The Line Between Numbers and Inference in Tennis Analytics

Three in the morning in Sydney. I open the data file for a match I have been following all week, and the statistics column is empty. No first-serve percentage. No points won on second serve. No line about break-point conversion. Only the two players' names on the first row, and below them a blank space stretching to the end of the page. I sit still, hands on the keyboard. The screen glows in the dark apartment. Outside the window, a few dim lights still burn over Sydney Harbour. In my head, the deadline is nine in the morning. Six hours away. And I have a decision to make: write with what I have, or write with what I wish I had. The trap of the data-analysis trade is not that you lack numbers. It is that you feel driven to fill the gap with some number, any number, so the piece looks complete. I have seen colleagues do it. I have nearly done it myself. And every time, what suffers is not the article but the reader's trust. I have worked in sports data analysis for eighteen years, counting from the days when I recorded every phase by hand and typed it into a spreadsheet. Born in Vietnam, raised on evenings watching sport through a flickering television signal, then to Australia to study sociology, where I fell into the craft of reading numbers. That road taught me one simple thing: data does not create itself. There is always a person, a system, a process behind every figure. And if you do not know where a number comes from, you do not truly own it. Tennis is among the most thoroughly measured sports. Since 2026, the Hawk-Eye system has been used at major events to judge whether a ball is in or out, and ever since, every serve and every approach to the net has left a trail of coordinates. IBM SlamTracker turns those trails into charts for television audiences. Data platforms such as StatsBomb and Tennis Abstract extend collection across the entire professional tour. Today, even a first-round match can generate thousands of data points. The paradox lies here: the more data there is, the easier it is to believe every question has an answer. Yet most tennis metrics tell only half the story. First-serve percentage tells you how often a player landed the serve, but not the quality of that serve. Points won on second serve tells you how well they escaped trouble, but not how the opponent returned. Every number is a slice, and every slice has an edge. This is why I begin every analysis with a question I ask myself: before believing a number, ask where it was born. When the data pipeline returns an empty result, the first reflex of an inexperienced analyst is to fill it. The correct reflex is to stop and trace. I divide this work into nine tiers, and each can collapse if the foundation is not solid. The first tier is technique and tactics. Here the analyst must identify a player's style: attacking or defensive serving, net-rushing or baseline-anchored, forehand- or backhand-oriented. But to identify style, you need point data by shot type. If the column is empty, every claim about style is guesswork. I once read a piece asserting a player had clearly improved the backhand, with no data at all on backhand points won. That kind of writing belongs to literature more than to analysis. The second tier is data and form. This is where numbers speak. First-serve rate, points won on second serve, break-point conversion, rally points won, deciding points won together form a picture of current form. But that picture only has value when you compare it with the same player's past and with the tour average. A break-point conversion of 40% sounds impressive until you learn the tournament average is 45%. A number has no meaning outside its frame of reference. The third tier is tournament structure and schedule. A player does not compete in a vacuum. They move from one surface to another, from one event to the next, and each transition leaves traces in the body and in the data. Entry density, gaps between events, the time of year all affect results. Ignore this tier and you will misread a defeat. I have seen a player criticised for a form slump while their schedule was half again as heavy as their rivals' over three straight weeks. The fourth tier is tour context and player standing. A world No. 5 and a world No. 50 carry completely different expectations. The same result means different things. This is where data analysis meets sociology: a number does not sit outside context, it sits inside it. When I studied sociology in Australia, I realised that quantitative and qualitative methods do not oppose each other; they complement each other. A rate means something only when you know who it belongs to, and under what circumstances. The fifth tier is rules and governance. Tennis has rules on serve timing, off-court coaching, medical timeouts. Each rule can become a variable in analysis. A player calling a medical timeout at a decisive moment can change the match, and the analyst must record it as a fact, not an accusation. The difference between recording and accusing is the difference between analysis and judgement. The sixth tier is team and player management. Coaches, fitness specialists, medical staff, commercial agents form the ecosystem around a player. A coaching change often comes with a change in playing style, and that is a signal worth tracking. But to measure the impact of such a change, you need at least ten matches of subsequent data. Without data, you are merely telling stories. The seventh tier is risk. Injury, ranking-points pressure, schedule overload can all be quantified if you have data. Without it, you are merely worrying on the player's behalf. In my trade, worry is not an indicator. It is an emotion, and emotion does not belong in a data table. The eighth tier is media narrative and expectation. This is the tier I consider most dangerous, because it is the easiest to inflate. A player winning three straight matches can be called a title contender by the media while the data shows they won only through luck at the deciding points. The gap between narrative and reality is exactly where the analyst creates value. If you merely repeat the narrative, you do not need data. If you verify it, you need a great deal. The ninth tier is industry transmission. Prize money, sponsorship deals, event investment, equipment technology form a chain in which an event at one end can ripple to the other. A champion can ignite racket sales in a distant market. That transmission chain runs by its own rules, and a good analyst reads it before it appears in financial bulletins. Those nine tiers are the skeleton of any serious analysis. But a skeleton only means something when there is flesh. When the data is empty, that flesh does not exist. And all that remains is an empty frame, handsome in form but hollow in substance. In my career, I have stood before that moment many times. In 2026, fresh into a newly founded sports site in Australia, I published a long analysis of Melbourne City's pressing metrics, built on GPS positional data. I showed that midfielder Luke Brattan had to run more than eleven kilometres per match yet produced barely more than one successful tackle, and that the team's pressing direction was misaligned. The piece was mocked as too dry. Three weeks later, the coach changed the pressing scheme, and the team won four straight matches. The lesson was not that data is always right, but that data is right only when you stay patient with it to the end. In 2026, I wrote a prediction about a national team at the World Cup based on expected goals. I was called a bookworm who knew nothing about football. But that team reached the final. After the tournament, a journalist from a major sports outlet contacted me to ask how I calculated a defensive metric. I spent two weeks writing code, cross-checking the data, and sent back a seventeen-page analysis. What I learned: a reader's scepticism can turn into trust if you are transparent about your method. In 2026, when the pandemic emptied stadiums, my prediction model collapsed. I had priced home advantage at 0.45 goals per match. After nine rounds without crowds, the figure fell to 0.08. I had to decline a commission to explain empty-stadium football, because I needed three more weeks of data before I dared conclude. When I finally published, I admitted I had been wrong not to account for the crowd variable. Since then, every piece of mine carries a section titled Assumptions That Could Be Wrong. Those three stories differ in setting but share one lesson. Numbers whisper. Those who listen will hear an entire match. The hasty hear only themselves. And in a major-tournament season, when everything moves faster, one's own voice more easily drowns out the data. Back to the empty file in Sydney. I have three choices. One is to write from memory of the match I watched, knowing memory is a poor data source. Two is to fill the page with estimated numbers so the piece looks substantial. Three is to call the data provider, verify the pipeline, and if necessary write a different piece. I choose the third. That choice lacks glamour. But it is the only one that lets me sleep after the piece goes live. There is something the analysis world rarely admits: an empty dataset is sometimes a gift. When you have too many numbers, you tend to believe everything can be measured, and that what cannot be measured does not matter. Emptiness forces you to face your own limits. It reminds you that you are a reader of traces, not a machine producing conclusions. The tennis-analytics industry is racing the opposite way. Platforms compete on the quantity of metrics: serve rate, spin speed, bounce height, ball trajectory, reaction time. Each new metric is a promise of deeper understanding. But that promise holds only if the user knows what they are looking for. A metric collected without a guiding question is like a door opening into an empty room. A complex model with hundreds of variables can predict worse than a simple one with five, if the variables in the complex model are sloppily collected. This is what amateur analysts often overlook. They believe more data means more accuracy. In reality, dirty data multiplies exponentially as variables increase. Every added variable is a new opportunity to be wrong. I once joined a project building a composite index for a tournament. At first we fed in twelve variables. Prediction was poor. We cut it to four. Prediction improved. Those four were collected by hand, cross-checked by two independent people, and had clear provenance. The other twelve came from an automated interface with an unknown error rate. We did not know what we were measuring, so we measured wrongly. The lesson: the quality of one number matters more than the quantity of numbers. And to judge quality, you must know where a number comes from. Who recorded it? Which system produced it? At what moment? Under which definition? Those four questions are the sieve every datum must pass before entering a piece. This is also why I always note the data version in every article. A season missing detail is like a match missing stoppage time. You may win, but you do not know how. And what you do not understand, you cannot reproduce. A conclusion that cannot be reproduced is not knowledge; it is merely an opinion dressed up in numbers. In a major-tournament season, the pressure to write fast is greater. Audiences are swept up in stories about national teams and players, and writers are swept along with them. But precisely then, data discipline must be held tightest. Because in a short tournament, a wrong conclusion can outlive the event itself. Fans will remember it, repeat it, and fold it into collective memory. An analyst's error is not only their own. Current data shows me a trend: tennis-analytics platforms are shifting from collection to verification. The question of the future will no longer be how many metrics we have, but which metrics we can trust. Whoever answers that first reshapes how fans understand the game. This is not a bold prediction. It is a cautious inference from what I have observed over eighteen years. And if the pipeline is empty again tonight, I know what I will do. I will make the call. I will wait. I will write when there is enough evidence, or I will write nothing at all. Readers deserve a verified truth, even when that truth is silence.