Empty Data: The Fragile Line Between Football Analysis and Fabrication

In the packed press room of Stade Vélodrome in the autumn of 2026, I sat in the third row — the only woman in the room. Marseille had just lost 1-3 to PSG. When I asked about the gap between the midfield and the left flank, a male reporter sneered: "Do women watch football with their emotions?" I did not answer with words. I opened my hand-drawn movement chart of all 22 players, reconstructed from video tape, and pointed to seven occasions when the left flank was left open. The room went silent. The following week, I had my first tactical analysis column.

Twenty-five years later, I sit in an office six thousand kilometres from Marseille, looking at a match analysis report sent by a data company. The report had every field populated: match name, possession figures, pass counts, heat maps, expected-goals charts. But when I traced it back to its source, everything was empty. No match was recorded in the database. No player was identified. The report looked flawless in form, and was entirely empty in substance.

That moment reshaped how I see the whole modern football analytics industry. The greatest danger is not bad data. Bad data can still be corrected, because it has a source to cross-check against. The danger is empty data presented as though it were real data — a conclusion with no origin, wrapped in the language of precision.

To understand why this happens, look at the football data supply chain that has formed over the past two decades. Every professional match today is encoded into discrete events: each pass, each shot, each duel is assigned a coordinate, a player, a timestamp. Companies such as Opta, StatsBomb and Wyscout collect millions of events each week, then resell them to clubs, broadcasters, bookmakers and content producers like me.

Based on my experience watching matches across more than thirty-seven years in the profession, I can see the shift clearly. Years ago, reporters like us drew charts by hand from video tape, the way I did at Vélodrome in 2026. Today, most of that process is automated. This is better for speed and coverage. But it creates a new weakness few notice: when the speed of production overtakes the speed of verification, the gap between the two gets filled with whatever sounds plausible.

Football data has become a commodity. And like any commodity, it has an incentive to generate returns even when quality is not guaranteed. A report with every field populated will be read, shared and cited — regardless of what is inside it. Form has become a substitute signal for content.

That is why I apply a principle young colleagues often find extreme: never trust a conclusion before you can verify its origin. A metric without a source is not data. It is an assumption written in the format of data.

I have also seen the darker side of this supply chain. Live data supplied to betting companies is the darkest side effect of the digitisation of sport. The same dataset sold to clubs to improve tactics is also sold to bookmakers to price risk. Where the data flows is not controlled, and ordinary fans do not know that every click on a match-tracking app can become raw material for a betting model.

In 2026, when José Mourinho's Porto beat Monaco 3-0 in the Champions League final, I watched the match eleven times over three days. I did not believe in luck. I traced a number few noticed: Porto had only 43% possession, yet created five clear scoring chances, while Monaco had just one. When people look at Porto 2026 and see a miracle, I see an equation waiting to be solved.

The core point is that Mourinho did not try to control the ball. He controlled the space his opponent was allowed to use the ball in. It is a problem of geometry: forcing the opponent to pass into areas where traps had already been set. A 43% possession figure says nothing about a team's quality. It only says who is holding the ball. Who controls space is the right question.

The Porto squad of 2026 was quickly dismantled after the title. Deco went to Barcelona, Ricardo Carvalho and Paulo Ferreira went to Chelsea, Maniche left the club. Their success became the opening form of another talent raid, and the cycle repeats at many smaller clubs whenever they reach a peak. The system produces the achievement, but the transfer market dismantles that very system.

Today we have more sophisticated metrics to measure tactical quality: expected goals, expected assists, passes allowed per defensive action. These tools are powerful — when they have a clear origin. Problems arise when someone cites an expected-goals figure without naming the model that produced it. Each model yields a different number, because each model defines a "clear chance" in its own way. Citing a metric without naming the model is like citing a report without naming the author.

A metric without a source is not data — it is a belief dressed up in numbers.

I remember a March evening, re-analysing a match in which the away side was praised for "superior possession". I checked the passing map and realised most of those passes happened in the defensive three-quarters of their own half — possession that created no threat. The metric said one thing; the geometry of the match said another. That is the most dangerous kind of bad data: data that is numerically correct but placed inside a false story.

VAR offers a similar lesson. Video assistant referee technology arrived to reduce errors, but it also produced a consequence few discuss: the final decision still depends on how a human interprets an image. For the same incident, three referees can reach three different conclusions. Technology does not remove judgement. It only moves judgement from the pitch into a closed room.

And here is the most important thing: modern football analytics operates on the same principle. We believe that adding more data will remove ambiguity. But data can only answer the questions it was designed to answer. Other questions — about intent, about context, about what was never logged — remain beyond the reach of any spreadsheet.

The transfer market is where this problem is most visible. Transfers are a market of hope, and hope rarely follows valuation. A striker with a high expected-goals figure in a smaller league can collapse entirely when he moves to a harsher league, because the model does not measure decision-making speed under pressure, the quality of surrounding team-mates, or cultural adaptability. Those variables exist. They simply are not entered into the model.

I once spoke with a scout about a highly valued young forward. The player had every good metric. But when I asked how he reacted to being substituted in the 60th minute, no one could answer. That is a data gap, and that gap is worth more than every beautiful number in the report.

The counter-intuitive view lies here: we are worrying about the wrong thing. The whole industry devotes resources to detecting bad data, while the bigger threat is empty data produced at industrial speed. Once the content-production process is optimised for volume, filling gaps with plausible material becomes the default, not the exception.

I once sat in a meeting where someone proposed "supplementing the missing metrics with estimates". The idea sounded harmless. But estimates born of nothing are not analysis. They are fabrication shaped like analysis. And when placed in a neatly titled report, the reader has no way to distinguish them from the truth.

Fate is not decided in the press room — but it begins to be written there. I believe this after more than three decades in the profession, and I believe its expanded version too: the fate of an analysis is decided at the data-entry stage, not at the conclusion stage. If the input is empty, every conclusion that follows is an illusion. No model, however sophisticated, can rescue an input that does not exist.

There is another variable the analytics industry often ignores: emotion. Emotion is the variable nobody enters, and so it is usually treated as noise. I disagree. A missed penalty in the 88th minute rarely concerns pure technique. It concerns the psychological state of the taker, the pressure of the stands, the memory of previous failures. These variables can be measured — through heart rate, through decision time, through individual head-to-head history. We choose not to measure them, then tell ourselves that what is unmeasured does not exist.

Collapse is not the end of the tunnel. It is the largest dataset life provides. After every defeat, I do not seek comfort. I trace the break points in tactics, fitness and psychology. But I can only do that when the data of the defeat is real. A defeat recorded honestly is precious material. A defeat analysed with empty data produces only a beautiful and meaningless story.

Football analytics will not advance by adding more metrics. It will advance when we dare to say "I do not know" in the face of a data gap, instead of filling it with a plausible-sounding answer. Football is a verifiable system of equations, and an equation is only trustworthy when every one of its variables is real. The next match will be the test. Read the analysis, then ask yourself: where is the origin of each number.