Empty Data, Full Reports: The Silent Flaw Inside Professional Football Analytics

Empty Data, Full Reports: The Silent Flaw Inside Professional Football Analytics

In 2026, at the FC Seoul training headquarters, I placed a forty-seven-page report on the meeting table. Inside were all thirty-eight matches of the K League Classic season, encoded into spatial cells: reception points, the distance between the two centre-backs when the team lost the ball, the angle of a central midfielder's turn before releasing a pass. The conclusion sat on page three. Hwang Sun-hong's side generated an average of 1.7 shots per match from the central channel, the lowest in the league.

The coaching staff flipped to the one-page summary, read it for forty seconds, nodded, and moved on. That night I sat alone in the office, folded the report into five small squares, and asked myself a question I have carried for nearly a decade: if I cannot transmit a correct finding in forty seconds, who is my analysis system actually serving?

A year later the question returned from another direction. In June 2026, in Nizhny Novgorod, I sat in the fourteenth row and recorded a single number across the entire first half of South Korea against Sweden: nine. Nine passes. That was the sum total of Son Heung-min's receptions in the opening forty-five minutes. The staff had drawn a 3-4-3 on paper, but on grass the average distance between midfield and attack stretched to forty-eight metres whenever the team was forced to press. The system did not collapse for lack of players. It collapsed for lack of room to breathe.

I wrote two hundred pages of notes on that match. I published one short piece. And I criticised myself for not having the courage to say aloud what the data had already screamed.

Context: the data factory of modern football

Today most European clubs and many K League sides no longer wait for an analyst to rewatch the footage. They buy data. Providers such as Stats Perform, Opta and Wyscout deliver thousands of events per match, every pass and every duel tagged with coordinates within seconds of the ball rolling. Platforms such as Hudl and Kitman Labs turn that raw feed into automated reports. The system reads the input, applies a pre-built analytical frame, and emits a document that looks complete: a title, tables, conclusions, recommendations.

People praise this maturity as a revolution, and it is genuinely useful. But across forty years of watching this industry, I have learned one simple thing: the more automated a system becomes, the higher the cost of a single error. Not because machines are dumber than people, but because we tend to trust anything that is neatly formatted.

A European club signs with a data provider and receives hundreds of reports in a single season. The readers are not data scientists. They are coaches, sporting directors, scouts. They have little time. They need answers. And when a document is laid out cleanly, they assume it is correct.

Analysis: the architecture of a report with nothing to say

Picture a two-stage analytical pipeline. The first stage reads the source document and breaks it into units of fact: which team, which player, which minute, what score. The second stage takes those units, applies nine professional analytical frames, and produces conclusions on tactics, finance, personnel, risk and media.

Here is what keeps me awake. The second stage can run smoothly even when the first stage returns a blank page. No title, no source, no information points, no entity identified. The first stage has failed. Yet the second stage never raises an error. It still builds the frame, still fills every cell, and for each missing field it writes a polite slash: insufficient information.

The result is a document thousands of words long, with clear chapters, a risk matrix, a transmission diagram, an information-value scorecard. It looks exactly like a professional report. Read closely, it says nothing at all.

This is the crux: a perfectly formatted system can make us forget that it never had anything to analyse.

In football, this failure does not live in a laboratory. It lives on the pitch, in the meeting room, and in the notebook of the punter. A club receives an automated scouting report on a South American striker. Five pages, a radar chart, a percentile ranking against positional peers. But if the input covers only three matches, and all three came against weak opponents, every number is technically correct and practically meaningless. That is the moment a ten-million-euro transfer decision is made on a document with no content.

I once wrote that a transfer does not buy a player, it buys a probability of success. But that probability is only worth something when the model is right. And the model is only right when the input data exists.

There is a subtler defect, and it sits inside the structure of many automated reports. One field asks the reader to judge the source based on the information above. But if there is no information above, the instruction becomes a loop with no exit. In football terms, it is a scouting report that says: rate this player based on the figures listed above, while the entire report contains no figures. The careless reader nods. The careful reader realises they have just read an empty circle.

The counter-intuitive angle: the fear is not that the machine is wrong

Most debate about artificial intelligence in football circles around whether machines will replace scouts. I think that is the wrong question. The real fear is not a model making a bad prediction. The real fear is a broken system that keeps emitting the signal of a working one.

There is a technical term for this: silent degradation. When a data pipeline hits a fault, instead of stopping and screaming, it returns a default value that looks harmless. The classification field returns unclassified. The source field returns not applicable. The entity field stays empty. Piece by piece, the system hides its own failure.

That is more dangerous than an obvious error, because an obvious error forces a fix. An error disguised as ordinary data spreads quietly. It flows into scouting reports, into player-valuation models, into the probability tables that betting firms sell to fans. And at the end of that flow, someone sitting in front of a screen in Hanoi or Seoul believes they are reading an objective statistic.

That is why I regard the supply of live data to betting companies as the darkest side effect of sports digitalisation. Not because betting is evil, but because it converts every model error into cash. When a data pipeline breaks and nobody notices, the loss does not stop at a meaningless report. It becomes a loss borne by whoever trusted the number.

The price of formal perfection

I have spent hundreds of hours redrawing match space. I believe in geometry. A triangle between three midfielders, a forty-metre gap between two lines, a seventy-degree angle as a full-back turns — that is the real language of football. But precisely because I believe in geometry, I also know that a beautiful drawing is not the same as a correct one. A neatly formatted table can conceal the fact that it was built from nothing.

On an empty ground, I can hear a defender breathing and a tactical scheme cracking. Inside an automated report, I hear nothing. That is exactly the problem. A good analyst stays silent when the evidence is not there. A system programmed to always answer does not know how to stay silent.

The day I realised that data does not judge, it only exposes, I understood one more thing: empty data exposes something too. It shows who built the system, who tested it, and who decided that a report looking full is better than an honest report admitting it knows nothing.

What I fear most is not error but a wrong model. Errors can be fixed. A wrong model requires someone willing to say that the entire analytical frame is standing on thin air. A data revolution is not measured by the number of report pages, but by the number of times a system dares to admit it does not know.

What to verify next

Next season, when you open a statistical table, ask one question: how many matches produced this number, and were those matches genuinely comparable? If the answer is unclear, treat it as a slash still waiting to be filled in.

For an analyst, the most important skill is not reading data. The most important skill is recognising when the data has not arrived. A tactical system only survives until it meets a bigger system. A data system is the same. It is only trustworthy when it meets an empty input and knows how to stay silent.