0.2 Seconds at San Siro: When Football Data Lies Without Warning

On an afternoon in March 2026 in Milan, I opened the movement-data file for twenty Serie A matches from the 2026-17 season and nearly believed it. Every cell was full. Not one error value, not one red flag, not one empty row. The dataset was so clean that it invited you to nod and move straight to conclusions.

Only one detail made me stop. AC Milan's expected-goals figure — xG — at their home ground, San Siro, was 1.85, while away it was just 1.02, nearly double. Yet the actual number of goals scored in both situations was level. A team cannot be twice as sharp at home and twice as blunt away within a single season, unless something in the measurement is telling a false story.

Modern football analysis lives on data. Every Serie A or Premier League match generates millions of positioning points per second, plus event data, physiological data and multi-angle footage. But most users only look at the final number: xG, PPDA, successful pressing actions, distance covered. Very few ask how that number was measured, under what conditions, and what happens if the measuring device is wrong.

At AC Milan in the 2026-17 season, the board gave me an unglamorous task: to audit the movement dataset of twenty matches. Nobody goes on television for rechecking a sensor. But I learned that in elite sport, most mistakes do not come from poor analysis, but from input data that was already broken before anyone analysed it.

Across a long season, data becomes habit. The coaching staff receive a report every week, skim it, highlight a few standout metrics and move on to preparing the next match. That very regularity creates a blind spot: when everything looks fine, nobody has time to doubt the foundation. But a season is a long chain of small signals, and physical or tactical signals only become a trend when someone is willing to look through the dust of habit.

I rewound the footage match by match. After many hours, the culprit appeared: a sensor in the south-west corner of the stadium was lagging by 0.2 seconds. Two-tenths of a second, shorter than a human blink. But when the system used that sensor to locate the ball during build-up from the goalkeeper, the delay caused every long pass by Milan to be assigned the wrong destination. The system raised no error. It quietly assigned the ball to another player, at another position, then calculated xG on that false data.

The result was a distorted but entirely plausible picture. On the numbers, Milan appeared as a team with a ferocious wide attack at home and a dead end away, while in reality they performed almost identically in both. The worst mistake in analysis is not an absurd number that makes people suspicious at once; the worst mistake is a plausible number born from a broken system. It does not shout. It whispers, and because it whispers nobody checks.

I wrote a fourteen-page internal report. In it I set out the error mechanism, its scope, the list of affected metrics, and a proposal for regular equipment calibration before every matchday. Head coach Vincenzo Montella used the findings to adjust the way the team circulated the ball to the right flank. In the last eight matches, Milan won five and secured a Europa League place.

This story is usually told as a victory for data. I tell it differently: it was a victory for data verification. Had I trusted that elegant spreadsheet, I would have recommended Milan pour everything into wide attacks at home and change the entire team structure when playing away — a false conclusion built on rotten foundations. Data only tells part of the story; the rest lies in whether people know how to listen. And sometimes what needs listening to is the silence of a lagging sensor.

Suppose I had not checked. The conclusion drawn would have been: Milan create far better chances at San Siro, so away from home they should sit deeper, cede territory and wait to counter. The coaching staff would have adjusted the defensive line height accordingly. But if the team's true attacking capacity was level in both settings, then dropping deeper away would only remove the possession phases they could have created. A wrong number does not merely ruin a report; it changes how a team walks onto the pitch. That is why I never quote a metric I have not cross-checked against at least two sources.

The cost of data infrastructure at a big club is far from trivial. Optical tracking systems, sensors in the ball, GPS devices in shirts, analysis software, engineering staff — all form a chain where a single misaligned link distorts the entire output. Yet that chain never cries for help. A camera out of calibration still transmits data steadily. An algorithm mislabelling players still produces fluent results. The cost of a silent error is always higher than the cost of a loud one, because a loud error gets fixed while a silent error gets trusted.

The process I proposed afterwards had three steps. First, cross-check at least two independent sources for every key metric. Second, state the measurement conditions beside each number, so a later reader knows the circumstances in which it was produced. Third, test spatial plausibility against footage — if the data says a player ran thirty-five km/h in the eighty-fifth minute, watch the footage again before believing it. None of those steps needs cutting-edge technology. They need patience, something modern sport increasingly has little time for.

There is a second story I always tell alongside the first. In 2026, thanks to the previous year's internal report, Sky Sport Italia invited me as a specialist commentator at the World Cup in Russia. In the Germany–South Korea match, on the seventieth minute, I posted on Twitter that Germany's back line was pushing up an average of 68 metres, that their pressing had failed seventeen times, and that South Korea already had twelve counter-attacks; if they did not drop the block, the goal would come from an aerial situation. On the 90+3 minute, Kim Young-gwon scored exactly to that script.

Thousands of accounts mocked me for turning emotion into arithmetic. But Gazzetta dello Sport republished the piece, alongside the distorted trapezoid diagram I had drawn of Germany's defence. The lesson I drew was not that the data was right. The lesson was that a number must be translated into a spatial image before readers remember it. I stopped writing “pushing up 68 metres” and began writing “the zipper has burst open to the valve box”. The same data, two ways of telling it, two very different levels of persuasion.

Both stories taught me one principle: every collapse has a precondition, it is just that few people bother to look beforehand. In Milan's case, the precondition was a lagging sensor. In Germany's case, the precondition was a back line that refused to drop after losing the ball. No collapse is sudden. There are only signs ignored because they were not yet loud enough to become a headline.

This is also why I always make time to replay the team-radio clips. The tone of an engineer's voice, hesitation in a negotiation, a long pause before someone answers — none of that appears on any dashboard, yet it often signals a technical problem before any sensor registers it. Data and people do not replace each other; they cross-check each other.

This principle reaches beyond match data. It applies to the transfer market too. A contract only looks good on paper until someone tries to fit it into a running system. A striker who scores twenty goals in another league can become a poor investment if the data on him was collected in a completely different playing system. People pay for the numbers, but the system is what decides whether those numbers can repeat.

The counter-intuitive point lies here. Sport has built a culture of worshipping the dashboard. But a dashboard is only as good as the sensor behind it. And the most dangerous state of data is not wrong data, but empty data.

An empty feed looks clean. No red cells, no error values, no warnings. It wears the most trustworthy appearance possible, then waits for someone confident enough to read conclusions out of thin air. I have just read an analysis report presented with extreme care: nine full dimensions of analysis, full tables, full section headings. But every content cell said “insufficient information”. A perfect skeleton with no body. The frightening part is that it still looked credible.

If someone reads that report and hastily fills the empty cells with guesswork, they will produce a conclusion that sounds highly professional but rests on nothing. That is the kind of mistake I fear most: a mistake with good form. Every tracking number should be placed on an operating table, not on an altar. And when there is no data, the most honest answer is to admit there is no data, not to perform convincingly.

I will carry this principle into the next matchday. Before quoting any metric, I will ask where it was measured, with what device, under what conditions. If a feed goes silent, I will not fill it with guesswork. Football does not reward empty confidence; it rewards those willing to check again. And the Germans that year forgot that football never forgives complacency — a lesson still fully valid for anyone holding a spreadsheet that looks too good.