The Null Result: When Football's Analytics Machine Learns to Say 'I Don't Know'

On a February morning, while Seoul was still curled up in the dry cold of winter, I opened a report file nine sections long. It had every heading, every table, every analytical frame any data room could dream of: tactical analysis, club finance, results and public-opinion cycles, league landscape, rules compliance, the dressing room, the risk profile, the media narrative, the industry transmission chain. Nine sections, not a single cell missing. Every table had a column for conclusion, a column for evidence, a column for risk level.

But as I read line by line, I noticed something strange: every conclusion was identical. Insufficient information to assess. Insufficient information. N/A. A vast analytical machine, built to dissect a team's matches, money, power and emotion, ended up saying a single sentence: I don't know.

What is worth noting is that the machine was not broken. It ran correctly. It was simply honest.

Over more than thirty years of watching football from Madrid to Seoul, I have read thousands of reports. I have seen forty-seven-page reports that a coaching staff read only on the summary page. I have seen data tables printed out and left on meeting-room chairs. But I had never seen a report dare to say I don't know so completely. And I had never been so worried.

A system designed to always answer will always find an answer — and that is the exact definition of fabrication.

The machine has a technical name. People call it a pipeline — a two-stage analytical duct. Stage one decomposes a source article into information points: title, source, type, core arguments, entities mentioned, author's stance. Stage two takes those points as its foundation and builds nine dimensions of deep analysis: tactics, finance, results, league context, rules, dressing room, risk, media, and the transmission chain of the industry.

It sounds reasonable. The problem lies here: when stage one returns an empty file — no title, no source, no entities, no arguments — stage two still has to run. It still has to output all nine sections. It still has to fill every cell. Because the frame was designed to always have an answer.

I know that feeling. In 2026, at the age of forty-seven, I threw myself into a tactical-decoding project for FC Seoul. I analysed all thirty-eight matches of the K League Classic season, building a database of the gaps between the lines. Hwang Sun-hong's team generated only 1.7 shots per match on average from central areas, the lowest in the league. I wrote a forty-seven-page report. The coaching staff looked only at the one-page summary.

That night I sat down and understood one thing: the problem was not that my data was wrong. The problem was that I believed a thicker report was a truer one. I reduced forty-seven pages to a geometric diagram of five boxes. From then on, every piece I wrote had to contain at least one spatial model.

That story taught me something I still carry: when a system is forced to always have an answer, it will start inventing answers.

Today, the analytics pipeline is no longer the concern of a single bookish journalist. It is infrastructure. Big European clubs hire entire data departments. Analytics firms sell reports on subscription to scouts, to coaches, to sports investment funds. A single match now generates millions of data points: every player's coordinates each second, speed, distance, touches per spatial zone, goal probability.

But there is a paradox few are willing to name. The more data there is, the less people accept silence. A data room cannot send its president a blank file. An analytics firm cannot sell a client a report saying we know nothing yet. So when the input is empty, the machine still has to spit something out. And that something, in most cases, is a very plausible story.

That is when I realised: the day I understood that data does not judge, it only exposes. And when it exposes an empty file, it is us — the operators of the machine — who are exposed.

Picture those nine dimensions as nine rooms in a building. Each room has a questionnaire on the wall. The tactics room asks: what shape does the team play, how is it executed, do the personnel fit, what is the key data. The finance room asks: where does revenue come from, what is the wage bill, what is net debt, is the transfer deal overpriced. The results room asks: where does the table position sit against expectations, what is recent form, who is under public pressure. The league-context room asks: where does this club sit in the food chain, how do resources compare with direct rivals, is there a risk of bigger clubs poaching core players.

Then the rules room, the dressing-room room, the risk room, the media room, the industry-transmission room. Nine rooms, nine questionnaires. And at the end of each questionnaire, there is always a waiting line: conclusion.

That empty line is the most dangerous thing in the entire building. It is not waiting for data. It is waiting for a sentence.

I once witnessed this from inside a meeting room in Nizhny Novgorod, in June 2026. South Korea lost 0-1 to Sweden. The goal came from an Andreas Granqvist penalty. In the stands, people talked about stoppage time, about bad luck, about a match the national team deserved to draw. But when I rewatched the footage, I saw something entirely different. Shin Tae-yong set up a 3-4-3, and Son Heung-min was completely isolated up top, receiving only nine passes across ninety minutes.

After the match, I dived back into all six Asian qualifying matches. The problem was not the game plan. The problem was the average forty-eight-metre gap between midfield and attack whenever the team had to press. I wrote a two-hundred-page note. Then I published only a short piece, and criticised myself for a lack of execution.

South Korea 2026: we did not lose on the pitch, we lost from the moment we believed we had won. That is a late conclusion. And I still ask myself: if I had had an automated analytics machine then, would it have dared to tell me the data was insufficient, or would it have spat out a fluent story about bad luck?

I think the answer lies here: a machine does not know fear. It does not fear silence. It does not fear saying there is not enough. And that is why it is dangerous.

Let us walk into each room and see what happens when the questionnaire meets an empty input.

The tactics room. The first question is about sophistication. With no shape, no style, no concrete tactical claim, every comparison is meaningless. The operator has two choices. One: write insufficient information. Two: imagine a team, assign it a 4-3-3, and analyse that shape as if it were real. The second choice is always more attractive, because it produces a readable report.

The finance room. No club, no financial statement, no named deal. So what is the premium rate? Unknowable. But if the machine wants, it can take an industry-average figure, attach it to an imaginary deal, and produce a conclusion that sounds highly professional about financial risk.

The results room. No table, no form curve, no fixture list. No trajectory can be drawn. But the machine can write about public pressure on an unnamed manager, because public pressure is something everyone believes exists.

And so on, nine rooms, nine choices between truth and story. The frightening part is that both choices look identical on paper. An honest empty report and a skilfully fabricated one share the same format, the same font, the same table layout. A reader can only tell them apart if they read to the very end, and notice that every conclusion is saying the same thing.

In analytics, this is called a null result. It is not a failure. It is a finding. When a medical experiment returns a null result, that is still data: it shows the hypothesis does not hold, or the sample is too small, or the measurement method is flawed. But in football, where everything is measured in fan emotion and sponsor money, a null result has almost no place.

No one buys a ticket to watch a null result.

That is why I call that nine-section report a rare find. It is one of the very few documents that dared to preserve the emptiness of its input instead of filling it. And precisely by preserving it, it inadvertently exposed a flaw far larger than a single faulty article.

That flaw sits in stage one. If the decomposition stage cannot read the source article — because the file was not transmitted correctly, because the format broke, because the language-processing step failed — then stage two has no way of knowing, unless someone installed a validation gate. And a validation gate is usually not installed, because it produces no visible value. It only blocks. And in an industry that prizes speed, blocking is seen as slowing down.

I have seen the consequences of missing validation gates in many places, not only football. In data rooms, people usually check whether the output looks reasonable, rarely whether the input actually exists. A report that looks reasonable goes straight to the decision-maker. And the decision-maker, under time pressure, acts on it.

This leads to a bigger question: what happens if that machine is not an exercise, but a product running in the real world?

Consider the transmission chain. A null report replicated at scale creates a layer of false information infrastructure. Scouts read it to pick players. Coaches read it to plan. Investment funds read it to value assets. Each layer believes the layer below has checked, and no one checks. The result is a decision system built on facts that never existed.

And here I must be blunt about an old worry of mine: what I fear most is not error, but a wrong model. Error is the business of one number. A wrong model is the business of an entire way of seeing. A fabricating machine does not produce error. It produces a wrong model, then replicates it everywhere until no one remembers where it started.

There is a deeper layer few want to look at. Data in modern football does not only serve clubs. It flows down into the betting market. Bookmakers pay to obtain live data, positional data, tempo data. They buy being one step faster than the public. That is the darkest side effect of sports digitisation: the same analytics pipeline feeds clubs and feeds the betting ledgers too.

When an empty report is pumped full of story, the ultimate loser is not the club. They have lawyers and their own data rooms. The loser is the reader who trusts the number presented as fact. And that person, usually an ordinary fan, has no way of knowing that behind the number lies a blank file.

Let me tell another story, to show the problem is not about new technology. Years ago, on a trip with a mid-table European club, I sat in the analysis room after a defeat. The coach opened his laptop and pointed at a distance-run chart. He said: we ran more than them. I looked at the chart, then at the footage. True, they ran more. But which direction were they running?

That is the question raw data cannot answer. Distance run is a neutral number. It does not distinguish between running to close a gap and running to chase a ball that has already passed you. And precisely because it is neutral, it is easily turned into a moral story: we tried harder.

I have seen gegenpressing turned into exactly that story. For a time, high pressing was hailed as a universal key. Then mid-table teams learned to counter it with physicality. They turned football into athletics: run, run, and run until the opponent's battery dies. Gegenpressing was decoded, not by a tactical genius, but by a distance-run chart handed to a coach who knew how to read it.

A tactical system only survives until it meets a larger system. High pressing is a system. The physicality of mid-table teams is a larger system in a very concrete sense: it does not need sophistication, it only needs endurance.

And here I want to talk about the transfer market, where a null result is most expensive. A transfer does not buy a player. It buys a probability of success. When a club pays one hundred million for a striker, it is not paying for what he has done. It is paying for the probability that he will keep doing the same for four years, in a different league, under a different coach, in a different system.

And that probability is computed from data. If the underlying data is contaminated, if an empty report has been pumped full, then that probability is a fake number presented as fact. An entire market can be mispriced because of a flaw in the decomposition stage.

There is one kind of fee I consider more toxic than a transfer fee: the signing fee for a free agent. It is often structured as a payment to the player and the agent, not to the selling club. So it slips out of the core scrutiny of financial fair play rules. On the books, the deal looks cheap. In reality, it can cost more than an ordinary transfer, and no one can verify it because it never passes through the front door.

This is a disguised null result: a deal where most of the data sits outside the field of vision. And when data sits outside the field of vision, a story replaces it. A story is always available. A story is always free.

So what is to be done?

First, accept that a null result is a valid output. A mature analytics system is one that can say I don't know without collapsing. It must have a gate at stage one: if there is no title, no source, no entity, no argument, then stop. Do not run on. Do not output nine sections. Return a single line: invalid input, re-run stage one.

It sounds simple. But in practice, installing such a gate demands something harder than technique: organisational courage. Someone must dare to tell their superiors that today there is no report, because the input does not exist. And in an industry where silence is treated as failure, that person usually pays a price.

Second, distinguish thickness from truth. Forty-seven pages are not truer than a five-box diagram. I learned this through my own mistake at FC Seoul in 2026. A long report can be just an empty report stretched out. Length is not evidence of depth. It is often evidence of indecision.

Third, question the origin of every number. Where does this number come from? Who measured it? Under what conditions? If you cannot answer those three questions, the number is just a story written in digits.

And here I want to be a little contrarian. Most people think the biggest risk of football analytics is wrong analysis. I think the biggest risk is correct analysis that answers the wrong question. A model can be technically flawless and utterly useless for a decision, because it measures something no one needs measured.

In the 2026 match where South Korea lost to Sweden, there were plenty of correct numbers. Correct passes. Correct touches. Correct shots. But the most important number — the forty-eight-metre gap between midfield and attack when pressing — was in no standard statistics table. It only appeared when I measured it by hand, frame by frame, after the match was over.

That is why I do not trust the statistics handed to the press right after the final whistle. They are correct, but they often answer questions no one actually asked. And in the gap between the real question and the ready-made answer, a story grows.

There is one more blind spot I want to name: the blind spot of execution. An analysis can be right, a plan can be reasonable, but if it is not converted into action on the pitch, it is only a document. In 2026, I had a correct analysis and a reasonable plan. I failed to execute it. I wrote two hundred pages of notes and published a short piece. The gap between two hundred pages and a short piece is my own execution blind spot.

And that blind spot, replicated at the scale of a machine, produces a generation of reports technically correct but dead in action. They sit on hard drives. They never step onto the grass.

I have thought a great deal about why that machine was so honest. And I reached a conclusion that may discomfort many: it was honest not because it is good, but because its input was so empty there was nothing left to fabricate. It was honest by necessity. If the input had held even a scrap of truth, the machine could have begun to weave.

That means the machine's honesty is not a virtue. It is a consequence. And a consequence can be reversed by changing the input.

This is what I want to leave with those building analytics systems: do not design a machine to always answer. Design a machine that knows when to stay silent. Well-timed silence is the hardest feature to build, and the most valuable one.

In football, we often praise teams that know how to attack. But champions are usually teams that know when not to attack. A good defence is not one that blocks every ball. It is one that knows which ball need not be blocked. The same principle applies to analytics: a good system is not one that answers every question. It is one that knows which question has no answer yet.

I return to that nine-section report. It has a section called the risk profile. And in that profile, the only risk flagged is the risk of the analysis itself. It is a rare moment when a machine points at itself. It says the greatest risk is not in any team, not in any manager, but in the very pipeline that produced it.

I think that is the truest thing in the entire document.

There is a question I still carry from Nizhny Novgorod to Seoul, from 2026 to today: if an analytics system does not dare to say I don't know, who is it serving?

The answer, I fear, is that it serves itself. It serves the need to answer, to present, to be trusted. It serves the feeling that everything can be measured, and everything measured can be understood, and everything understood can be controlled. That is a very human belief. And a very dangerous one.

A model does not control a match. It only describes a match that has already happened. And if it describes wrongly, it does not merely misdescribe one match. It teaches us to see all the following matches wrongly.

That is why I tell younger colleagues: learn to write a null result before you learn to write a conclusion. Anyone can write a conclusion. A null result demands honesty toward the data and humility toward yourself.

And in this big-tournament season, when the whole world is swept up in flags and stories, when every match is dissected frame by frame, I want to restate one simple thing. Data does not tell stories. It merely exists. We are the ones who tell stories. And the best storyteller is not the one who tells it best, but the one who knows when to stop.

The day I understood that data does not judge, it only exposes, I thought it was a liberation. Now I think it is a responsibility. Because if data only exposes, then whoever holds it is responsible for what is exposed — and for what is concealed.

I still keep that nine-section report in a separate folder, not because it has analytical value, but because it is a reminder. Every time I open it, I see a vast machine saying I don't know. And every time, I ask myself whether I have the courage to do the same, when an entire industry is waiting for me to give an answer.

That is the question I leave for the next match. Not for the team. For myself.