The US Open Is a Data Centre With a Stadium Attached

For two weeks at the end of August the tennis world watched Flushing Meadows, and it watched it through more instruments than any sporting event in history. Every serve was clocked. Every rally was counted. Every ball that landed within a hair of a line was judged not by a human eye but by a bank of cameras running at hundreds of frames a second.

What is easy to miss, while you are watching the tennis, is how completely this has changed what a professional player knows about their own game — and how recently that change happened.

The US Open Is Now a Data Centre With a Stadium Attached

Consider what a player and their team walk away with after a single match in New York.

They have the speed and placement of every serve, along with the bounce point measured in centimetres. They have the contact position of every groundstroke. They have rally-length distributions, shot-tolerance numbers, court-position heat maps, and distance covered broken down by point outcome. They have the direction of every ball their opponent hit off both wings, and the pattern that opponent falls back on at 30-30.

Twenty years ago, essentially none of this existed in a form a coach could use between matches. A team would rewatch a broadcast tape, count things by hand, and argue about what they saw. The analysis was real, but it was slow, partial and heavily coloured by whoever was doing the counting.

Electronic line calling is the part of this the public sees, because it is the part that changes a decision in front of them. The US Open went to fully automated calling in 2020 and the rest of the tour has steadily followed. But line calling is the least interesting thing the cameras do. The same tracking that decides whether a ball caught the line is producing a complete three-dimensional record of where every ball and every player was, several hundred times a second, for the entire match.

That record is the real product. Line calls are a by-product.

The Shift: From Watching to Measuring

The important change is not that there is more information. It is that the information is measured rather than remembered.

Human memory of a tennis match is famously unreliable, and it is unreliable in a specific direction: we remember the dramatic points and forget the ordinary ones. Ask a player how they served in a match they lost and they will describe the double fault at 4-5. They will not describe the twelve first serves that landed short in the box and gave the returner a free swing, because none of those moments felt like anything at the time.

The data does not have that bias. It counts the boring points exactly as carefully as the dramatic ones, and it is usually in the boring points that matches are actually decided.

This is why professional teams have reorganised around it. The modern coaching box does not just watch — it reads. The question has moved from "how did that look?" to "what does it say?"

The Part That Never Reached You

Here is the frustrating bit. Almost none of this trickled down.

The tracking systems used at Grand Slams cost hundreds of thousands of pounds to install and require a dedicated camera rig in a permanent stadium. They cannot be taken to a club court on a Tuesday evening. The gap between what a top-100 player knows about their forehand and what a good club player knows about theirs is not a gap in intelligence or in coaching quality. It is a gap in measurement.

A club player will typically hear the same handful of observations for years: get your racket back earlier, bend your knees, follow through. All useful. All completely unquantified. Nobody can tell you whether your contact point is ten centimetres further forward than it was in March, because nobody measured it in March.

And these are not small things. On a serve, contact lasts roughly four thousandths of a second and happens above and slightly in front of your head — the one place in the entire motion you can never look. You can feel that a serve was poor. You cannot see why.

What We Built, and Why

OnCourtAI exists to close that specific gap. Not to replicate a stadium camera rig, but to take the part of it that actually helps a player improve and put it in their pocket.

You film a few shots on a phone. The system finds the individual strokes inside the video, estimates your body position frame by frame, and scores the components of the technique — contact point, preparation, kinetic chain, balance, follow-through, wrist lag and the rest — against reference patterns built with coaches. You get numbers, and you get them per component rather than as a single verdict.

The OnCourtAI app currently holds analyses of 15,249 individual shots from 792 players across 89 countries, built up over the past year. That is not Grand Slam infrastructure. It is a phone, a tripod and a spare ten minutes, and for the purpose of finding out what your forehand is actually doing, it turns out that is enough.

Why Component Scores Matter More Than a Single Number

The single most useful thing we took from watching how professional analysis works is this: never trust an aggregate.

An overall score can hide two large changes cancelling each other out. We have seen exactly this in our own data — players who substantially improved their serve contact position while their hip and shoulder rotation went backwards by a similar amount. They had started reaching with the arm instead of driving with the legs. Their overall serve score barely moved, which would have told them nothing was happening at precisely the moment a lot was happening, and not all of it good.

Broken into components, the same data tells you exactly what to do: keep the new contact point, get the legs back into it.

This is how a professional team reads a match, and there is no technical reason a club player should not read their own game the same way.

What the Technology Genuinely Cannot Do

It is worth being straight about the limits, because the marketing around sports technology is not always careful.

Measurement tells you what is happening. It does not always tell you why, and it never plays the shot for you. In our data, faults that can be corrected once they are pointed out — where you meet the ball, where the racket finishes — respond very well to video feedback. Faults that require rebuilding a movement pattern respond much more slowly. Spin generation is the clearest example: it is the weakest component in the amateur game across our whole dataset, averaging just 35 out of 100, and it is stubborn. That one still needs a coach, a basket of balls and a lot of repetition.

Technology has not replaced coaching at the top of the game either. It has changed what coaches spend their time on. Less time counting, more time deciding.

The Real Legacy of a Tournament Like the US Open

Every few years the sport absorbs a technology that was once exotic and makes it ordinary. Hawk-Eye was a novelty, then a challenge system, then simply how line calls are made. Ball tracking was a broadcast graphic, then a coaching tool, then the backbone of the sport's entire data layer.

The next step in that sequence is the one that matters most to everyone reading this, and it has already started: the same class of analysis, without the stadium. Not a broadcast graphic you watch, but a measurement of your stroke that tells you something specific and actionable about what to do at practice on Thursday.

The players in New York have had this for years. The interesting development is not that they have it. It is that you can now have it too.

Where to Start

If you want to see your own numbers, the fastest route is a single video of ten to fifteen shots from side-on, filmed at chest height. Our guide on filming for AI analysis covers the setup in more detail, and it makes a genuine difference to the accuracy of what comes back.

Film the stroke you are least confident about. That is almost always where the biggest number is hiding.