Data Based Coaching: How to incorporate data-driven decisions into your coaching workflow
/By Kieran Doyle-Davis (@kierdoyle)
I will start this with a disclaimer: this is not how Borussia Dortmund or Manchester United incorporate analytics into their coaching workflow, it’s not even how the Colorado Rapids incorporate data into their coaching. It is a look at the opposite side of the same coin Carl Carpenter examined earlier this week, only for a smaller school without access to the biggest and baddest equipment.
I am an assistant coach and data analyst with the University of Toronto Women’s Soccer program. I’ve been in my role for about a year now, in which we went from finishing 6th in our division (barely making the playoffs), to finishing 4th and going on a Cinderella story run to a national bronze medal. This corresponded with a totally new staff who deserve all of the credit. We’ll look over the timeline of a week, which is generally the unit these things work in for us at the University of Toronto. We are one of the lower budget programs (particularly in comparison to NCAA D1 schools), so this is written with the hope that you may be able to adopt it for yourself, whatever level you work at!
We play double matchday weekends, with one match Saturday, one match Sunday generally. You have the occasional midweek game, but generally you are preparing your week for two potentially very distinct opponents at the same time, all while working on you and building your game model.
Saturday and Sunday
Our matches are all filmed via a single camera set up from our press box at the stadium, which gives us quite a good angle to analyze video and tactical things accurately. Occasionally our games are streamed and we have access to the broadcast feed as well. On away matches we often don’t have a stadium with press box to film from, so we just use a hill, a bleacher, or whatever vantage point allows for the best camera angle available.
We don’t have a league wide data provider, so every game is hand coded afterwards on review of the match. I’ve been using some public event data trackers, which have a really nice interface. @PeterMcKeever’s Open Football Club is an excellent one which allows you to add whatever context flags you want for shots (and also has a nice xG model built in!), but be sure to fill out every single field if you want it to work. Similarly FC Python’s event tracker is quite good, and Ben Torvaney has a simplified shot tracker as well. They all are useful in their own way. Once we go through, we can start to look at how our evaluation of the performance matches up with what the data starts to say.
This shows the xG values by team for the game. We are in blue and Ryerson is in orange. The size of the dot reflects the value of the xG from the shot. Ryerson scored from their PK, thus why they don’t have a pink dot.
After both our weekend’s matches have been played and coded, we start to look at our periodization plan and how we can build that week’s concepts and prepare appropriately for the week ahead, both in terms of addressing the failures of the week before and incorporating things identified for the next set of matches. Sometimes this means throwing out a week’s worth of trainings scheduled for pressing if we need to focus on defending in our own third a lot more, but usually it’s more about small tweaks to what we’re already working on.
Monday
Unless we have a midweek match, the players have Monday off. It’s good to give them the break away, at the end of the day they are student-athletes not athlete-students, and they need the day to stay on top of their work (especially after a weekend of matches). Monday consists of three major processes: (1) clipping our film for individual player packages and team film sessions later in the week, as well as any telestration. In terms of telestration, we can draw lines and highlight players and areas on pitch, to essentially narrate a story to draw players attention to a specific content in a given clip. (2) clipping opposition film for team film sessions later in the week, (3) planning the week’s sessions. For example, in the match above we had a lot of trouble generating high quality central shots from open play. From the video, we recognize this as an inability to attack depth due to a deep block, and poor movement when the ball was in wide areas to try and find that space or combine and connect through. From the opposition video, we expected similar from the following weeks opponent. As such, the week consisted of a lot of attacking phases or combination exercises designed to find the space in the final third.
This shows the xG values by team for the game. We are in blue and Carleton is in orange. The size of the dot reflects the value of the xG from the shot.
Tuesday
Tuesday players receive their individual video packages, specific to their position. This will include some clips from the week previous matches, sometimes older clips if relevant, and some clips from the week ahead. The sessions will be planned out with room for adjustment depending how they go, especially if we’re bringing out new exercises. Tuesdays are generally a pretty intense day, with a bit of high intensity work and transition-y exercises and then some big picture stuff from there depending on the week’s topic of focus.
Wednesday
Wednesday is competition day, the player’s favorite. We do some game model focused ball-work to start, then it’s all games. Finishing games, passing games, small sided games, *really* small sided games (like 2v2+2 in the penalty area small). Goals count for individual points, winning a round counts for individual points, and a scoreboard is tallied and posted in the changeroom each week. At the end of the year there’s a prize for the top three finishers. Wednesday is a shorter session, where we get to do lots of individual work at the end.
For example, we had a striker who underperformed her xG significantly during the regular season, but her numbers were pretty great all around. She averaged 0.73 xG per 90 minutes, held the ball up well, got into good spots, connected well with others, and put in a lot of defensive work. We were really happy with her performance, but for a player it can be really hard to constantly get good looks and not finish them, even if it’s just a cold streak. So we looked at the data, and the majority of her misses came from a combination of 1v1’s at pace when she was played through and first time finishes. So any extra finishing work we did was focused on those scenarios specifically. Now, the obvious comment is that this is a super small sample size and any improvements are just variance smoothing out, which is true. But players aren’t ball-kicking robots and feeling like you’re working to fix the thing that’s going wrong is almost as important as continuing to do the great things and knowing the goals will eventually come. In our winter season (pre-COVID19 cancellation), she’s scored in every match.
Thursday
Thursday is a big picture day, we have team video before the session, then train with a focus on the things we saw in the context of that week’s theme. If we’re doing defensive things, how do they generate shots? Where do they generate shots? If we’re focusing on in possession, where do they give up shots? What are they trying to limit? Sometimes we can see this straight from video (e.g. their left back is always out of position), or sometimes we see big clusters on shot maps (maybe a bad zonal marker on set pieces, or poor protection of a cutback). We try not to stray too far from our own principles, but you also want to exploit what you can.
Friday
Friday is all about set pieces and match prep. We’ll do some very light ball work and passing exercises to get everyone going, feeling sharp ahead of the weekends matches. We’ll do a bit of finishing and small sided work, but a lot of the time is focused on set pieces. We bring the iPad on pitch to go over the weekend’s routines for each match, either with previous video or with animations. A lot of our catalogue is gently borrowed from some of the great set piece people on twitter like Stuart Reid.
We can go through the different iterations, both on the offensive and defensive side until we’re comfortable with where the squad who will be dressing are at for the weekend. This is secret sauce stuff, so no details here.
And that’s more or less the week. It’s far from ground breaking, but even at the lowest level we can use hand-coded data to provide actionable insight which has shown a real year-over-year improvement for basically the same squad. Here’s hoping for more of the same.
