Inside the Brøndby Analysis Room: When Badminton Data Reaches Its Own Limit

Brøndby Training Hall, 7:40 a.m.

Brøndby Hallen, on the outskirts of Copenhagen, has its own smell in January mornings: varnish, floor rubber, and stale sweat left over from the night before. I sat in the top row, laptop open, watching Anders Antonsen run a drill he has repeated more times than anyone could count: smashing shuttles into a square of white tape at the far left corner, over and over, until the coach called a halt.

At the eighteen-minute mark I started timing the interval between consecutive smashes. It averaged 3.4 seconds. The standard deviation was so small I checked my stopwatch twice. One man, smashing into one spot, at one rhythm, for nearly twenty minutes, and his body never wavered.

If you only read the data I wrote down that morning, you would see a nearly flat line. You would conclude the session had nothing to say.

You would be wrong in a very understandable way. And that is why I am writing this.

A Japanese analyst in the middle of Danish badminton

I was born in Japan, graduated in broadcast journalism from the University of Copenhagen, and stayed in Denmark. My job title is sports data analyst, but the actual work is closer to translation: I translate human movement into numbers, then translate the numbers back into human movement. The second translation is always harder than the first, and I have made enough mistakes to know it.

Denmark is an odd market for this work. Fewer than six million people, yet badminton sits close to the position of second national sport behind football. A small country produced Morten Frost, Peter Gade, Poul-Erik Høyer Larsen, and most recently Viktor Axelsen, Olympic champion in Tokyo in 2026 and Paris in 2026, plus world titles in 2026 and 2026. Per capita, that is one of the strangest ratios in European sport.

Danes do not like explaining that anomaly with numbers. They explain it with stories: the cold of a Jutland winter, provincial halls with three hundred seats that still fill for third-tier events, nine-year-olds driven to practice at five in the morning in the dark. When I bring a spreadsheet to a coach here, the first reaction is almost always the same question: "What story is this number telling me?"

That is the right question. It took me years to learn to answer it without falling into the standard trap of the trade: believing that what can be measured matters more than what cannot.

The first metric I calculate: rally length

Back to Brøndby. After logging the smash rhythm, I opened the dataset I build myself from BWF World Tour matches in the 2026 season. I call it a dataset, but it is really a huge spreadsheet of more than four thousand rallies labelled by hand, work I usually do in the evening, alone, with noise-cancelling headphones.

The first thing I look at is rally length — the average number of shots before a point ends. In men's singles at Super 1000 level, the figure I get hovers between nine and eleven shots, depending on the event and the surface. For Antonsen in the 2026 season, it sits roughly 1.5 to 2 shots above the tour baseline.

That sounds small. Multiply it out. A three-game match runs 70 to 90 minutes, with roughly 150 to 200 points played. An extra 1.5 shots per rally means 250 to 300 extra shuttle contacts — each one a movement, a load-up, a drop of the centre of gravity. The difference between a rally-extender and an early finisher is not in the beautiful technique. It is in how many times the legs are asked to absorb it.

That is why I start with the legs, not the smash.

In that session, Antonsen's drill was not measuring smash speed. It was measuring the ability to reproduce the same movement after fatigue had arrived. By minute thirty his interval was 3.7 seconds. By minute forty, 4.1. My line began to bend downward, and the bend itself was the real information.

A player is rarely beaten by his opponent's hardest smash. He is beaten by second 3.9 instead of second 3.4.

What the sensors cannot measure

The Danish national team has used racket-mounted and shoe-mounted sensors for years. I once sat through an export session, and the first feeling was vertigo: swing speed, racket-face angle, impact force, distance covered, jump count, heart rate, reaction time. Everything has a number.

But there is one thing the system cannot measure, and it took me a long time to accept that this is not a technical shortcoming. It is a limit of the category itself.

Between points, the interval runs 15 to 25 seconds under the current 21-point rally scoring system. Inside that interval, something happens that no sensor records: the player decides what he is going to believe in. The straight smash or the drop. Extending the rally ten more shots or ending it now. Believing in his body, or believing the opponent will break first.

PPDA cannot measure the heart, but it does point to where the heart is beating. Rally length in badminton works the same way. It cannot tell me what Antonsen was thinking at second 3.9. It only tells me where he was suffering, and where he was still standing.

I once wrote a piece arguing that a young player's defensive metrics were terrible. A coach called and asked one question: "Did you watch the third game?" I rewound the tape fourteen times in the editing room until three in the morning, and realised I had missed the most basic thing: he was playing on an ankle that had not healed, and the whole ugly defensive profile was the consequence of a conscious decision — cut the jumping, concede a few points, save the leg for the later rounds.

I sent an apology email and rewrote the piece in two versions: one by the numbers, one by the eye. Both were right. Both were insufficient.

Antonsen and the sentence handed down by numbers

There is a type of player the analytics world files in the wrong drawer. Antonsen is that type.

For years the popular description of him on international forums was "lacks a kill shot", "no smash heavy enough to finish at the top level", "too safe". Those remarks are not entirely baseless. Put him next to Axelsen at peak form and the contrast is brutal: Axelsen's smash has been recorded above 400 km/h on favourable serves, a number that makes crowds gasp. Antonsen has no weapon like that, and he has never pretended otherwise.

But look at the data from a different angle and the picture changes colour.

Across the 2026 season, compiling his matches at Super 750 and Super 1000 level, Antonsen's win rate in rallies lasting more than 15 shots was markedly higher than his win rate in rallies under 6 shots. He wins more when the match goes long. That is not coincidence — it is a tactical choice executed systematically across hundreds of rallies.

I call it the strategy of accumulation. It produces no moments that make the television highlights. It produces moments in the third game, at minute sixty, when the opponent has hit seventeen heavy smashes in a row in game two and the eighteenth now lands three centimetres past the line.

A rally-extender does not play to score points. He plays so that his opponent has to score them while exhausted. The philosophical gap between those two approaches is far wider than the technical one.

What interests me is how the public reads it. When a big hitter wins, we call it character. When a rally-extender wins, we say his opponent played badly. Same result, two readings, and the second reflects an aesthetic prejudice rather than a question of effectiveness.

At Brøndby that morning, I asked about the taped-square drill. The answer was short: "I don't train to play better. I train to play the same at minute seventy as I do at minute seven."

That is a claim you can measure. And when you measure it, it holds.

Head-to-head: when the sample is too small

One of the strongest temptations in this trade is over-reading head-to-head records.

If Antonsen has faced someone twelve times and won eight, that sounds weighty. Break it apart. Four of those twelve came before 2026, when neither was at his peak. Two happened in indoor arenas with strong air-conditioning currents, where the shuttle flies completely differently from a sealed hall. One ended in retirement after fifteen minutes of game one.

After removing the non-comparable cases, I have five matches left. Five matches is too small a sample to conclude anything tactical. It is only enough to raise a hypothesis.

Here is what I always have to remind myself. In football, a season has 38 matches, and a player may accumulate hundreds across a career. In singles badminton, a top player competes in roughly 15 to 20 events a year, three to six matches each, and most of the matches that matter most happen only once in a lifetime. Our denominators are alarmingly small.

With a denominator that small, every tactical conclusion is a hypothesis dressed up in numbers. I write that not to dismiss analysis, but to insist that badminton analysis needs more humility than football analysis, not less.

My way around it is to change the unit of observation. Instead of counting matches won, I count rallies. A player competing in 20 events a year, 180 points per match, is playing more than 3,500 points a season. That is a large enough sample to talk about tendencies — provided you accept that you are talking about tendencies, not outcomes.

The contrarian angle: correlation is not causation

I have an old lesson I retell often, and this time I will retell it through badminton.

In 2026, as a student, I wrote my thesis on a Danish football club famous for developing young talent. I calculated PPDA — passes allowed per defensive action — and found they pressed intensely, about 2.1 units below the rest of the league. They finished seventh. The grading panel called my paper "dry as stale bread".

I sat alone in a Copenhagen café wondering why numbers that clear could not convey any heat.

The answer came later, and it applies directly to badminton: a metric that describes behaviour does not automatically explain outcomes, because between behaviour and outcome there is always a gap that data cannot fill. That team pressed hard because they could not control the ball. Their style was the consequence of a limitation, not the cause of a success.

Badminton is full of the same trap. The player with the highest distance covered per match might be the hardest worker in the draw, or he might be the one being pulled around the most. On the spreadsheet, those two cases are identical. On court, they are opposites.

There was a 2026 World Cup match I analysed for a Danish broadcaster. I wrote that the national team pressed "chaotically" because the PPDA was very low. A former international criticised me live on air with a single question: "Have you watched the tape?" After rewatching it fourteen times, I understood the low figure came not from chaos but from a deliberate defensive plan: concede space in the middle to seal both flanks. That was a decision. A metric cannot distinguish between a decision and a mistake.

Denmark against France was not a failure of data. It was a failure of mine, for thinking data was everything.

I keep that lesson to this day, sitting over badminton footage. Every time I see a beautiful metric, I ask: what behaviour produced this, and was that behaviour a choice or a consequence?

When the rules change: the biggest hidden variable

There is one factor almost nobody accounts for when reading badminton data: the scoring system can change, and when it does, nearly all historical data loses part of its comparative meaning.

The current 21-point, best-of-three rally system has been in place since 2026. Before that, badminton used the service-holding system, where a match could stretch for hours. In 2026 the world federation proposed an 11-point, best-of-five format; it was rejected. More recently there have been trials with a 15-point, best-of-three format.

From an analytical standpoint, this is not merely a matter of numbers. The scoring format decides which kind of player gets rewarded: the shorter the system, the more the smash is worth; the longer the system, the more endurance is worth. If the format changes, every model I built on 2026 data will need recalibrating from scratch.

So I keep one rule: when I read a badminton metric, I must know which format produced it, on what surface, at what altitude, and in a hall with or without air conditioning. A metric without context is not a metric. It is a number severed from its body.

I once reached a conclusion about home-win rates from 120 matches played without crowds during the pandemic. The home advantage dropped sharply. But what stayed with me after weeks in empty arenas was not the figure. It was the sound of a shuttle meeting a racket with no roar answering it. The echo of a rally in an empty hall is one of the loneliest sounds I know.

The dead season taught me this: an empty arena is the final test of data. In silence, data becomes suspiciously pure, and that reminds me that most of the chaos in sport comes from people, not from tactics.

The industry current, from racket factories to provincial academies

During football's transfer window everyone talks about money. Badminton runs on a different cycle but has its own cash flows, and they travel along paths viewers rarely notice.

Start with the racket. A professional uses ten to twenty frames a season, each restrung after a few sessions, with tension tuned to the surface and the opponent. It is a small market with extremely high attachment: when a leading player switches brands, he brings hundreds of thousands of potential buyers in his home country with him.

Then the tournament system. The BWF World Tour is tiered from Super 100 to Super 1000, alongside the World Championships and team events such as the Thomas Cup, Uber Cup and Sudirman Cup. Most of these events take place in Asia, where broadcast rights are largest and sponsors most generous. Denmark is one of the few European countries with enough weight to keep a Super 1000 event on the calendar — and holding that slot is a permanent commercial fight, not a given.

Then the academies. Denmark runs a centralised model: young talents from provincial clubs move to the national centre in Brøndby at the appropriate age, live in dormitories, train together. The advantage is a shared standard. The disadvantage is that when everyone trains on the same syllabus from the age of fourteen, the room for idiosyncratic styles narrows.

I once discussed this with a veteran scout I deeply respect. He said something I wrote down: "You can model technique. You cannot model whether a fourteen-year-old can survive a Norwegian winter."

That is another way of saying the same thing. In football, transfer models tend to overvalue young players with striking metrics and undervalue what I call dressing-room chemistry: the ability to fit in, to endure loneliness, to accept six months without minutes. Badminton has an equivalent: a youngster may have top-of-age-group footwork and swing speed, but if he cannot stand training alone in a hall at six in the morning in a country that is not his own, every metric is worthless.

I know this not only from theory.

The blind spot in the analysis room

Last year I consulted on data for an international club competition. During the transfer window I recommended a Danish club sign a young defensive midfielder I had found through data: 11.8 km covered per match and 6.2 ball recoveries per match, both near the top of the range. A veteran scout warned me about cultural integration. I ignored the warning.

Four months later he was struck from the squad list.

I tell this story not for the sake of self-flagellation, but to point at what I think is a structural blind spot in sports analytics: our models are very good at predicting what can be measured, and nearly useless against what cannot be measured — yet in sport, the second group usually decides the first.

In badminton that blind spot has a more specific shape. I can measure distance covered, reaction speed, serve success rate, win rate in long rallies, and dozens more. I cannot measure what it is like for a player who has spent ten months rehabilitating an injury and is now competing in a body he no longer fully trusts. I cannot measure a player in the final match before retirement. I cannot measure a player who has just become a father.

None of that invalidates data. It only puts data in its place.

Signal for the next round

Back to Brøndby. The session ended at 9:15. Antonsen walked off, sat on the bench, drank water. I looked at my spreadsheet: a gentle bend at minute forty, then flat again around minute fifty. He had found a new rhythm after fatigue rather than collapsing.

Numbers only tell the past; badminton lives in the future. What I saw that morning was not a prettier smash but the ability to reproduce a movement while the body objects. That is the kind of signal I trust more than any flashy figure.

If I had to write down three signals to track in the next cycle, they would be these: first, how rally-extenders hold their rhythm in the third game, because that is where their tactical edge turns into points. Second, how young players adapt to a dense tournament calendar — scheduling is the most undervalued variable in every model. Third, how the debate over the scoring format develops, because if the rules change, my entire frame of reference has to be rebuilt.

I do not believe in luck. I believe in what luck conceals. And what luck conceals in elite badminton is usually a very small, very boring detail, repeated thousands of times inside an empty hall at 7:40 in the morning.

You see the smash. I see second 3.9.

We are both right. We are just standing on opposite sides of the same number.