Every elite sports team in the world employs analysts. Most employ a department of them. The numbers you see on a broadcast — expected goals in football, true shooting percentage in basketball, sprint distance in rugby — are the consumer-facing tip of an iceberg. Below the waterline are private models that influence transfers, training plans, in-game decisions, and even how players are paid. This is what those models actually measure, and why the public versions of these stats are usually less interesting than the private ones.
The three categories of sports data
First: event data. Every pass, shot, tackle, dunk, or pitch is logged with a location, a time stamp, and an outcome. Companies like Opta (football), Statcast (baseball), and Second Spectrum (basketball) collect this in real time. Event data underpins every traditional public stat — goals, assists, completion percentage. Second: tracking data. Player and ball positions are recorded 25-30 times per second by stadium cameras or wearable GPS units. Tracking data is what enables expected goals, advanced spacing analyses, and "ghost defender" simulations. Third: physiological data. Heart rate, sprint distance, jump height, sleep quality, hydration, biomechanical load on joints. Wearables like Catapult and WHOOP dominate this layer, and the data is largely private to the team.
Expected goals (xG), in plain language
xG is the most cited modern football stat. The model takes every shot ever recorded — millions of them — and learns the probability that a shot from a specific location, with a specific body part, after a specific type of build-up, becomes a goal. A penalty is around 0.78 xG. A long-range header is around 0.04. A team that creates 2.5 xG of chances and scores 1 goal had a bad finishing day; a team that creates 0.6 xG and scores 2 goals had a lucky one. Over a season the gap between actual goals and xG closes for almost every team — meaning xG is one of the strongest predictors of future scoring.
Why basketball runs on possessions
NBA analytics rebuilt itself around the per-possession framework. Points scored per offensive possession (offensive rating) and points allowed per defensive possession (defensive rating) are the metrics that matter, not raw points scored. Teams that play fast inflate their box scores without necessarily being better. The corollary is the rise of the three-pointer: a 36 per cent three-point shooter delivers 1.08 points per shot, beating a 50 per cent two-point shooter at 1.00. Once teams realised this on a per-possession basis, three-point attempts roughly doubled across the league inside a decade.
Tracking data and the "ghost defender"
With 25 frames per second of every player’s position, you can ask counterfactual questions. "What would the defensive shape have been if a different defender had pressed?" "Did the goalkeeper come off their line at the optimal moment?" These models — sometimes called ghost defenders — let coaching staff evaluate decisions independent of outcomes. A wing-back who made the right decision but lost the ball is rated higher than one who made the wrong decision but got bailed out by a teammate.
Load management and injury prediction
Wearable physiological data is increasingly used to predict and prevent injury. The strongest signals are sudden acute spikes in workload (one match where a player runs 20 per cent more than their rolling 28-day average) and chronic under-recovery (sleep quality declining for two weeks). Modern training staff manage rest periods, substitution timing, and selection partly off these signals. The modern criticism that "the schedule is too crowded" is partly a statement that the load-vs-recovery balance can no longer be managed within the calendar.
The limits of analytics
Data is excellent at decisions that repeat thousands of times — set-piece routines, shot selection, defensive shapes. It is weakest at decisions that are rare — one-off knockout matches, in-tournament momentum, locker-room dynamics. The best teams treat analytics as a way to improve baseline decisions, not as a replacement for the experience and intuition of coaches who have seen the rare situations many times. The analytics revolution did not replace the head coach. It made the head coach a better-informed decision-maker.