Sports Analytics Weekly by kubeia.io - 12/2026

Does changing manager actually improve performance, or do teams often recover anyway?

That is what this paper set out to examine across fifteen seasons of Premier League football, using a methodology that is more demanding than the usual before-and-after comparison. Rather than simply measuring points gained after a dismissal, the authors compared each managerial change to a carefully matched counterfactual: a team in a similar competitive situation, with a similar recent performance trajectory, but without changing coach.

The paper starts from one of the most persistent problems in football analytics: separating what belongs to the player from what belongs to the environment around him. A player’s observable production, whether measured through pass completion, xG contribution, or other event-based outputs, is always entangled with tactical context, teammate quality, opposition level, and game state. As the paper puts it, “a player’s observable metrics… are not a pure function of their individual ability”. This immediately creates a practical difficulty for recruitment and projection. A striker scoring regularly inside a dominant possession structure may not reproduce the same output elsewhere, while a midfielder whose numbers look modest could in fact be constrained by system effects.

Let's build a Football AI system to dig deeper into match stats! We'll use computer vision and machine learning to track players, determine which team is which, and even calculate stuff like ball possession and speed. This tutorial is perfect if you want to get hands-on with sports analytics and see how AI can take your football analysis to the next level.

All data modalities and tasks are occupied by foundation models.

All? No! One small modality still holds out against them: tabular data.

But this resistance is crumbling.

TFMs are a fundamental shift, not just a performance trade-off

TabPFN opened the era of foundation models for tabular data. For small and mid-sized data, tabular foundation models now outperform other ML algorithms (see TabArena).