A Conversation with Sportlogiq

By Dr. Brent Smale

Analytics is taking the hockey world by storm and every team is trying to get that upper edge whether it’s in regard to scouting, tactics, or performance benchmarking. One company that is at the crux of all this is Sportlogiq, which is headquartered in Montreal, Canada. Sportlogiq was founded by former Olympic skater Craig Buntin and Mehrsan Javan, a PhD graduate in computer vision and machine learning. These two have teamed up and secured millions in funding to provide high level data to the Swedish Hockey League, 27 NHL franchises, and media partners such as TSN, Sportsnet, and RDS. The hockey mind behind all of this is Mr. Christopher Boucher. Christopher is the VP of Sports Development, Analytics and Hockey Operations at Sportlogiq and was generous enough to spend some time talking to us at Apex Hockey about hockey analytics.

Thanks for sitting down with us Christopher. To start off, please tell us a bit about yourself regarding your hockey background and how that led you to Sportlogiq.

Thanks for having me. I began as a player and then a goalie coach who ran some clinics so hockey has been kind of what I’ve done forever. A long time ago I started wanting to learn more about the game so I started tracking every event in a hockey game. I started back in 1991 with a VCR bought with my first student loan and started recording games, where I would just pause and record every event I could such as passes, hits, dump ins, etc. I started writing about my analyses and got a little work with some NHL teams and then Craig reached out to me with some technology they had. We sat down, combined our systems and low and behold, that’s how the hockey side of Sportlogiq was born.

How long would it take for you to get through a single game when you first started?

Back then it was a long time – probably about six hours per game. Once the tech got a bit better such as an iPad, I could finish a game within 30 minutes of it being done. I was getting pretty good at it because I was doing it so much but yeah, it was kind of a fixation. Hockey has been the one constant for me forever.

So, what exactly is Sportlogiq and what kind of services do they provide?

We use my system to track events in a game by using some of Mehrsan’s PhD work to digitize the ice during a game’s broadcast. Through this camera calibration, we can see where on the ice everything is. Combining this knowledge with some BodyPose technology, we can teach the system through millions of images of what each player is doing on the ice. This allows us to provide data to the Swedish Hockey League, 27 NHL teams, 20 AHL teams, and broadcasting stations such as TSN, Sportsnet, and RDS. I should also mention that we also provide analytics for soccer and football in addition to our work with hockey.

Is Sportlogiq able to provide information on both an individual player and team level?

Yes, we absolutely can. We are able to track every possession play such as every pass, deke, dump in, shot, blue line carry, etc. Then on the defensive side we have every stick check, body check, blocked pass, etc. All of these events tell us exactly where the puck is and tell us which certain players are doing well. For example, is the player creating off the rush changes, turning the puck over, fronting the shooter, boxing players out, or creating space by his feet or with a pass.

On the team level, we are able to provide for example pre-scouting reports; how does the opposition treat a dump in or a cycle. Then also game summaries that are more than just the box score of goals, assists, and shots. We can really dig down to the granular level of how a team is performing and how a win or loss is being generated.

You often hear about the resistance to accept analytics some traditional hockey minds have. Do you believe traditional scouting and analyses can coexist with the novel methods that people like yourself are coming up with?

I actually believe that they are fundamentally the same thing. If we start digging down into what players are actually doing with the puck and that granularity of those events, you’ll be able to see that player’s strengths and weaknesses. We’re actually producing the same material that scouts are producing; we’re just able to do it with both more of a macro- and microscope. So, it’s really the same thing, we’re just able to do it with the broadcasts and provide that objective consistency.

What are some shortcomings of traditional scouting compared to your novel machine learning approach?

The foremost shortcomings would be sample size and consistency. Teams are taking multiple scouting reports from multiple scouts and trying to merge them together to find commonalities and that’s how they determine how to rate a player. We’re doing the exact same thing but just with added consistency. I think scouting will evolve from less of what the players are doing on the ice to who they are off the ice and what their situations are like. Perhaps even moving some scouts from the ‘hockey minds’ to those with a psychology background so that they can actually learn more about the players. I think teams would really benefit from this as I believe the hockey stats will soon take care of itself.

On the reverse, what would you say are the shortcomings of the analytics side?

The biggest shortcoming is understanding what the data is telling us. Sometimes people will cherry pick information to fit their narrative. That’s not how this works; the game is flowing and everything is interconnected. For example, Chicago leads the league in offensive zone possession time but they’re still struggling this year. Someone might come in and say that this must mean the data is not any good but instead, it’s just because they have players who rather than dumping the puck in, prefer to carry it and keep possession. If you just take one stat, you’ll never get the information you wanted to get.

How do you go about quantifying those immeasurable factors like ‘heart’ and ‘leadership’?

My argument for this is that it will show up on the ice regardless. If it’s heart or grit, the player would be expected to have the puck more by winning more puck battles or getting to loose pucks quicker. All of this will show up in the data because of the granularity of our system. If it’s not translating into the data, it’s not impacting the game. We unfortunately can’t quantify how a player impacts their teammates emotionally but a traditional scout wouldn’t be able to do this either; it’s really just a perception. I’m not discounting it, I just don’t think anyone can quantify it no matter their method.

With all this being said, I have a feeling that you believe the traditional statline of G/A/+- is insufficient. Are there any specific stats that you believe are very important that are traditionally overlooked?

For sure. If we wanted to start in a broader sense, there’s a player’s Expected Goals and Expected Assists because that’s going to give you something a little more repeatable. It also shows you what a player is doing regardless of what their linemates are doing. From there we like to focus on a player’s ability to get the puck to the middle of the ice and we call that a scoring-chance generating play and what this would be is any pass, drive, shot or deflection on net, loose puck recovery from the slot. It’s all of these things combined that allow you to see how much a player is producing and we pro-rate that per 20 or 60 mins. This allows us to compare apples to apples across elite players who can create offence.

You’re putting a heavy emphasis on the slot there but do you think that underserves snipers like Ovechkin, Laine, and Stamkos who are notorious for taking those one-timers from outside of the slot?

What’s important to remember is that about 52% of all goals are scored from the inner slot, which is a small triangle formed from the sides of the net to the midpoint of the hashmarks. Additionally, 78% of all goals are scored from the slot so that’s why we put so much emphasis on that area because that’s where the bulk of goals come from.

Well now that we’re starting to get deeper into the actual stats, do you have any favorite or distaste for any stats?

Well my favorites are obviously the ones I created – they’re like my babies. As for the ones I like, scoring-chance generating plays and possession-driving plays develops player ratings based on their impact on the game. These are the ones I lean on a lot when writing my reports. The ones I can’t stand are +/- and pure shot attempts. Everybody and their mother now talks about Corsi and there is value in it that should give you some indication of future performance but it’s more of a results-based metric and not necessarily a process-based metric. I prefer the process-based stats; the process of creating a shot or the process of defending.

I’m happy you brought up Corsi. Currently the only non-defenceman in the top 20 Corsi For rankings is Dylan Larkin. Why are so many people placing such an emphasis on an apparently biased stat?

I think it all started with the example of the Colorado Avalanche 2014/15 season where they severely regressed from their stellar 2013/14 campaign as their Corsi numbers predicted they would. Everybody kind of hung their hat on that and said look at what we can do with this stat. So that specific season made everyone including teams buy into it. However, you can’t fault people for trying to build metrics on public data and learn about the game using the information that they have. I see it as a good thing that there is interest and a love for the game urging them to learn more. It only turns negative when something is overvalued or thought of as more important than it actually is and that’s what happened with Corsi.

Speaking of overvalued, what is your view on blocked shots? That’s the difference between Corsi and Fenwick but there are two camps: the first saying blocked shots are good because you stopped a puck from going on net but the alternative is that if you have to block a shot, you failed to prevent that shot from being taken in the first place.

Yeah, I actually think both of those are correct. If your only recourse is to block the shot, then it’s a good play. If you’re blocking a lot of shots, it definitely indicates that you’re defending too much. So, both of those things are right but if you have the choice between a player who is defending a lot compared to a player who isn’t defending a lot, you will always choose the player who isn’t defending a lot. What indicates that is the player’s possession time and time spent in defensive zone; that’s going to give you the information you want to build a successful team. If I’m a manager and I see that a defenceman is blocking a lot of shots, I need to investigate further to determine why he needs to be blocking so many shots. I love those questions though because they are the ‘why’ questions and they’re the most fun to jump into.

One scenario where it is more black and white though is on the penalty kill because you are supposed to be killing time. To do so, you need to have more mobile players that can get to loose pucks and dump the puck out. That should be your focus instead of trying to block oncoming shots. I feel that there has been more of a shift in the coaching mindset towards this but there are still some guys who just want the shot blockers.

This leads very well into something I would love to get your input on. Don Cherry is famous for going on Coach’s Corner and arguing for defenders to get their sticks out of the shooting lane. Do you agree or disagree with Don on this?

His argument is absolutely anecdotal – it’s not based on the numbers at all. There’s definitely an advantage to getting your stick in there and blocking the shot. The amount of times you end up deflecting the shot into the net versus the amount of times you deflect the puck away from the net is not even close. It’s Don Cherry and anecdotal so those two things go together really well. 

Then putting yourself in the Coach’s position, how does Sportlogiq help coaches and young players improve their game?

I think using the data that we have we learn what’s important to both players and teams. Just looking at the skating side, we can look at for example who’s getting to loose pucks first or beating other players one-on-one or a player’s ability to escape. Beyond that, the BodyPose data is something that can definitely help in terms of how you are skating – your leg extension, your knee angle for example – holds value to young players. It’s all about seeing what the best players are doing and making that the focus for training younger players more than anything else.

Another area we hold an advantage is the amount of certain game situations we see. For example, the D-zone loose puck recovery by the defenceman after a dump under pressure is very strongly related to amount of time spent in your zone. Now we have something tangible that we can actually focus on in practice. We also have the information about decision making and the appropriate action to take based on the probabilities of successful future outcomes. If we take this tangible information and start putting it into practice, there’s a huge positive upside to be had.

And if someone is looking to learn more about Sportlogiq, where would you direct them?

Well, we keep most of our stuff pretty close to our chests because of our contracts with the teams but our website is Sportlogiq.com, Sportlogiq on Facebook, and on Twitter we are @Sportlogiq. There you can find any news and some other hockey insights.

Fantastic. Thanks, Christopher for taking the time out of your busy schedule to sit down with us – We really appreciate it!