The Empty-Data Paradox: The Most Expensive Thing in the European Transfer Window

Three in the morning, and the analytics room in Lyon is still lit. I open the file on a 21-year-old full-back the club is considering paying 12 million euros for. The spreadsheet has fourteen columns. Nine of them are full of numbers: minutes played, passes per 90, ball recoveries in the opponent's final third, average distance covered, sprints above 25 km/h, duel win rate. The other five columns are blank. No training-load data. No soft-tissue injury history. No aerial metrics split by zone. No one-on-one record against specific forwards.

And the final column is empty – the one I need most to answer the only question that matters: can he sustain this intensity across thirty-eight matchdays.

At nine the next morning, the recruitment committee meets. They will decide. And they will decide by filling those empty cells with something else: highlight reels, an agent's testimonial, a rumour in the press. I have sat in many of those meetings. I know exactly what happens next.

In statistics, a blank cell is called a missing value. How you handle it determines the entire reliability of every conclusion that follows. A properly built analytical framework must carry a missing-data rule: when a field does not exist, every related conclusion must be flagged as insufficient information to assess. No inference. No filling with guesswork. It sounds dry, but that is the line between analysis and invention.

Professional football has not drawn that line. Clubs still make eight-figure decisions every transfer window while only six of ten input columns are populated. The four blank ones get filled with something almost impossible to verify: narrative.

I have followed European football for more than thirty years, the last fifteen through a data lens. Every window, the same mechanism repeats. When a club lacks information about a player, the market does not go quiet. The market fills the gap. A player with no clear injury data gets described as durable. A player with too small a defensive sample gets described as having potential. A midfielder with twenty appearances in a weak league gets described by the one good game he had against a strong side.

In Vietnam the story runs at a slower rhythm but with the same nature. V.League still lacks a synchronised positional-data system covering every match, and most clubs assess players through video and scout reports. When a club wants to sign a foreign player, the file is almost always missing three things: actual minutes at high intensity, soft-tissue status by muscle group, and adaptability to a tropical climate. Those blanks get filled by an agent's recommendation – a source with a very obvious incentive.

During a transfer window, I grade rumours in four tiers: confirmed by the club, confirmed by the agent, corroborated by an independent second source, and single-source only. The last tier accounts for most of the traffic, and most of the errors. A rumour without a second source is not information; it is an empty cell written in words.

The transfer window is a market of gaps. Money flows from one data-thin place to another, and in between sits a stream of rumour acting as substitute data.

In the 2026-18 season, Lyon beat Marseille 3-2 in a match where expected goals said the opposite: Lyon created 1.6, Marseille 2.3. I wrote about it and got hammered. Three years later, the same people who hammered me called that metric the standard. Numbers never lie, but they know how to hide. Our job is to make them talk.

People see the goal. I see the gap between two centre-backs stretched apart by PPDA.

PPDA – the passes an opponent is allowed per defensive action – is the metric I use most when assessing a pressing system. Ahead of the 2026 World Cup round of 16 in Kazan, the PPDA gap between France and Argentina reached nearly four units. I wrote that Argentina's midfield would break vertically, and the match ended 4-3 exactly along the lines that metric had drawn. PPDA is not a number. It is a measure of a collective's patience against a dead ball.

In 2026, the game stopped. I rebuilt Lyon's training programme around GPS data: locomotion load, change-of-direction load, the ratio between high-speed load and base load. When the league resumed, the squad's muscle injuries fell from 12 to 5 over the same window. A season inside a bubble, yet the GPS still recorded every breath a player took. Nobody outruns data.

And yet most of the data used in recruitment is far thinner than that.

Distance covered is sold to fans as a measure of effort. It is meaningless unless split by pitch zone and match state. A midfielder running 12.4 km may simply be chasing the ball while his team is losing. A centre-back running 9.1 km may have shut down an entire left flank. I once reviewed a report ranking midfielders by total distance; the team topping that list finished the season in the bottom half, because their players were running to correct their own positional errors.

Possession is the same. Sixty-three percent means nothing if seventy percent of the passes happen inside your own half. I once took one of those matches apart: 63% possession, 41 passes into the final third, 2 clear chances. Beautiful number, poor product.

In the transfer market the problem is worse because of sample bias. A 19-year-old in the Belgian second tier may have positional data for just 6 matches. A 27-year-old in the Premier League has 34. Put the two files side by side and normalise per 90 minutes, and you create the impression the samples are equivalent. They are not. The error margin on a 6-match sample is many times that of a 34-match sample, and any valuation model that ignores this is selling belief.

Lyon sold Bradley Barcola to Paris Saint-Germain in August 2026 for a reported fee of around 45 million euros. That deal had enough data on both sides: minutes, successful dribbles, top speed, wage bill. Michael Olise moved from Crystal Palace to Bayern Munich in June 2026 for a reported fee between 50 and 53 million euros, after a season in which his per-90 attacking metrics ranked among the best in the Premier League. That file was thick, and Bayern paid accordingly. In the summer of 2026, Kylian Mbappé moved from Paris Saint-Germain to Real Madrid on a free transfer, and the entire negotiation revolved around clause structure and image rights rather than a transfer fee. Three deals, three kinds of file, one principle: money follows the thickest data.

It is also why I always read the release-clause structure before I read any other number. A release clause tells you what the owning club thinks the player is worth; the wage bill tells you how much they actually believe in him.

Correlation is not causation. Everyone knows that, and almost everyone forgets it when a young player scores seven goals in eight games.

The counterintuitive angle I want to push further: more data does not automatically produce better decisions. If the added data is itself skewed – if only players in big leagues are fully tracked – then adding it amplifies the old bias rather than correcting it. This is the blind spot of the analytics community itself, and it makes recruitment models dangerously self-assured.

And this is where I have to argue against myself. For years I rejected all qualitative description. I treated the word mentality as an empty phrase. But there are things data does not measure: how a player reacts after losing his position, how he organises his team-mates, how he receives the ball when his side is a goal down in the 80th minute. Those observations cannot replace expected goals or PPDA; they fill the cells the spreadsheet does not have a column for. The difference is that they are verifiable observation, not a story retold from rumour.

The most dangerous place is the space between the two. When a club lacks data and also lacks enough observers, the only thing left is the agent's story. And an agent has no statistical obligation.

In this transfer window, I will read in this order: minutes played at high intensity over the last two seasons, soft-tissue injury history by muscle group, and release-clause structure. Those three answer three different questions, and none of them appears in a highlight reel.

If a deal can only be defended by highlights and rumour, I default to treating it as an empty cell. Football is not a game of luck. It is a game of probability, and the winners are the ones who can read the table.

That spreadsheet is still on my hard drive. Five blank columns, never filled. The club signed him anyway. The following season, he made eleven appearances.