Empty Tables, Full Confidence: The Silent Failure Inside Football's Data Pipelines

I watched one clip three times last season. The goalkeeper took the ball well outside his own box, drove a long diagonal to the right touchline, the ball landed perfectly, the wing-back collected it and stepped inside, the stadium erupted. Two hours later the clip had crossed two hundred thousand views. In the same week I pulled that goalkeeper's save data for the previous three seasons, specifically on low-driven shots. The line went down. Three seasons, three declines.

What the club did was rational. They paid for the clip. The clip is real, the data is real, but the two are answering different questions. Budgets get built from the first question. Results on the pitch come from the second.

I went looking for the transfer fee and kept finding an operating system. The fee is the symptom, not the story. The machine is the pipeline that hands a club its decisions — who scouts, which variables get measured, who reads those variables, and who is authorised to say "this information is not information." That last question is the least asked and the most expensive.

In November I got to see the notes from a club's recruitment meeting — I will not name the club, the source stays protected. Five names on the shortlist. Beside each one: age, estimated fee, agent. The column explaining why those five names and not five others was empty. The table was valid. The decision got made anyway. That is what this piece is about.

The information infrastructure football built over the past fifteen years is genuinely unprecedented. Liverpool put Michael Edwards at the centre of recruitment in 2026 and brought in the physicist Ian Graham to run research in the same period. Five years later the machine was ready. Mohamed Salah arrived from Roma for £36.9m in June 2026 and delivered 44 goals across all competitions in 2026-18.

I was a mid-level reporter at a Liverpool-based digital outlet then. My job was simple: build a standardised transfer-return spreadsheet combining xG, pressing recoveries and wage-to-output ratios. I ran the template across all twenty Premier League clubs and published twelve pieces in six weeks. On Salah the model projected 20-plus goal contributions; he delivered 44. Traffic rose 42 percent and the newsroom adopted the template.

What happened next happened inside clubs too. Brighton's Tony Bloom translated a betting-risk model into a club model. Brentford closed their academy in 2026 and went data-first on recruitment. Every Premier League club built some version of this layer. Standing here in 2026, English football does not have a shortage of data.

The problem sits elsewhere. What got built is a reporting layer, not a validation layer. A schema promises format, not content. Football's decisions walk straight across that gap.

The pipeline fails in three places. The first is acquisition — getting the material in. If an outlet sits behind a paywall, if the page renders in JavaScript, if a region block applies, the system comes back with nothing. The same thing happens inside a club: when the scouting database holds fewer than four matches per player for a given league, the model still returns a valid result. The result is meaningless.

The second is extraction — the schema being satisfied. A report can carry a headline, a date, numbers, names, and still be empty if it lacks the one sentence explaining what the subject means. The five-name shortlist is the case in point. The format was perfect. The argument was missing.

The third is decision — who is reading. Nobody on a board reads the raw table. They read the summary. And a summary generated from an empty table always says the same sentence: "no red flags." That is the deepest confusion in the system. The absence of a flag is not a clean bill of health. The first is missing information. The second is a validated decision.

To see where that distinction costs the most, look at financial compliance. Everton's ten-point deduction was announced on 17 November 2026 and cut to six on appeal in February 2026. Nottingham Forest lost four points on 18 March 2026. Manchester City faced 115 charges filed in February 2026. In each case the club's internal model needed one number — the rolling three-year loss. When that number arrives late, in a different format, or when auditor and finance team disagree on it, the model returns a valid-but-empty result. The board reads it as compliant.

The goalkeeper market is the cleanest example of the failure. In August 2026 Chelsea spent £71.6m on Kepa Arrizabalaga, a world-record fee for a goalkeeper. In the same window Liverpool spent £66.8m on Alisson Becker. The gap between the two fees was £4.8m, which means the market priced them as near-equivalents.

My years of watching matches tell me that near-equivalence was not an accident. It was procedural. A pipeline measures what is cheap to log. Long passes, distribution distance, pass completion — those are discrete events, visible on camera. Positioning before the shot, set position, near-post coverage, the angle of the hand on a low-driven ball — those are hard to encode. So the model inflates the weight of what it can see and deflates the weight of what it cannot. The market then prices two different profiles on the same variables, and the error surfaces three seasons later.

On set pieces I have seen the opposite, and that is where the optimism lives. At the 2026 World Cup in Russia I tracked dead-ball efficiency across all 64 matches. France's four set-piece goals and 38 percent aerial duel success were visible on the table before the final. After the 4-2 result plenty of people described it as a story of leadership. The set piece looked like luck until the efficiency table disagreed. What it actually was: delivery routine plus second-ball structure, repeatable design rather than heroism.

The comparison matters. Where a pipeline measures repeatable behaviour, decisions improve. Where it measures something visible but transient, decisions become bets.

The agent's role has to be added here, because the supply chain of information is not neutral. The package an agent sends a club carries twelve columns — matches, goals, assists, sprints, duels. It carries no column for provenance. The analyst ends up placing a tier-one journalist's verified figure and a tier-three rumour in the same table. The club's decision then rests on format rather than source.

My own record offers a direct case. In March 2026 the Premier League stopped. Anfield's 53,394 seats went empty with Liverpool 25 points clear. I ran a daily financial impact tracker estimating roughly £3.2m of lost matchday revenue per home game and conducted remote interviews with fourteen club executives. Across a twelve-week series I held a hard 6 p.m. filing deadline.

What I learned then sits at the centre of this argument. A deadline is a gate. It does not produce accuracy, it produces a decision. Facts were changing daily, but the 6 p.m. file had to go, which forced an explicit call on which information was sufficient and which was not. The biggest gap in club data departments is exactly that declaration: nobody states in writing what counts as sufficient.

This is where I have to step outside English football. In Bangladesh and across South Asia the problem inverts. The pipeline there did not break; it was never built. Scouting runs on networks, memory and lobbying. There is no fear of the empty table because there is no table. What is missing is more valuable — institutional memory. A club repeats the same mistake a decade later because the mistake was never written down. Capital, ownership and labour markets are structured so differently across the two that the same model will not transfer. In England data spend is a line item even for a Championship club. In Dhaka it is not a line item at all.

Fairness requires conceding the conventional view. More data means better decisions, and Liverpool 2026 is the strongest evidence available. That club did not buy players. They bought repeatable decisions — Salah at £36.9m delivering 44 goals, with Mané, Firmino and Robertson arriving through the same method. The model works. I will not pretend otherwise.

But the risk has moved. The 2010s problem was too little data. The 2020s problem is too much validated-but-empty output. Confidence scales faster than accuracy. An organisation with no gate in its pipeline becomes more certain when it receives more data, not more correct. And certain organisations commit more money to wrong decisions, because nobody in the room is left to ask the question.

The second uncomfortable point is organisational, not technical. We blame the model when the real question is who holds the authority to stop the pipeline. In most clubs, nobody does — because the person who could stop it is the person who commissioned it. Asking permission from yourself to switch off your own machine is not a control. That is a governance defect, not a software defect.

The third point concerns my own profession. Journalism carries the same failure mode. When a source files an empty report, when a feed returns a valid but zero-length list, we often publish it as "no news." It is actually "no information arrived." The ethics of the trade live inside that difference.

The fix is startlingly cheap, which is the problem. One assertion is enough: if the title is null, or the information list is empty, the pipeline should raise an error and halt. In club language — a provenance field, an entity registry, a hard gate. All three are low-cost work. None of them produces a highlight reel, so none of them gets funded. Organisations invest in what is visible. Nobody sponsors the invisible validation layer.

Watch where football's capital is actually going. Clubs are pouring billions into ownership, broadcast rights and multi-club networks, while those networks share no common standard for verifying information. Five clubs in one group write scouting reports in five different schemas. Information flows. Validation does not.

I learned more about football from a revenue gap than from a highlight reel. The gap tells you where the decision went wrong. The reel only tells you who ended up being right.

The market prices talent. The smartest clubs price the process that finds it. The clubs ahead over the next three years will not be the ones with the largest data department. They will be the ones with a person who can stand up in the meeting and say: this table is valid, it is also empty, and we are not deciding on it.

So the question for your club: the next time the model returns a clean table, who in that room holds the authority to say it is not a result?