From Moneyball to Tracking Data: How Sports Analytics Actually Reaches the Field

The book came out in 2003 and the argument has been settled for a while. Nobody in professional sport is debating whether to use data. The debate now is quieter and more practical: why so much of the data collected never changes a single decision.

Collection stopped being the hard part

Every major American league runs continuous tracking. MLB’s Statcast captures pitch and batted-ball movement from camera systems accurate to a fraction of an inch, with the automated ball-strike system measuring to roughly one-sixth of an inch. The NFL puts RFID tags in shoulder pads and feeds Next Gen Stats. NBA arenas run optical tracking on every player and the ball.

A single game now produces more structured measurement than an entire season did twenty years ago. The data exists, it is clean enough, and it arrives fast. None of that is the bottleneck.

The bottleneck is translation

A model that predicts soft tissue injury risk from workload data is only useful if a strength coach changes a session because of it. That requires the output to arrive before the session is planned, in a format the coach reads without a data scientist standing next to him, with a confidence level he can weigh against what he saw with his own eyes on Thursday.

Most analytics departments produce beautiful work that fails one of those three tests. The number is right and lands on Monday for a session that happened Saturday. Or it lands on time wrapped in a dashboard with forty filters that nobody opens twice. Or it arrives with no sense of

how certain it is, so the coach treats a model that is right 58 percent of the time the same as one that is right 92 percent of the time.

The unglamorous fix is a data pipeline

Behind every analytics department that actually influences decisions sits infrastructure nobody writes articles about. Tracking feeds, medical records, GPS from training, video tags and scouting notes arrive in different formats, on different schedules, with players identified differently in each one. Somebody has to reconcile all of it and keep it reconciled through transfers, injuries, loan spells and name changes.

Teams routinely underinvest here because the work is invisible. They hire three analysts and no data engineer, then wonder why every question takes a fortnight to answer. The ratio that works in practice looks more like one engineer for every two analysts, and the first six months are spent on plumbing rather than models. Organisations without that capability in-house usually bring in AI and data engineering support for the pipeline layer and keep the sport-specific modelling with their own staff, which is the right division of labour.

Expansion franchises meet this in sharper form. The Toronto Tempo and Portland Fire both entered the WNBA in 2026 with no historical data of their own, no agreed file formats and no institutional memory of which measurements their coaching staff would trust. Building the pipeline first and the models second is not a preference for a new club. It is the only order that works when season one starts in May.

What good output looks like

The most useful analytics output tends to be the least impressive to look at. A single page reaches coaching staff the morning after a match. It lists six players, a one-line reason for each, and a recommended adjustment. No charts. A link at the bottom for anyone who wants the working.

That page was the output of a system processing millions of rows. The restraint was the point. Coaching staff have maybe four minutes of genuine attention for analytics on a Monday morning, and a product that respects that limit gets used.

Where human judgment still wins

Tracking data describes what happened with precision and says almost nothing about why. It records that a midfielder covered less ground in the final twenty minutes. It cannot tell you whether that was fatigue, tactical instruction, a niggling calf, or a player protecting himself after a hard tackle in the 62nd minute.

This is why the “replace the scout” framing was always wrong. The useful version is a scout who walks into a meeting having already seen which thirty players out of eight hundred are worth watching. The model narrows. The human decides. Coverage of how clubs actually combine the two, such as analytics reporting on American sports, tends to be more honest about this than conference talks, which always present the model as the hero.

A reasonable test for any analytics project

Before commissioning the work, write down the decision it will change, the person who makes that decision, and the moment in the week they make it. If you cannot fill in all three, you are building a report, not a tool.

That test would kill a meaningful share of analytics projects in professional sport, and most of them deserve it. The departments producing real value are not the ones with the most sophisticated models. They are the ones whose output lands in front of the right person at the right hour in a form that takes ninety seconds to read