Tennis Data and the Trap of the Empty Analytics Board

The clock in Brisbane read 2:17 in the morning. On the left monitor sat an Excel sheet with 87 columns of data I had built for the current Grand Slam season; on the right monitor was the live feed from a quarterfinal. Then the feed stopped. It was not the network, and it was not the provider. The analytics board simply returned a blank, every cell reading N/A, as if the match had never been recorded. I sat still for ten minutes, watching two screens face each other like strangers sharing a room. It was the first time in nine years on the job that I realized an empty data board can teach more than a full one. The truth is, my trade lives on numbers that never sleep. Every serve is measured, every footstep is tracked, every point is logged before the fans finish a sip of water. Hawk-Eye records thousands of data points per set. Systems such as Tennis Abstract, UTS, and the internal platforms of the ATP and WTA track everything from first-serve points won to successful net approaches. Yet most analytics boards I read each week, including some I write myself, are empty of meaning. They contain numbers, but not information. They show that something happened, but never why it mattered. I joined Sports Illustrated in 2026, starting as a fact-checker. The first job of a fact-checker is not to find a beautiful number but to find the number that is lying. I learned that a player's first-serve points won can jump simply because his opponent served badly in that set, with nothing to do with his own progress. I learned that a player winning 70 percent of net points may have approached the net only three times all match, a pretty number with no meaning. And I learned the most important lesson: an analytics board has value only when it dares to admit its own limits. That is why I begin every morning with a single question: which number today might be lying to me? The question led me to an uncomfortable conclusion. Much of the modern tennis analytics industry operates as a machine for producing confidence, not a machine for producing understanding. We measure more than ever, but we understand less than we think. To make this concrete, consider a typical week in a Grand Slam season. On Monday morning, the serving data for all 128 players at a major is updated. The first column is first-serve points won, the second is second-serve points won, and the third is breaks suffered per game. It looks scientific. But when I rank all 128 players by first-serve points won, the ranking almost matches the seeding. No surprise. No discovery. A beautiful board that is empty of information, because it only repeats what everyone knows: strong players serve well. Real analysis begins when I split serving data by situation. A player may hold 92 percent of ordinary service games but only 61 percent of set-deciding games. That 31-point gap is where the story lives. The overall serving rate does not lie, but it also says nothing. The serving rate in pivotal games is what separates a champion from a runner-up. I once built a simple model to predict Wimbledon quarterfinal results from four variables: first-serve points won over the first three rounds, return points won, tie-break win rate for the season, and average minutes per match. The model called three of four quarterfinals correctly that year. But in the semifinals, it missed both. I spent two days understanding why. The answer lay in a fifth variable I had ignored: rest days between rounds. One player had won his quarterfinal in four tight sets on Wednesday night, while his opponent had finished his on Tuesday afternoon in three. That 26-hour gap appeared in no technical column, but it decided the semifinal. This is the lesson I repeat to myself every week: data never lies, but those who read it make excuses. When my model is wrong, I do not blame the data. I blame my belief that four variables were enough. The arrogance of an analyst lies not in using too little data but in thinking he has enough. In tennis, there is an even more uncomfortable paradox than a wrong model: the paradox of overpraised metrics. Take second-serve points won. For years it was treated as a measure of nerve. A player who held a high rate was said to have ice in his veins. But when I separated the data of 50 top-100 players across two seasons, I found that players with aggressive second serves, those who chose risk, often had lower second-serve points won but won more service games. In other words, the celebrated metric was measuring caution, not courage. It reflected the opposite of what people believed. That was the second time I realized a board can be empty in a subtler way. It returns no N/A. It returns a positive number, but that number measures the wrong thing. More dangerous than emptiness is confidence in the wrong place. I call this the decorative metric. A decorative metric is a number included because it is easy to collect, easy to chart, and looks scientific, not because it offers predictive power. Aces are an example. They glitter, they are easy to count, they look good on screen. But when I checked the correlation between aces and Grand Slam match win rate, it was far weaker than the media imagines. A player can strike 20 aces and still lose, because serving is only half the match. The other half, returning, is where data is often treated unfairly. In most boards I read, the return column has a single line: return points won. But to know how good a returner really is, I need to know how many return points he wins on second serves, what percentage of second serves he attacks, and whether he still swings when down 0-30 in a return game. Without those three numbers, the overall figure is just an overall figure. Carlos Alcaraz taught me this lesson most clearly in recent seasons. His stats often tell one story, while his matches tell another. His return points won at many events are not among the leaders, yet he wins more return games in deciding sets than almost anyone. The difference is that Alcaraz does not distribute effort evenly. He concentrates his attack in the games where he senses fragility. That is a tactical skill no average stat line captures, because it requires time-series data rather than aggregate data. Jannik Sinner moves in the opposite direction. He has almost no technical weakness in the stat sheet: high first-serve percentage, steady return points won, few unforced errors. When I look at Sinner's board, I find no gap to attack. But that very stability creates a new kind of emptiness: a perfect board leaves no room for surprise. His opponents cannot prepare by targeting a weakness, because there is none. They must prepare by accepting that they will have to play above their usual level for at least three sets. That turns every Sinner match into a psychological battle before it becomes a technical one. Here I must address the limits of my own method. A stat board, however detailed, is a snapshot. It cannot describe the flow of a match. It cannot tell me when a player began to change tactics, when he lost faith in his serve, or when he decided a set was not worth exhausting himself for. I have spent years building point-by-point tracking boards, but I have yet to build a decision-by-decision one. The biggest shock of my analytical career came in 2026. I built a prediction model from six major tournaments of historical data, using Elo and qualifying results. It ranked Brazil as the top candidate with a 23.4 percent title probability. I was confident enough to write that the data had revealed the champion. Brazil fell in the quarterfinals, while France, ranked only fourth at 11.2 percent, lifted the trophy. That 2026 taught me that a 95 percent probability still leaves a 5 percent that smiles, and that a model can be right about a trend but wrong about people. After that shock, I removed the word "certain" from my analytical vocabulary. I replaced it with "confidence interval." I began publishing the limitations of my model at the end of every piece, because I understood that numerically literate readers will forgive an error but not a concealment. In tennis, disclosing error matters even more than in football, because the sample is smaller. A player competes in 60 matches a season, but only four Grand Slams, and at each major a top player may play just 21 sets. With such a small sample, a single upset can shift an entire trend. I have seen a tie-break win rate move 12 percentage points after just two tie-breaks. That is why I always tell readers to approach tennis data like a long-distance hiker, not a weather-app user. One season taught me this most beautifully. In 2026, when tennis returned after the pandemic in empty stadiums, I ran a before-and-after comparison. The results echoed what I had seen in football: players served more cautiously without crowds, pressure eased but so did adrenaline. From the empty stands I could hear the match breathing, and that breathing was slower, steadier, less panicked. But I also had to concede that my sample was only a few dozen matches, and every conclusion belonged in the category of the unverified. That honesty did not weaken my writing. It made it more credible. When an analyst admits he does not know something, readers start trusting him on what he claims to know. This is why I believe most modern tennis analytics boards suffer a structural flaw: they are built to answer the wrong question. Metrics such as first-serve points won answer how well a player serves overall. The right question is how well he serves in decisive games, against a strong returner, in the fourth set of a three-hour match. The second question requires situational data, and almost no public board provides it. I call boards that answer the wrong question empty analytics boards. They have numbers. They have charts. They have colors. But they have no answer. Worse, they prevent readers from asking the right question, because they create the feeling that everything important has already been measured. It should also be said that not all emptiness is the analyst's fault. In tennis, data on squad depth and mental state barely exists in quantified form. No metric captures how many hours a player slept before a semifinal, or whether his coach is going through a personal breakup. When I write about a season, I must concede that my model ignores almost every human variable. This is where I part ways with many colleagues. Many believe the growth of data will gradually fill these gaps. I do not. I believe some gaps will never be filled, because they are not in the nature of data. A missed serve at a decisive point may be technical, psychological, surface-related, or simply something only the player knows. Data can narrow the explanatory gap, but never close it. That is why the format I always choose for myself is one that admits imperfection before presenting conclusions. I call it marked-hole analysis. Instead of hiding missing-data zones, I flag them. Instead of saying "data shows," I say "available data shows, but missing data could change the conclusion." This style is uncommon and sometimes irritates my editors. But it is honest. There is a second reason I hold this line, and it concerns what I consider the darkest side of sports digitization: live data sold to betting companies. When every point, every serve, every movement is logged in real time, the greatest value of the data is not with the fans. It is with parties able to turn it into odds before the audience understands what is happening. Fans get charts; betting firms get an information edge. I do not write to serve either side of that game. I write to say that an empty analytics board is not only a technical problem but an ethical one. An empty board makes fans believe they understand a match, when in fact they are being given a feeling of understanding. And a feeling of understanding, when deceived, leads to wrong decisions, in betting, in commentary, and even in how a young player views his own career. I once watched a young player change his serve motion because a stat board showed his first-serve rate below average. He spent the winter fixing it, and the following season he served more safely, his first-serve rate rose, but his service-game win rate fell. The board was right. He did what the board advised. And he played worse. That is the price of an empty analytics board: it optimizes a column, not an outcome. This brings me to a conclusion I have held for years: in tennis, the most important metric is not the one you measure but the one you choose not to measure. A player can win 65 percent of net points and still lose, because the other 35 percent fall at decisive moments. A player can hold serve 90 percent of the time and still lose a set, because the remaining 10 percent is the twelfth game of the fifth set. Aggregate numbers, however alluring, always hide the distribution. And the distribution is where the match actually lives. There is one lesson I learned from a match with no data at all. It was a match I watched with my eyes, no stat board, because the data system crashed mid-match. I sat and watched two players fight for over three hours in a quarterfinal. No xG, no ratios, nothing but what I saw. When it ended, I realized I understood that match better than any other that week. I could not cite a number, but I could tell the story. That is why I believe the best tennis analysis is not the one with the most numbers but the one that knows when to use numbers and when to tell a story. Data is an excellent guide and a terrible master. A good analyst uses data to find the path, then leaves it behind when entering the forest. This is the part where I must be honest with myself. Over the years I have written many pieces I am proud of. I have also written many I regret. The ones I regret most are not the ones with wrong predictions. They are the ones where I used data to manufacture confidence I did not have. The ones where I presented a conclusion as a law, when it was only a hypothesis. The empty pieces, wrapped in beautiful numbers. I have fixed that. I have learned to write sentences like "I do not know," "the data is insufficient," "the model may be wrong." I have learned to mark the hole instead of covering it. And I have learned to trust that readers are smart enough to accept an analyst who admits his limits. Looking back at the past season, I see that most tennis debates, about who is best, who will win, who is declining, are built on empty analytics boards. We argue about the number and forget that the number never carried context. We compare players at different career stages, on different surfaces, against different opponents, and call it objective truth. Objective truth in tennis, if it exists, lies somewhere else. It lies in a player's ability to win a point when every metric is against him. It lies in the moment a player chooses to hit the line while the safe shot is down the middle, because he knows the safe shot will lose him the match in the long run. Those moments appear in no column. Yet they decide the match. That is what I want readers to carry into any tennis analytics board, including mine. Ask what the board measures, and whether what it measures is what matters. Ask what was left out, and whether the omitted part is larger than the kept part. Ask whether this beautiful number is hiding a hole larger than itself. An empty analytics board is not a board without data. It is a board with data but without truth. The difference between the two is my entire job. As the Brisbane sun rose, I closed both monitors and went to sleep. My board was still empty. But that evening, when the next quarterfinal began, I knew exactly what to look for. I looked for the moment when numbers could no longer speak and only people could. That is when the match begins. That is also when my work begins. The next major season will bring a new data board. It will be fuller, more detailed, with more columns. But I know it will still have a hole. My task is not to fill that hole with false numbers but to mark it, so readers know that a match, however measured, is always larger than the sum of its numbers. I will keep asking whether my model omits a variable, and whether my confidence interval is truly wide enough to contain the truth. I will keep writing marked-hole analyses, because I believe honesty is the highest form of analysis. And if you ask me who will win the next tournament, my answer will not be a name. My answer will be a question: what are you measuring when you ask who is best? Because the answer depends entirely on which column you choose to look at, and on what you choose to ignore.