The Report With Empty Columns: Tennis Analytics' Real Shortage Is Data, Not Models

At eleven at night a file sat open on my screen and every column in it was empty. Title, source, match, player, statistics, date — null across the board. The entire next day's analysis was supposed to stand on that file. Ten years ago I would have panicked at the sight of it. Five years ago I would have tried to fill it in — from memory, from instinct, from approximation. Tonight I closed it and made tea. An empty column is a finding too; it just isn't a finding about tennis. It is a finding about the system that measures tennis.

I first understood this in 2026, when the Hawk-Eye data from the last edition of the Chennai Open sat on my laptop. Twelve matches, fourteen columns, and inside them the pairing of Leander Paes and Rohan Bopanna. The serve-placement chart looked clean until we opened the column where the doubles court goes quiet. Second-serve win percentage on the deuce court read 48, against a tour average of 54. That pair was winning its quarterfinal. Winning and being solid are not the same thing, and a scoreboard paints both in the same colour.

Context

Tennis's data supply chain has three floors, and each floor charges a different rent. At the top sit the Grand Slams and the Masters 1000s, where Hawk-Eye, serve speed, rally length and return depth are all public. In the middle is the ATP Challenger Tour, where point-by-point data arrives half-built and some weeks consist of nothing but a scorecard. At the bottom are the ITF and junior circuits, where a player's story does not travel further than his coach's WhatsApp group.

The 2026 edition was the Chennai Open's last; from 2026 the event became the Maharashtra Open in Pune. That same year I worked with a San Francisco sports-data firm that held a contract with the Maharashtra Open. There I applied PPDA to the first-round matches of Indian wildcards. PPDA — passes per defensive action — is a football metric, and I redefined it for tennis: balls played per defensive return against the second serve. The wildcards averaged 12.3; the qualifiers 8.7. A local Pune paper cited the report and I received a full-time offer.

This is where tennis's money and tennis's attention take separate roads. Analytics budgets flow toward Slams and singles; the largest share of professional players lives in Challengers and ITF events. A Challenger week — flight, hotel, coach, balls — costs somewhere between fifteen hundred and twenty-five hundred dollars on the basis of conversations inside academies, and the ranking points it returns are countable on two hands. No public database tracks that cost per point. The most expensive information in tennis is the information the scoreboard never records.

Source quality becomes the sharpest question here. Official tour statistics, tier-1 media reporting and a social-media clip carry three different reliability ceilings, and dumping all three into one table is the most common error in the field. I state the source tier behind every claim, because a conclusion built on a clip can do more damage than any model.

Core

The first lesson hides in the denominator. Behind that 48 percent at Chennai sat forty-one points on second serves to the deuce court. Twelve matches, forty-one points — that indicates a direction, it does not deliver a verdict. A five-point gap against the tour average looks dramatic, yet on a sample that small a swing of one or two points is ordinary. Since then my rule has been fixed: publish the denominator beside every rate, and where the sample falls below thirty points write "directional," never "conclusive." Pasting a heat map onto a blog is easy; typing "n=41" underneath it is hard, and that hard part is what separates an analyst from a reporter.

The second lesson came from a load curve. In 2026, with tennis shut down, I was consulting remotely for an academy in Bengaluru. GPS data from a twenty-year-old's 2026 Challenger matches landed on my desk. His average distance covered per point was 12.4 metres, eighteen percent above the Challenger average for players of his style. I flagged it as an injury risk. Six months later he suffered a stress fracture in his left foot. Distance covered can look like courage until the load curve starts predicting the injury. And the stress fracture was no miraculous warning; it was a data point that arrived before the diagnosis.

My own caution belongs in the same paragraph. One player, one season, one metric — that is correlation, not cause. I attached a data-limitations section to that report and I still attach one today, because an injury story slides into a character story faster than any other kind. A distance-covered figure does not say how hard a player worked; it says how much ground was asked of him.

The third lesson is about translation. Football's vocabulary is easy to drop into tennis and dishonest in most of the places it lands. There is no clean tennis equivalent of expected goals, because tennis's unit is the point, not the possession. When I used PPDA I first defined what counted as a defensive action here — a returned ball against the second serve — and only then applied it to wildcards against qualifiers. Definition first, number second. Reverse the order and what comes out is decoration, not analysis.

The fourth lesson is structural. Every major Indian trophy sits in a doubles or mixed draw, and nobody has built the analytics scaffolding for doubles. Serve placement means something entirely different there: a wide serve from the deuce court forces the net man into a decision on his first step. If he is winning 48 percent behind a second serve, the real question is not whether the serve was poor; the real question is whose positioning structure it was built to serve. Wildcards, Davis Cup rubber order, Olympic partner choices — I read all three as structure rather than sentiment. State the intended design first, then grade the selection against it.

The fifth lesson produced that empty file. Tennis's most consequential stories rarely arrive in the shape of a statistic. Ranking points roll on a 52-week cycle, which means points earned in this week last year expire this week and must be defended again. I call it the points-defence cliff, and it shows up on a calendar, not in commentary. A player returning from long-term injury enters on a protected ranking, the ranking frozen at the time of the injury. The medical time-out is an in-match treatment break whose tactical use remains a standing governance argument. Surface switching is the same category of problem — reaching grass from clay in five weeks is a distinct physical load, and at Challenger level there is no data measuring it. All of this travels in prose, not in tables, and that is exactly why it is the first thing to vanish from a data-driven pipeline.

Contrarian

I carried a prior into the Maharashtra Open: that wildcard selections had no return-pressure structure behind them. The PPDA numbers agreed with that prior — 12.3 against 8.7. The danger is precise here, because when the data agrees with a position you already hold, you stop auditing. So I will write what I wrote then: PPDA did not call the wildcard a mistake; it showed that the selection had no return-pressure structure. The structure may have been something else entirely — a local gate, a sponsor clause, a young player given a big court. Until the intent is stated out loud, a selection is neither a gift nor a blunder.

The second counter-intuitive reading comes out of the empty file. Empty data is not an accusation. That a file contains no reference to doping, match-fixing or a disciplinary matter does not mean something was there. Turning absence into evidence is the worst failure available to an analyst, and declaring the file empty was the only way to avoid it.

The third cuts against my own identity. When an old read survives contact with the data, write it up with the same seriousness you reserve for a debunk. Tennis's oldest truth — that holding a top-100 slot means grinding points in Challengers every week — still survives. Setting it aside as old news is not analysis; it is laziness.

Takeaway

Two things will hold my attention over the coming months. First, the price at which point-by-point data opens up at Challenger level; if low-cost video tracking reaches academies, the data density of the Challenger tour changes inside three years. Second, which player's ranking the 52-week rollover hits first during the Asian hard-court swing. On the day a commentator holds up a 48 percent figure to bury someone, the first question should not be how bad it was. The first question should be how many points sat in the denominator.