Challenging thinking by showcasing new concepts
Each year, the days following the conclusion of the OptaPro Forum give us an opportunity to pause and reflect on how our industry is evolving and how we, as practitioners, approach the challenges we face, not only in terms of application but also how we share insights with coaches and players.
Since launching the Forum in 2014 we have often talked about how the event can be used as a barometer for where analytics sits within the wider industry.
As well as being oversubscribed for the second successive year, what is perhaps most significant was the increase in the number of delegates attending from national football associations and leagues outside of the UK, highlighting how industry stakeholders are increasingly recognising the value of data analytics to their member clubs.
2020 included some Forum firsts, with two presentations utilising a full season of tracking data courtesy of the Belgian Pro league; two presentations derived from research questions put forward by Swansea City and FC Nordsjaelland; and an applied presentation on the impact of the new goal kick rule, produced by our analysts following consultation with performance staff working at the English FA.
With analysts working at Women’s Super League clubs and data scientists attached to teams in League One also in attendance, it only further reinforces how the application of data processes is no longer a key differentiator between clubs in a single competition. The challenge now is how to utilise data more effectively than your rivals – to generate insights that are not only relevant but can be used to inform tactical decisions, coaching plans and selection, as well as key recruitment decisions.
This is one of the reasons why the Forum remains a key date in an analyst’s calendar. It remains a unique opportunity to learn about new approaches which have been pioneered with the sole objective of finding innovative answers to key questions posed within a club environment week in, week out, during the season.
Communication, communication, communication
This event has always aimed to showcase the cutting edge of football analytics and this year we continued to showcase methods that enable us to challenge traditional thinking while still focusing on how we can create data concepts with clear tactical application.
During the early years of the Forum, one of the key litmus tests for a presenter was ensuring that their project could be easily applied within a club’s analysis workflow. Whilst this continues to be an important consideration, the pace of evolution within the analytics space means that things have moved on – it is no longer just about the application of a concept, it is also about ensuring applicable outcomes are communicated to coaches and players in the most effective way.
Having the ability to speak the language of a coach has been reinforced at a number of previous Forum guest talks, notably by Luke Bornn in 2018. The introduction of the two club-led submission categories this year had a notable impact on how this challenge was approached by the respective presenters, who benefitted from access to each club’s analysis team to tailor their presentations.
In the case of Vignesh Jayanth, whose presentation focused on identifying effective strategies for FC Nordjaelland to break down an opposition low block, you could see that he had worked closely with his mentor, Joe Mulberry, to understand the club’s problem in the context of their game model. He then took the decision to label his strategies in line with the coaching staff’s existing play book, as a method of generating buy-in and demonstrating how data can help solve a key tactical challenge on the training ground.
Continuing on the theme of communication, Steven Houston, from Williams Advanced Engineering, used his guest talk to moot the idea of analysts rehearsing their meetings prior to presenting key findings to coaches. This is a practice already adopted by analysts working in Formula One and poses a question to analysts working in football – do their existing communication processes ensure the most important takeaways are relayed in the most effective way? That is something for us to all consider, irrespective of our role or industry.
Our headline talk today is from Steven Houston of @WilliamsAdvEng, who is considering what football can learn from F1 and Motorsport in relation to the application of analytics. Horsepower. pic.twitter.com/nsTfqKKr7h
— OptaPro (@OptaPro) February 5, 2020
Enhancing an analyst’s working processes
One of the overriding objectives of the Forum is to bridge the gap between the analytics community and professional analysts. As well as tactically-driven presentations, the Forum also showcases technical innovation, highlighting how existing working processes can be enhanced as well as being made more efficient.
Both of these elements were evident in a project presented by Karun Singh, which applied deep learning processes to tracking data to identify recurring match scenarios across a full season, potentially accelerating a match analyst’s daily workflow by removing the need for manually labelling video and searching for similar passages of play.
By leveraging ‘self-supervised’ labels, he outlined how his system could free up several hours of an analyst’s working day through potentially eradicating the need to review huge amounts of video to unearth key team trends.
Karun’s approach to identifying efficiencies reflects a wider trend in how practitioners working outside of professional sport now possess a better understanding of what an analysis process looks like. This means that the Forum can offer people a platform to use their core skills, which in Karun’s case was software engineering within applied AI, to showcase innovation and provide us with a small glimpse into how technology is going to revolutionise our industry in the decade to come.
Thinking about how we approach technical scouting
In addition to performance analysis, the Forum also provides an opportunity for recruitment analysts to escape the frenetic day-to-day environment of monitoring and assessing players from key scouting markets and reflect on their processes for assessing players to fit short, medium and long-term objectives.
One idea presented this year came from Stats Perform’s Paul Power and Will Gürpınar-Morgan, who introduced the concept of ‘Role Discovery’, which moves on from the idea of assessing players based on position – and instead focusses on profiling players based on their role in a team.
After highlighting the limitations of profiling players based solely on position, the pair demonstrated that, by applying unsupervised learning processes, teams can potentially start to analyse players based on a role within a wider team style. This means that recruitment analysts could potentially group players together based on the key tendencies and characteristics required to fulfil each role within their own team’s game model.
The role of the technical scout is still relatively new, having first emerged in the industry around 15 years ago, but it is probably the one role within clubs which is being redefined the most as technology, data and video evolves. However, whilst the tools at their disposal have changed significantly, how much has the actual assessment criteria of players actually changed?
By considering if scouting by role is more effective than scouting by position, we can start asking objective questions about our existing approaches and whether we are generating the optimum player analysis to make sound recruitment decisions.
Before we break for lunch, @counterattack9 and @WillTGM are demonstrating how unsupervised learning can automatically learn player roles across the pitch – creating data driven summaries. Intelligence. pic.twitter.com/IZPSMoPzqZ
— OptaPro (@OptaPro) February 5, 2020
Insights into the future
We were really fortunate to have Steven Houston deliver our guest talk this year. Steven was able to draw on his previous experiences of working in football to provide insights into what the sport can learn from analytical processes pioneered in F1.
One valuable takeaway was a reminder for us as analysts to ‘squeeze the lemon’ – to maximise what we have from the data that is already available to us. A good example of this was David Quartey’s presentation into how throw-in strategies can result in goalscoring opportunities, which reminded us of how throw-ins remain a relatively underutilised set piece opportunity in the sport.
At the other end of the scale, Steven was also able to give us a brief glimpse into how data-driven insights, harnessed through AI technology, are driving decision-making in F1.
It’s pretty clear that like most industries, AI and Machine Learning will dominate our thinking in this decade as we look to utilise the power of this rapidly advancing technology to create new data sets, process it faster, and identify patterns and trends that will impact decision making in every area of sport.
In F1, whether it is using body pose technology to identify marginal gains at pit stops, or simulating performance of different drivers in the same car, there are obvious parallels of how this technology could be applied within performance analysis in football.
Similarly, recruitment departments in football often talk about projecting a player’s future performance, by simulating how they would perform in different match scenarios. Given the advances in technology it is only a matter of time before this type of analysis becomes a reality.
Looking back on the Forum as a whole, my key takeaway would be that we have a vast depth of talent in this industry that are challenging us practitioners to never stand still. On behalf of everyone at Stats Perform, I would like to congratulate all the presenters and contributors for inspiring more focused innovation in 2020!
Video of all 2020 OptaPro Forum presentations, together with poster write-ups, will be available from the Stats Perform website later this spring.