“We want to push the boundaries of AI-based tactical football analysis”

Manuel Stein is CEO and founder of Subsequent. The company is a provider of real-time skeleton data extraction and tracking and innovative motion analysis based on simple video recordings. In 2023, Subsequent has been awarded as Germany’s “Digital Start-up of the Year” by the Federal Ministry of Economics and Climate Protection. In this interview, he talks about his journey towards a more context-oriented data analysis and visualisation in football.

Artificial Intelligence is on everybody’s lips. How do you use AI at Subsequent specifically?

At Subsequent, we have two pillars. The one is real-time data acquisition from simple video recordings which you can take with your smartphone, for instance. The other pillar is the motion analysis based on this data. Machine learning and neural networks are our tools to gain meaningful insights from data in a highly accurate, yet cost-effective approach. We can use this for different sport disciplines, but also in markets such as home fitness and neurological rehabilitation or even automotive and retail. However, we started our journey with the AI-based tactical analysis of football matches. This is something particularly challenging to automatically compute.

Manuel holds a football in his hand and shoots into the camera.

Why is that?

Football is a complex team sport. The behaviour of players is highly interdependent. Many things happen in situations where the ball is not involved. And there are different coaching philosophies which focus on different tactical aspects. However, I believe that insights which provide a real added value can come from automated context-oriented data analysis and visualisation.

What does this enable?

There is highly relevant contextual information which can be extracted from video data of football matches. Our approach enables us not only to track the ball and players in real-time, but also to automatically analyse interesting aspects such as free spaces, interaction spaces, dominant regions, or passing options. We want to push the boundaries of what is possible with AI-based tactical football analysis. Today, this is still mostly done by manually tagging and annotating match recordings. That is very time-consuming and expensive. Therefore, in order to make our automated analysis user-friendly and verifiable, we have also developed a technique for automatic in-video visualisation. This is not only useful for coaches and analysts to speed-up and enrich their tactical analysis of football matches. It can also be used for player scouting and for TV broadcasters to provide more meaningful tactical information for football fans.

You obtained your PhD from the University of Konstanz and worked in academia. Why did you decide to found Subsequent?

At university, I had already started to work on novel methods for the analysis of motion data. Yet, the availability of data was a huge obstacle. Sometimes data were not available at all. And sometimes the conditions to use them imposed by data owners were not acceptable for a university. Thus, my vision was to find own solutions. Our approach at Subsequent works with existing hardware and is easy-to-use and affordable for end-users. That way, we want to democratise the acquisition and analysis of motion data.

You are based in the city of Constance. What do you like most about living and working there?

I really love the beautiful region around Lake Constance. Professionally, the proximity to the University of Konstanz is very important for us. The university is one of Germany’s Universities of Excellence and it is a worldwide leader in the area of visual movement analytics. As a company, we benefit a lot from our close cooperation with the excellent researchers there.