Google DeepMind is bringing its soccer AI to the real pitch. On June 11, the Institute announced that Brazilian club Palmeiras was the first team to meaningfully build TacticAI. DeepMind says the system can simulate on-field scenarios and predict open play dynamics up to eight seconds in advance. The deal, announced at the Google for Brasil event, also involves the Brazilian Federation of Futebol (CBF).
TacticAI is not new. DeepMind first detailed this issue in March 2024 in a Nature Communications paper developed in collaboration with Liverpool FC through a multi-year research partnership. What is new is that top clubs are moving their systems from research demos to live open play analysis.
How TacticAI reads games
The system relies on geometric deep learning and graph neural networks. Treat each player as a node with characteristics such as position, velocity, and height, and model how those nodes interact. TacticAI respects pitch symmetry, so it can generalize from a relatively small number of training examples. This is a real constraint in elite football where labeled tactical data is lacking.
advertisement
In its original form, TacticAI focused on corner kicks. It can predict which players are most likely to receive the ball, estimate shot chances, and generate alternative setups that coaches might try. The Palmeiras partnership extends that approach to broader open play, using broadcast-style visual data rather than specialized sensors.
Why the coach took it seriously
The headline results from Liverpool’s survey still play an important role. In blind comparisons, the club’s human experts preferred the TacticAI-suggested setup over the actual match configuration 90% of the time. This number is what separates TacticAI from novelty. Coaches who do this for a living almost always rate the machine’s tactics as good or better than their own.
There are limitations worth keeping in mind. The 8-second prediction window and live open play ambitions exceed peer-reviewed corner kick accomplishments, so they remain notable claims rather than published results. Still, it would be a meaningful vote of confidence for a club competing for major titles in Brazil and South America to bet on this tool.
This also applies to a broader pattern of DeepMind pushing its models into everyday products and partnerships, much like Google’s broader AI rollout that recently brought Gemini to Google TV settings. It will take a full season to determine whether TacticAI changes on-field results, but the experiment is now a reality.
