Google DeepMind has deployed a football prediction model that can forecast player movement up to eight seconds before plays develop.
TacticAI uses geometric deep learning to model how players move across the pitch. The technique treats spatial relationships as graphs rather than flat pixel grids, letting the system reason about positioning in ways that standard computer vision struggles with. It ingests broadcast-style footage rather than specialist tracking cameras, which matters for clubs without expensive sensor infrastructure. São Paulo club Palmeiras is the first team running it for live open-play analysis.
Eight seconds is a real window in football. That is long enough for a midfielder to shift position, a winger to adjust a run, or a coach to call a press trigger. Most analytics tools work from post-match data; a system delivering recommendations mid-play is a meaningfully different product. The broadcast-vision approach also widens the addressable market: clubs that cannot afford expensive positional tracking licenses can still plug in.
Whether those predictions translate to actual wins is the question that comes after the press release.