“That’s him!”: how TwelveGPT captures players in words
We have solved a problem which no-one else has solved in the football data world.
We have found a way to write reliable reports on players, which summarises their skillsets in a language which scouts understand. When I presented TwelveGPT to clubs in Brazil at the end of the month, the feedback I got time after time was the same: I would show an analyst at the club a player on TwelveGPT, they would look at the text and say,“yes, exactly, that’s him!”
To get an idea what I mean by “yes, that’s him!” compare ChatGPT description of Erling Halland and that from TwelveGPT….
Comparing how TwelveGPT and ChatGPT writes about Erling Haaland.
ChatGPT does a good job at first, it covers his physical presence and even gets his height correct, but as we read on, we start to see how ChatGPT starts to make things up. It claims Haaland excels at everything: dribbling, providing teammates, as well as scoring goals. He doesn't! This is clearly captured in the second text from Twelve. It is also consistent with the data: he specialises in a small number of skills: the most important of which is poaching and does not get heavily involved in games. As we see in more detail below.
There are lots of components which make TwelveGPT work so well, but I want to emphasise one of them here: the texts that come from it are based on Twelve’s statistical models of football, and not the other way round. We have, over the last few years, built up experience working with leading clubs on how to use metrics to scout players. We have then used that experience to build, what we call qualities (like involvement, poaching etc. ) from event data. So when TwelveGPT says something, it is based on our best practice. ChatGPT works the other way round, it trawls the Internet for data, and then tries to summarise that by building its own model of that data. This is why ChatGPT makes things up! It doesn’t understand football or data.
The power of TwelveGPT does not lie in summarising Haaland. Most people in football know how Haaland plays and everyone in the world knows he is good. It lies, as one footballing director we work with keeps telling me, “in finding hidden gems.” It is a tool which is invaluable for clubs who maybe can’t afford an expensive data scientist, but do want to find the best young striking talent in, for example, Slovenia. Like Nejc Gradišar, “a youthful talent in the 1. SNL, is a striker who is excellent at threatening the goal, making runs, poaching chances, and winning aerial duels.”
Larger clubs can also use TwelveGPT, as a second opinion, or as help for data analysts in writing reports. As one scout at a Championship club told us after he trialed TwelveGPT: “I can see the club chairman loving this”. Suddenly everyone can talk numbers.
The open trial period for Twelve GPT has now finished, as we onboard our early adopters. But if you are interested in utilising Twelve GPT within your club then please contact us using the button below and we can give you temporary access to try it out.
