Player Valuation and the Dressing-Room Blind Spot: A Data Diary from Marseille

In August 2026, Marseille was hot enough that the ceiling fan in my office could not spin properly. I sat there with 1,204 shots recorded by hand on A4 paper, stacked by matchday. Opta had just released the first xG table for Ligue 1, and my whole office was excited as if they had found the holy grail. I was not. I needed to verify it myself before believing it.

Two weeks later, I had my answer. The correlation between xG and actual goals in the first half of the 2026-18 season was 0.84. Not 1.0, not 0.6, but 0.84 — enough for me to begin building my own striker valuation dataset, and enough for me to understand that the remaining 16 percent is where the real work begins. In the summer of 2026, I learned to trust something no one had yet named: xG.

That fact is the starting point of nearly a decade spent answering a seemingly simple question: what does the transfer market use to value a player, and what does it get wrong? The answer does not lie in fashionable numbers. It lies in the gap between the dataset and the dressing room.

What I have concluded after nearly ten years is this: transfer data models overvalue young potential and undervalue dressing-room chemistry — and that gap is where the market quietly loses money.

When I started as a transfer market administrator in Marseille, the job was still run by eye and by relationships. A scout would watch ten matches, write a report, mark three names, and the club president would decide on instinct. In the late 1990s, computers arrived. In the 2000s, event data per action became standard. By 2026, xG had entered the daily news cycle.

The problem was never the quantity of data. The problem was how people used data to value players. One Ligue 1 club bought a 21-year-old striker with 18 goals in the second division for 12 million euros, then was disappointed after eighteen months. Another club paid 4 million for a 28-year-old midfielder with no standout metrics, and that player became a pillar for three seasons. I saw both scenarios often enough to stop calling them luck.

The market had a wrong frame of reference about value. It believed a player's value sits inside the player. But as someone who works with data, I saw value sitting in the relationship between the player and the system around him. A player is a variable. The team is a function. For most of my life, as an observer, I am a constant — I only change the variable to see how the function responds. The player is a variable, the market is a function, but most of my life is a constant.

So I started from raw data. Not from providers' rankings, but from the shots, passes, and presses I counted myself. My belief in a metric must pass three checkpoints: sample size, confidence interval, and context. Without all three, the metric is just a pretty number.

After confirming that xG correlated at 0.84 with actual goals in the first half of 2026-18, I split Ligue 1 strikers into four groups: high xG and high goals, high xG but low goals, low xG but high goals, and low in both. The grouping was not statistically novel, but it gave me a valuation tool most colleagues lacked.

The second group — high xG, low goals — was the most undervalued by the market. The third group — low xG, high goals — was the most overvalued. In the winter of 2026, I submitted an internal report to Marseille's leadership: a striker in the second group was priced below his true value, and a striker in the third group was priced above his. The report did not make me famous. It only earned me a reputation for being slow, because I refused to conclude before I had enough data.

But I never cite a new metric without stating sample size, confidence interval, and match context. 1,204 shots was my sample size. The confidence interval around 0.84 is a number I still keep in my notebook. And the context — Ligue 1 in the first half of 2026-18 — is something I repeat whenever someone wants to extend the conclusion beyond that range.

From that dataset I drew a principle that later became a pillar of how I read the market: do not buy goals, buy the process that creates goals. The principle sounds simple, but in 2026 it ran against how clubs still operated. They bought goals because goals sell tickets. They did not buy xG because xG does not sell tickets. And precisely because of that, high-xG, low-goal players are usually the market's best bargains — provided you verify that this is not a sign of technical decline.

Thanks to my Marseille dataset, a sports newspaper invited me to contribute to the 2026 World Cup. I was 58, watching 64 matches and counting each team's PPDA. PPDA is the number of opponent passes allowed before each defensive action — the lower it is, the more aggressive the press.

In the semi-final between Croatia and England, Croatia allowed England only 8.2 passes before each defensive action, while England allowed Croatia 12.5. I wrote a prediction that Croatia would win through their pressing in extra time. They won 2-1. I did not shout in celebration. I reopened the spreadsheet to hunt for outliers — because when a prediction of mine is right, my first reflex is to check why.

After the tournament, a wave of articles praising PPDA swept the press. And I started to feel uneasy. Croatia won a tournament with low PPDA? Then PPDA is only a letter. If a low PPDA were sufficient to win, every aggressive pressing team would win. Reality says otherwise.

What I learned from the 2026 World Cup was not that PPDA is good or bad. It was that PPDA is only one part of a wider frame. A pressing midfield needs defenders who push high, a goalkeeper who can play out with his feet, and forwards who make runs to turn recovered balls into chances. If one link is missing, a low PPDA becomes a hole behind the midfield. I shifted from writing about matches through emotion to presenting PPDA, distance covered, and successful presses. The phrase "the team pressed better" appears only when the numbers truly support it.

In 2026, the pandemic restarted football in empty stadiums. My editor assigned me to cover the Bundesliga because I already had data experience from the 2026 World Cup. I was 60, sitting in Marseille, analysing 81 matches played in empty stadiums in the 2026-20 season.

The result made me reread the dataset three times. Home teams won only 26 percent of matches, compared with 43 percent before the pandemic. Seventeen percentage points. A gap large enough that it could not be noise.

I wrote the report "Empty Stands Kill Home Advantage." An empty stadium is the finest laboratory for a data obsessive — because it removes a huge variable that normally cannot be isolated: noise, crowd pressure, and referee influence in front of a crowd. With empty stands, we measure pure home advantage, free of interference.

A Ligue 2 club, Le Havre, used this report to lower the price for a young striker in strong form at home. He scored many goals in front of home fans, but when home and away splits were separated, the number dropped noticeably. Le Havre offered a lower fee, and the deal moved in that direction. It was the first time I saw my data report directly affect a player's price. I was not sure whether to be happy or worried.

From 2026, my writing began separating home and away metrics in every table. I remind readers not to trust pre-lockdown records when evaluating a player. A striker with 15 home goals and 3 away goals is a different player from one with 9 on each side — even though the totals are identical.

In 2026, the empty-stadium report reached Canal+, which sent me to Qatar for the 2026 World Cup at age 62. There I encountered a tactical phenomenon the experts praised endlessly: the inverted full-back.

Achraf Hakimi was cited for 142 sprints and 2.3 chances created per match. Those numbers were correct. But when I dug into positional data, I found the corridor behind him was empty for 34 percent of the time. That is an enormous figure. In elite football, leaving a flank empty for a third of the match is an invitation to attack.

Morocco stayed safe for most of the tournament, but not because of Hakimi. It was because their centre-backs ran above 31 km/h. They were fast enough to cover the space the inverted full-back left behind. I wrote a note warning that this tactical fashion only holds if the defence has enough speed. In the match against France, the opponent attacked Morocco's right flank relentlessly. The necessary condition was met, but the sufficient condition was not.

The lesson here is principled, and I apply it to every tactical trend. Before praising a new system, I list the necessary and sufficient conditions. The necessary condition is what the system requires to operate. The sufficient condition is what your squad must have so the system does not collapse against strong opponents. The inverted full-back needs speed at centre-back. Aggressive pressing needs a goalkeeper good with his feet. Possession needs midfielders who can pass through lines. Without the sufficient condition, a trend is just a fashion.

This is also why I do not worship trends. I believe in necessary and sufficient conditions. A tactic can look beautiful on the whiteboard, but if the squad lacks the right personnel, it becomes a burden. And the same applies to the transfer market: a player with beautiful metrics can fail at a new club if the system around him does not fit.

There is another variable raw data cannot capture: cash flow and financial-reporting pressure. During my years in Marseille, I watched many clubs shift from private ownership to shareholder models, and a few moved toward listing.

When a club goes public, it sells part of the fans' emotion. A football club's stock is not a producing asset; it is a memory asset. Shareholders are not buying future cash flow, they are buying belief in the club's story. That creates a paradox: long-term sporting decisions often conflict with the market's short-term profit expectations.

I once watched a club sell a promising young player mid-season to balance year-end accounts, even though the sporting department wanted to keep him one more season. Sportingly, it was a poor decision. Financially, it was a reasonable one. When financial-reporting pressure weighs on sporting decisions, the outcome is usually decided by the person in the boardroom, not the person on the bench.

This is the blind spot of transfer data models. They assume clubs act purely for sporting reasons. In reality, clubs act for financial reasons, and sporting data is only one variable in a larger objective function. A player valuation model that does not model the club's cash flow is missing an important variable.

Recently I have spent time watching esports, not because I love it more than football, but because it gives me a new laboratory. A mouse click on an esports screen carries the shape of a pass — both are decisions made in an extremely short window, and both can be modelled.

What I observe in esports is a trend football went through about twenty years earlier: professionalisation turning people into the product of an assembly line. When a discipline is organised as a professional league, training is digitised, and individual play is smoothed to fit an optimal framework. The player becomes an assembly-line product: input selected by metric, output measured by metric.

I find this worrying, not because of efficiency, but because of what is lost. Individual play — the thing that creates difference in decisive moments — often does not survive a process of absolute optimisation. It is the same problem I see in football academies: a 17-year-old striker with a distinctive style is often moulded to fit, and by 22 he has become a player like every other player. In esports, this process happens faster because player careers are shorter.

This is the part I must handle most carefully, because it runs against my own work. I have spent my life building data models. But I know their limits.

A metric that correlates with success does not mean it causes success. xG correlates with goals, but xG does not score. PPDA correlates with pressing, but PPDA is not pressing. Home win rate correlates with the crowd, but the crowd does not run for the players. Every time I see someone turn a correlation into a cause, I remember Croatia and the PPDA lesson.

The biggest blind spot of data analysis is not bad data. It is the belief that the data is complete. When a player valuation model relies on xG, progressive passes, and sprint counts, it ignores variables that cannot be measured by metrics: dressing-room integration, pressure tolerance, personal motivation, and the relationship with the manager. These variables appear in no dataset, yet they decide whether a transfer succeeds.

That is why I have an odd reflex: whenever a prediction of mine is right, I quietly hunt for the exceptions. When I am right, I do not rush to celebrate. I reopen the spreadsheet to find cases that contradict my conclusion. If I find no exceptions, I begin to doubt the sample size. If I find too many, I doubt the conclusion itself. Before drawing any conclusion, I write at least three hypotheses to explain the result, then try to refute them. Only the hypothesis that survives refutation deserves to enter an article.

This is also why I hold no arrogance toward analysts who work differently. I was doubted when I believed in xG in 2026. I know what it feels like to be called mad for being ahead of the crowd or off it. The difference between me and others is not that I am right. It is the method I use to verify what I believe. When someone presents a different view, I try to understand their frame before judging their conclusion.

Some matches are won on the pitch but lost on the data table — I choose the data table. But I choose it with humility, because I know my table also has blanks. A cancelled match is not lost points, it is a lost diary page. And every lost diary page is a chance to understand football more deeply that I do not get.

I am 66. At this age, I am old enough to know a number never tells a story unless we ask. And the question I am posing for the next cycle is not which metric will become fashionable, but which metric is being abused to hide what we do not want to see.

I am tracking three signals next season. First, how clubs value dressing-room chemistry. If someday a metric can measure integration, the market will change. Second, how financial-reporting pressure shapes mid-season deals. When a club sells a young player for accounting reasons, that is a signal that the governance model is overriding the sporting model. Third, how data models handle necessary and sufficient conditions. A good model does not just predict outcomes; it states the conditions under which the prediction holds.

I am not sure I have enough time left to see the market value dressing-room chemistry correctly. But I am sure of one thing: when it happens, people will call it a data revolution, and they will forget it began with people like me, sitting and counting every shot on A4 paper in Marseille in August 2026, refusing to believe a metric until I had verified it myself.

The question I leave for the next cycle is not which metric is right. It is: if you have a player with high xG and low goals, do you have the courage to buy the process instead of the result? And if you have a player with beautiful metrics but a dressing room that will not accept him, are you clear-headed enough to price what the dataset cannot measure?