiq was more accurate — and in a game of inches, 4–8% data accuracy matters a lot.
How do you keep from getting drowned in data and zero-in on actionable information?
We’ve really spent time with the scouts and the coaches to listen to their conversations. We listen to the subjective things that they say and then try and get the data to say that.
For example Brendan Gallagher’s success is often chalked up to his “gritty” play, so we had to figure-out how do you capture “gritty” data. We broke it down into where he spends his time (in the corners, in front of the net), the number of body checks vs. stick checks, the number of events he was engaged in, the number of times his hits got the puck back.
But we all “know” Gallagher is great, what’s the point of quantifying it with data?
Brendan Gallagher was a 5th round draft pick, 147th overall. Recruiting him was a bit lucky. Finding these diamonds in the rough is now an established standard in baseball, but hockey lags far behind.
A team knows who they are going to pick in the 1st round of the draft, and maybe which players they want in the 2nd. When it comes to the players they want in the 3rd, 4th, 5th rounds, the scouts haven’t spent that much time looking at all of these players. The whole bottom third of the prospect list isn’t getting that much visibility, and multi-million dollar contracts are getting signed with out much validation. What we can do is take an entire season of CHL footage, or KHL footage, or footage from Europe and put thousands of athletes into a database and really start seeing the specific attributes of players that might have gone unnoticed. We are finding things where people just aren’t looking; you can even scale this right down through to a parent filming their child in youth league.
[Sportlogiq also works on most YouTube and cellphone video — it’s safe to say Big Data just got bigger. Considering the amount of hockey footage out there, there is now a deluge of new data that just became available because of Sportlogiq’s software. Craig had to ask his co-founder and computer vision Ph.D, Mehrsan Javan, to make a few calculations to get an estimate of just how much we’re talking about…we are still waiting.]
So beyond scouting how can Sportlogiq help a team?
Often times a perception of what a player does is off from what you actually see reviewing a game. There was another team we work and had a debate about how engaged a player was in the game. They believed a player was really involved in a game, but when you actually broke it down and looked at each of the individual plays we all said ‘nope’. The number of puck touches weren’t there, the number of hits weren’t there. Perception is a huge factor when a coach has to look at 20 players.
We trust our guts for those decisions where the objective information isn’t there to find the clear, right choice. Data helps us find clarity in more and more complicated decisions, and make our guts more informed for when we still need to turn to them.
Have you been working with the Habs?
When we sign a team on we agree that we won’t disclose which teams we’re working with or what players we’re looking at. What I can say is we are working with 4 teams in the NHL right now.
Could this type of data have changed the outcome of last night’s 4–1 loss for the Canadiens?
To be honest, I think they had a system going and they know what they are doing and if you tried to disrupt that you might be swimming upstream. That said, if today they were to start using the Sportlogiq data for draft choices and start the first day next season using the tool for coaching, I think you would see a different team. There would be significant improvements across the board.