That's bait
I've had an unusual disinclination to write blogs lately.
It's not for a lack of material. My LinkedIn feed throws shiny posts at me, trying to bait me into writing, but no. Enough stuff has built up now to make for a good grab-bag, though, so here we go with the links and segues...
In case you don't read to the end, have a good (northern hemisphere) summer. Get Goalside will probably be back in inboxes at some point. Feel free to get in touch in the meantime.
Statsbomb released data from 2023-2024 for five top women's leagues: the WSL (England), Liga F (Spain), Frauen Bundesliga (Germany), Serie A (Italy), and the NWSL (USA). We like this, this is a good thing. Related, the lines to submit proposals for this year's Hudl Performance Insights conference are now open.
US Soccer changed the paradigm by giving Mauricio Pochettino a laptop to use during water breaks in pre-World Cup friendlies, putting him an air-conditioned office away from becoming a Daily Mail bête noire. He used it to show players video clips, but this is a step towards the dream of teams being able to see their own beautiful data in real-time.
Speaking of seeing beautiful data, the BBC and FIFA are both determined to make 'fetch' happen, offering 3D versions of World Cup matches. This follows shortly after La Liga turned some Real Madrid highlights into EA FC. In the latter two cases, video game engines are at-hand to render pose data (the BBC are working with immersiv).
I still don't really understand what these are for, other than trying to work out how to make the tech work as a product. I get it, it feels like there's something to be had there, but, personally, I've got a mental block around near-real rendering of actually-real football. I'd rather the Australian Open Mii's or the NFL's Disney collab. Perhaps this is like agentic AI, where the gains will be made in the professional space before being brought over to the consumer world in better packaging.
If you've got ideas for what near-real 3D rendering of football can be used for, let me know. Also let me know if you've got ideas/opinions about what I should do with my data-and-game-model AI assistant project that I spoke about a couple of weeks ago.
Back in the real world, The Cutback published a report on next steps for women's football - 'From Visibility to Viability: A blueprint for the next phase of women's football' - based on interviews with execs and experts across the game. It's not strictly 'football analytics', but growth of women's sport and women in sport more generally are all good things worthy of Get Goalside's digital ink.
In a similar 'slightly tangential but Get Goalside wants to mention' vein, Thom Lawrence has moved on from Statsbomb/Hudl. One of the 'analytics Twitter' cadre, his work and attitude (curious, built on strong technical foundations, funny) has been very influential on me. I'm sure a lot of others share the same view, and we wish him well with his deserved break and with whatever comes next.
Another figure who's moved on to something new is Arsenal's former Head of Football Platforms, Danny Karbassiyoon. He's started a company, Sentrum, whose website tagline is 'The decision layer for' and then a rolling series of job titles. Alas, that activates Get Goalside's 2023 trap card 'So... everyone's a decision-maker now'.
However, the real reason this is bait is that the Sentrum content marketing is what Get Goalside would write if Get Goalside was a credible industry figure:
Two of the clubs I spoke to had actually tried [to build data infrastructure internally]. In both cases, smart people with real resources took a run at centralising their data infrastructure internally. Both eventually stopped.
The reasons were consistent. Building the initial version is possible. Maintaining it alongside actually running a football club is the part that breaks. Data providers update their APIs. New sources get added. Staff turn over and institutional knowledge walks out the door.
Speaking of staff turnover, King's League has made ripple-sized headlines this week for cutting jobs as part of a restructure, the extent and nature of which is disputed by a statement from King's League employees. Like 3D-rendering of real football, King's League has been a long-time Get Goalside case of 'this is an interesting development but there are dots I can't quite connect'.
Now, onto a competition which does seemingly know how to sustain itself (although has also pushed through expansion plans), this summer's World Cup will feature FIFA Power Rankings. This possession value model-sounding set of rankings will also be "the first major global collaboration with Aramco, FIFA's exclusive Energy Partner". Some may be surprised at this partnership, but not Get Goalside! Retro is firmly in, and this is merely a retro callback to the Opta Castrol Index of the late 2000s/early 2010s. You think solar can power these numbers? Pah!
The rankings will mean that "[p]erformance will no longer be judged by opinion alone," according to Arsène Wenger in his role as FIFA’s Chief of Global Football Development, who will presumably be expecting social media to be a quiet place during the tournament. Interesting too is the fact that "[s]cores will be made available no later than four hours after the final whistle of each game", giving an insight into FIFA's data collection and processing pipeline.
Related to FIFA's data collection processing, a paper on generating event data automatically from broadcast tracking data was released by a FIFA team last month. I've not yet read it in full, but we like this, this is a good thing.
Another paper that I've not yet read in full but am eagerly looking forward to is– [puts finger to ear] sorry, I'm being informed that Hudl's Defensive Responsibility metric does not come with a free white paper. Still, there's a video with a nice animation. According to it, the new model:
[P]rovides a better way to measure each player's defensive contribution and sees the game like an expert coach. By combining metadata from the defending team's formation with millions of historic events, the model understands that a centre-back is more likely to defend actions nearer their own goal than a midfielder, whose defending is spread across the pitch.
That's bait.
There's an interesting implication in blog though. When it comes to the section on centre-backs, players who make fewer defensive actions than the model expects aren't described as 'bad' or 'inefficient'. Instead, to take RB Leipzig's Castello Lukeba as an example, it suggests that "he relies more on his recovery pace and positional containment within Leipzig's structure than meeting and engaging with the play". Does this mean that even when adjusted for defensive responsibility we can't say 'higher number = better'? It doesn't seem unreasonable, but feels like a challenge to communicate.
'Not unreasonable but a challenge to communicate', must be a data scientist! – is the segue to the final note: recent job ads of the English Football Association. They've closed now, so I'll link to Dominic Jordan's post about them. He described them as follows:
The first role sits within the elite coaching development function [...] explicitly focused on the discovery, development and progression of English coaches. [...]
The second role is a more traditional but research-heavy data science and machine learning position. The emphasis here is on deep learning, tracking data, building novel metrics, player modelling and in-depth game models.
National football associations ramping up their data operations is fun and interesting. If you view FAs as their national teams, they have less to do than clubs, with only one major tournament a year (combining the senior men's and women's teams). But, of course, both have qualification processes, and then there are the age-group teams too. And then there's the other work of an FA, like the coach education that one of these jobs is aimed at. Their roles are both as competitors on multiple fronts, vehicles for player (and coach) development, and stewards of the national game's health.