Realtime sports analytics before it was fashionable

Realtime statistics during a match taught me something I still use in software design: data has to arrive while it can still change the next decision.

Sports analytics is easy to misunderstand. After the game, every number feels clean. During the game, the data is noisy, incomplete, emotional, and time-constrained. Coaches and players do not need a perfect report. They need a better decision before the next possession.

That makes the problem much closer to operations software than most people expect.

Decision Value of a Statistic

$$V(stat, t) = Impact(stat) \cdot P(action \mid stat) \cdot e^{-\lambda t}$$

A statistic with high retrospective value may have low live value if it arrives too late or cannot suggest an action. A simple observation delivered at the right moment can matter more than a complex metric discovered after the match.

Game simulation gives the same lesson in another form. A match is a state process. The current score, time, fouls, possession, player fatigue, defensive shape, and shot quality all change expected outcomes.

Possession Value

$$EV(s) = \sum_a P(a \mid s) \cdot Reward(a, s)$$

The exact model can be simple or sophisticated. The useful part is the discipline: define the state, observe transitions, estimate value, and feed the conclusion back into human judgment quickly enough to matter.

Video analysis adds a spatial layer. The same event has different meaning depending on position, spacing, timing, and defensive pressure. That is not far from route intelligence or operational monitoring. Geometry and time turn events into context.

The human side is just as important. During competition, nobody wants a lecture from a model. The information has to be small, credible, and actionable. It must respect the tempo of the environment.

That early work made me appreciate a pattern I later saw in finance, delivery, telemetry, aviation, and city systems. Realtime analytics is not about producing more numbers. It is about selecting the few signals that can change the next decision.