
I went to a conference on AI and technology at a secondary school and came across a talk I wasn’t expecting: one by the then chief technology officer of Córdoba Club de Fútbol. My team.
He had come to show us what they were developing. An internal platform that used AI for scouting and for analysing player statistics, something like an assistant for the coach.
Anything to do with data and AI is my weak spot, so I found the project fascinating. And as I listened, two things were happening at once: I liked his idea more and more, and I kept thinking of a thousand ways to improve it.
When the talk ended I rushed home. I wanted to start building it right away.
That is how footactik was born. It answers the questions a coaching staff asks every week: who should play, who is close to their limit, who could be signed and how the opponent plays.
For the next match it suggests a starting eleven and an approach, with the probability of winning, drawing or losing. It studies the opponent’s style and compares it with yours. It tracks the minutes each player has accumulated over the last four weeks and warns when someone is heading for an overload. And it estimates what a player will be worth in a year, so you know who to renew before their value rises.
The scouting doesn’t search by position but by style of play. You ask for players similar to your fullback and it may return a midfielder, because they play the same way even if their profile says otherwise. It works both for finding talents and for covering the positions where the team is weak, and it even builds a shadow team: two market alternatives for each position in the eleven, within a budget.
And there is a tab that hardly anyone has, the honest one. It compares what each model predicted with what actually happened, and with what someone using no model at all would have got right. If the model doesn’t beat that by enough, it says so. It is the lesson from my thesis applied to football: not trusting a result until I have put it to the test.
The biggest obstacle wasn’t the code, it was the data. I needed it to be free and to be good, because everything rests on that. And the data on a second division player, or on one who has just come up from lower leagues, is nowhere near the quality of what you find for the top flight.
Even so, I think it turned out quite attractive. And the whole thing runs on a laptop, without paying for any service.
I showed it to the chief technology officer. I explained that his talk had motivated me enormously and that I had built it simply in case it helped him have better ideas for his product. He thanked me and said he would take a look.
I haven’t heard anything since. I still think there was something big there.
Technical summary
A local football analysis platform for Córdoba CF. It collects data from Sofascore, Transfermarkt and FBref and uses five machine learning models for lineup selection, injury risk, scouting, player valuation and opponent analysis. Scouting uses autoencoder embeddings to compare players. The platform runs locally without paid services.