The Hidden Calculus of the Beautiful Game: Systems Thinking in Soccer Analytics

In this conversation, soccer analytics veterans Mike Treacy and Joris Bekkers explain that the fluidity of soccer is not an impenetrable barrier to data, but a complex distribution problem that can be broken down. By moving beyond simple possession metrics to skeletal tracking and game-state modeling, teams are shifting from subjective scouting to a rigorous, portfolio-management approach. This transition reveals a simple truth: the most valuable insights are not found in the highlight reel, but in the bad data of skewed game states and the quiet movements that prevent goals before they happen. For leaders in any high-variance field, this shift offers a clear advantage: the ability to separate raw performance from outcome-based noise, allowing for decisions that optimize for long-term durability over immediate, flashy results.

The Illusion of Randomness: Why Soccer is Finally Solvable

For years, the conventional wisdom held that soccer was too chaotic for the Moneyball treatment that changed baseball. Where baseball is a series of discrete, repeatable events, soccer is fluid and continuous. However, Treacy and Bekkers argue that this was merely a limitation of compute and data density. The shift from simple on-ball event data to high-frequency positional tracking, and now, full-body skeletal pose data, has allowed analysts to break the game into manageable, modelable chunks.

The game is just more complex. So it needed more compute and needed more data, needed more knowledge.

-- Joris Bekkers

This reveals a systemic shift: when you have enough data, art and intuition are often revealed to be high-speed calculations the human eye simply has not been trained to track. By mapping player movements at 25 frames per second, teams are no longer guessing who has grit. They are quantifying how a player closes passing lanes and prevents high-probability scoring opportunities, which are events that, by definition, never happen on the scoreboard.

The Portfolio Management Trap: Why MLS Differs from Europe

Treacy highlights a distinction between European leagues and the MLS: the third vector of portfolio management. In Europe, the focus is often on recruitment and first-team performance. In the MLS, the rigid salary cap and roster rules force teams to treat players like assets in a financial portfolio.

A player value is not absolute; it is relative to their cap charge. A player who is a bargain at a senior minimum salary might be a bad investment if they occupy a Designated Player slot. This forces a systemic discipline: teams are not just scouting talent; they are balancing a cap-constrained portfolio where the objective is to maximize points per dollar. This mirrors volatility trading, where the goal is to manage exposure under strict regulatory and budgetary constraints.

The Rich Fountain of Bad Data

A non-obvious insight from the conversation is the danger of raw outcome data. Treacy points to a match where a team went down to nine men and played an absurdly high defensive line to compensate. The resulting goals were not representative of the players actual skill or the team typical strategy; they were artifacts of a specific, skewed game state.

It was a rich fountain of bad data.

-- Mike Treacy

Most teams, when evaluating a player at the end of a season, look at per-90-minute outputs. If those outputs are polluted by bad data from anomalous game states, the model breaks. The competitive advantage goes to the teams that have the discipline to censor these outliers, effectively cleaning their dataset to understand what the team looks like at an equal game state. This is the difference between reacting to noise and understanding the underlying signal.

The Translation Gap: Why Models Do Not Coach

Even with perfect models, a massive bottleneck remains: the translation of machine-learning output into actionable coaching advice. A neural network cannot explain its reasoning to a manager. The most successful teams have bridged this by using data to identify specific video clips, a bottom-up approach where the data serves as a search engine for the human eye. The insight here is that data does not provide answers; it creates questions that require an expert analyst to translate into the language of the pitch.

Key Action Items

• Audit Your Bad Data (Immediate): Identify the anomalous events in your operations (the nine-men-down scenarios) that skew your KPIs. Determine if these data points are helping or hurting your predictive models.
• Adopt a Portfolio Lens (Next Quarter): If you manage resources, stop evaluating assets or talent in isolation. Map them against their cap charge or cost-to-benefit ratio to see if you are over-allocating capital to low-leverage positions.
• Bridge the Translation Gap (Next 6-12 Months): Stop presenting raw model outputs to decision-makers. Build a workflow that uses model signals to curate specific, actionable examples (like video clips) that visualize the why behind the data.
• Prioritize Preventative Metrics (12-18 Months): Shift focus from tracking successful outcomes (goals) to tracking the prevention of failure (closing lanes). This requires effortful, non-obvious tracking, but creates a lasting competitive moat.
• Control the Controllables (Ongoing): Accept that human error and randomness are fixed parameters. Build systems that are robust enough to handle the noise of the game without requiring constant, reactive adjustments.