AINEWS
The Professor Who Says To Boot the Ball Out of Bounds on Purpose
A decade of machine learning analysis from KU Leuven's Sports Analytics Lab is quietly rewriting how top soccer clubs think about possession, starting with the opening kickoff.
A decade of machine learning analysis from KU Leuven's Sports Analytics Lab is quietly rewriting how top soccer clubs think about possession, starting with the opening kickoff.
On the opening kickoff of a soccer match, some teams deliberately send the ball the length of the pitch and out of bounds on the opponent's end, surrendering possession before a single pass is played. Conventional wisdom calls it a waste. A decade of machine-learning analysis from a Belgian university lab now treats it as a setup.
Jesse Davis, a professor of computer science at KU Leuven and head of its Sports Analytics Lab, has spent more than ten years applying machine-learning models to soccer, with adjacent work in basketball, volleyball, and field hockey, according to MIT Technology Review's profile of the lab. The lab's work has shifted how professional clubs evaluate rosters, assess in-game tactics, and detect recurring patterns, the same report notes, and Hugo Rios-Neto, the data recruitment lead for Royal Sporting Club Anderlecht, calls it "the most influential sports analytics lab in soccer."
The clearest demonstration of the lab's method sits in a finding that would have sounded heretical a generation ago. A 2024 paper colloquially known as "Boot it" showed that intentionally launching the opening kickoff deep into the opponent's end can be a prime scoring setup, even though it looks like giving the ball away. The study trained tree-ensemble models on roughly 1.4 million passes and around 60,000 throw-ins, including 2022 World Cup data, and found that when the ball reaches the middle third of the pitch within ten actions of a goal, the sequence is more dangerous than equivalent possessions that start with a controlled build-up. The mechanism is positional, not mystical: a long ball pins defenders near their own goal, compresses the field, and forces throw-ins or goal kicks in dangerous zones where the pressing team can attack the second ball.
That kind of counterintuitive result is the lab's signature product. Davis and his collaborators have built tree-ensemble models that ingest event data and tracking data to score actions, possessions, and tactical choices, then run counterfactual analyses to ask whether a different decision would have produced a better outcome. The approach treats soccer less as a narrative of stars and moments and more as a system of repeatable decisions under uncertainty. Clubs have responded by rebuilding their recruitment pipelines, their set-piece routines, and their in-game adjustment protocols around what the models flag.
The work is not without limits. The data that flows into these models comes from leagues and vendors that sell it at prices smaller clubs cannot afford, which means the lab's strongest findings tend to reflect the leagues that can pay for tracking. There is also a real gap between what the research shows and what actually shows up on the pitch. A paper can identify a pattern in a million passes and still never change a manager's behavior on a Saturday. Rios-Neto's "most influential" label, attributed to him by MIT Technology Review, is a practitioner's verdict rather than an audited scorecard, and it is worth keeping in that lane.
What to watch next is whether the "Boot it" logic scales beyond the opening kickoff. The same dataset covers tens of thousands of throw-ins and tens of thousands of goal kicks, and the lab's modeling framework can be turned to any set piece that resets possession. If second-ball value holds up across leagues and seasons, expect more clubs to rehearse the "give it away on purpose" play until defenders stop treating it as a mistake.