The Analytics Room's Favourite Child: When Serve Data Becomes a Blindfold

In the commentary booth, there is a line I tell myself every time I go on air: the data does not lie, but it tells only half the story, then falls silent at exactly the most dangerous moment. That night in Indian Wells, I sat behind the glass looking down at the main court. A young player was serving in the eleventh game, the score 5-5, 30-30. The screen showed numbers as pretty as a painting: 71% first serves in, 83% of first-serve points won, 12 aces. The host sitting beside me nodded: he is serving beautifully.

But I was not looking at the screen. I was looking at the returner's feet. Three games earlier, he was still standing behind the baseline, waiting for the ball to bounce before reacting. In this game, he had stepped in, planting one foot inside the court the moment his opponent tossed the ball up. He was reading the spin. He was starting to guess right. And the on-screen graphic — loyal to the past — was still praising a server who had, in reality, been decoded two games ago. He lost that game. He lost the match. But in the post-match report, his first-serve numbers still looked good. Very good.

That is why I began to doubt the very numbers I once worshipped.

The tennis data revolution began with something very simple: understanding the trajectory of the ball.

In 2026, Hawk-Eye was formally introduced into the challenge system at Wimbledon and the US Open. For the first time in the history of this sport, a serve on the baseline stopped being a matter of the human eye. Since then, every serve, every forehand, every approach to the net has left a digital trace. Then came IBM SlamTracker, then Hawk-Eye Live replacing line judges, then tracking systems with dozens of cameras recording every footstep, every angle of hip rotation. Within less than twenty years, tennis shifted from a sport read through feeling to a sport read through algorithms.

I have watched that process from inside the analytics room. And I have seen a paradox: the more data there is, the easier it becomes to confuse what is easy to measure with what actually matters.

Take the simplest example. When a player serves, three things are measured instantly: speed, first-serve percentage, and the rate of points won when the first serve lands. These three metrics appear on screen before the point even ends. They are intuitive. They are easy to grasp. And they make viewers believe they understand the match. But the problem is this: none of those three metrics says anything about the person on the other side of the net.

A server hitting 70% first serves against a returner standing deep behind the baseline means something entirely different from the same figure against a returner crowding the baseline. The same number, two different stories. The same first-serve points-won rate, two opposite consequences. The stat sheet cannot distinguish these two situations, because it is designed to measure what the server does, not what the opponent is preparing to do.

This is where I must state my view on serve metrics clearly. They are not wrong. They are misread. And they are misread most by the very people who consider themselves data-literate.

First-serve percentage is a textbook vanity metric in tennis. It makes people feel safe. It makes a player look disciplined. But it measures courage at a very low level. A server who simply pushes the ball in safely at moderate speed can hit 75% first serves — and then be punished on every subsequent shot. A server willing to strike the line at maximum speed may hit only 60% — but when it lands, those are near-unreturnable points. The graphic loves the first player. The second player wins the big matches.

I once treated such a server as the analytics room's favourite child. He had everything a spreadsheet loves: a high first-serve rate, a dominant first-serve points-won rate, few double faults. We built an entire predictive model around him. Then came a quarterfinal against an elite returner, and he served 76% in — and lost in three sets. Not because he served badly. Because he served so safely that his opponent only had to stand in one spot and wait for the ball. The analytics room's favourite child must eventually stand on his own two feet. And that night, those feet did not hold.

The same happens with break points saved. On television, people celebrate a player with 70% break points saved. It sounds impressive. But remember that in a three-set match, a player may face only five or six break points. The sample size is so small that one lucky shot, one line call, or one net cord can change the statistic in a way that cannot be repeated. We are measuring luck and calling it nerve.

I spent a month and a half after Euro 2026 sitting down to check where I had gone wrong in my predictions. I built a spreadsheet comparing predictions to reality, not to congratulate myself when I was right, but to find my blind spots when I was wrong. And my biggest blind spot, repeating across many matches, was that I trusted clutch metrics — metrics computed from a handful of crucial moments — as if they reflected a fixed quality of a player. They do not. They reflect a few moments with a small sample size.

If you want an example of the power of data used correctly, look at the return points won metric. This is the most undervalued statistic in the sport. A player winning 25% of return points is already excellent. 30% is elite. 35% is almost an anomaly. But because it is small — because it does not produce a pretty number like 12 aces — it rarely appears on screen. Fans read a player by what he does when serving. But the greatest players in history were built on return points. Novak Djokovic is not famous for his serve. He is famous for turning every opponent's serve into an opportunity. That is a truth that serve metrics never tell.

There is something I learned from a personal project I ran in 2026, when every tournament froze because of the pandemic. I collected data from 312 matches across Europe's top leagues, comparing results with crowds and with empty stadiums. The finding startled me: the home win rate dropped from 46% to 38% without crowds, but the average goals per match rose slightly, from 2.67 to 2.81. Home advantage had vanished, yet the games became more open. I wrote a 5,000-word analysis and sent it to two editors. After two weeks of silence, one replied: this is the most original angle of the year.

I tell that story not to boast. I tell it to point to a principle I apply to tennis: public data can generate proprietary insight, but only when you ask it questions others do not ask. The problem with most tennis analysis today is not a lack of data. It is a lack of questions. People measure what is easy to measure, then report what they measured, then call it analysis.

Look at how we treat surface specialists. For years, a player who won a lot on clay was branded a clay specialist, and one who won on grass was called a grass specialist. This labelling persists because it turns a short run of results into an identity. The truth is that surface affects style, but much of what is called specialization is simply a coincidence of scheduling and a few lucky results early in a career. Once a player is labelled, he starts serving to match the label. The label becomes a self-fulfilling prophecy.

And this is where I want to go against the crowd.

What tracking cannot measure is pressure. And pressure is the single biggest variable in elite tennis.

I once watched a player with the highest first-serve points-won rate in a tournament walk into a final and serve like a completely different person. It was not technique that changed. It was not fitness running dry. It was the hand shaking. It was the breath shortening between two tosses. It was the eyes flicking up to the stands before serving. None of that lives in any camera system. It lives in the body of the person standing at the baseline, and no model can reach it.

I had such a moment at Euro 2026. In the 60th minute of the semifinal between Italy and Spain, I looked at real-time pressing data and said on air that Italy's pressing was declining sharply, that they would be forced into a substitution around the 70th minute, most likely Chiesa. Five minutes later, Mancini pulled Chiesa off in the 65th. A colleague beside me blurted out: how is that possible. The line was clipped and spread across social media. I received thirty-five calls from different networks over the next two days.

But I also received a warning from my superiors: do not turn yourself into a prophet, because the audience will set the bar too high. I never forgot that. Since then, every time I present an analysis based on real-time data, I state the limits of that data clearly. I state what it cannot reflect: player psychology, a sudden tactical surprise, a coach's decision no one predicted. Because data is only seasoning. People are the main course.

Back to tennis. There is another limit of data that few mention: it is computed on averages. And elite tennis is not decided by averages. It is decided by the most important points. A player can serve at a flawless average all match, yet serve badly on exactly three crucial points — and lose. On the report, he is still the better server. On the scoreboard, he is the loser.

That is why I always re-check every number with a single question: what is the evidence for this claim. If I cannot answer it, I do not write. If I can answer it but the confidence interval is too wide, I say so to the audience. I believe in a principle of transparency: I am 70% confident about this, 55% about that. Better to state your uncertainty honestly than to pretend certainty. A prediction with an honest confidence interval remains useful even when wrong. A prediction with no confidence interval is only useful when right — which means it is useless most of the time.

I have a growing worry about how data is being used at the youth-development level. Young coaches, under pressure for results, begin optimizing the metrics measurable at the U18 level: height, serve speed, physical power. They neglect what is harder to measure but decides a long career: feel for the ball, reading the game, patience in long rallies. The physicalization trend in adolescence is eroding the technical soil. We train young servers to hit at terrifying speed at seventeen, then wonder why they disappear at twenty-three, through injury or because no one taught them how to play a point when everything else has run out.

Most predictive models of young talent fail because they measure what has happened, not what can develop. They reward early bloomers and punish players who need time. We call it science, but it is really an opinion written in numbers.

So when someone asks me how far to trust data, I answer with another question: what question are you trying to answer. If the question is how fast this player serves, data answers very well. If the question is whether this player will win in the decisive moment, data hands you a pretty number and a polite lie. The boundary between those two kinds of questions is the boundary between useful analysis and decorative analysis.

There is something I always remind myself when I open a spreadsheet: a spreadsheet does not know what longing is, and we should not pretend otherwise. A player does not serve for a percentage. He serves because he wants to win, because he fears losing, because he wants to prove something to the person in the third row, because a memory of failure still burns in his chest. None of that appears in any column. But it appears in every ball.

The night in Russia in 2026 stays with me as a lesson about making safe predictions. Before the quarterfinal penalty shootout between Russia and Croatia, I analyzed on air that Russia had practiced penalties forty-five minutes a day throughout the tournament, but Croatia had a goalkeeper who had saved three in the shootout against Denmark. I predicted Croatia would win 5-4. The result: Croatia won 4-3. A young colleague texted: why didn't you commit to a more specific number. I realised I had made a safe prediction only because I feared being wrong. For a month afterward, I rewatched all sixty-four matches of the tournament, noting every play I had misjudged, to find the blind spots in my thinking.

I retell that story because it relates directly to tennis. People often confuse data giving us confidence with data giving us understanding. They are entirely different. Confidence makes us predict recklessly. Understanding makes us predict carefully while still daring to give a number. I learned that the most honest way to use data is to make a bold prediction with a clear confidence interval, then be ready to admit error the moment reality proves otherwise.

So, looking at the rest of the season, I am not searching for the player with the prettiest stats. I am searching for signals the data does not display. I am searching for a player beginning to read his opponent's serve half a second earlier. I am searching for a returner stepping half a step forward in the decisive games. I am searching for a player who, after losing a set to love, serves the next point harder rather than softer. Those are the real signals. They are not on the screen, but they are on the court.

The analytics room's favourite child must eventually stand on his own two feet — and fans, fortunately, can see those feet without needing a single spreadsheet. The next match you watch, try one thing: switch off the graphic in your head, and look only at where the returner stands. You will understand the match in a way no number can retell.