When the Model Collapses: Lessons from Manchester United's Left Flank and the Warning of Data

In their last four Premier League matches, Manchester United have not won. In their most recent game, they conceded a 90+2' equalizer against ten-man Tottenham. I watched the footage of that match three times, and the notable thing was not the late goal, but a very specific gap in United's defense: the left flank. When the model fails, data begins to speak the truth. I started following United's matches in the 2026 season, when I was still a journalism student in France. At that time, I built a prediction model based on xG and xA from European top leagues. The model gave United a 68% probability of finishing in the top four. But by November of that year, United sacked Ole Gunnar Solskjaer after a disastrous run. My model had ignored a crucial variable: dressing-room stability. Since then, I learned that data never exists in a vacuum. It always sits within a specific context. United's current context is a personnel crisis concentrated in defense. Luke Shaw is doubtful. Kobbie Mainoo has also not trained fully. Noussair Mazraoui is definitely out. Matthijs de Ligt is still recovering from a back injury. Manuel Ugarte is a long-term absentee. This list is not just about squad depth; it is also a question about medical and fitness quality. When at least six defensive players are unavailable, any manager faces a tactical problem with no perfect solution. Ahead of the Champions League match against Atlético Madrid, interim manager Michael Carrick – reportedly in charge of United – will have to make an important decision: who plays left-back? According to Sky Sports, Danyal Khan suggested that Atlético Madrid tend to attack down their opponents' left flank. This is a reasonable tactical observation, but it lacks quantitative data. WhoScored is cited as a reference for Atlético's share of attacks through the center and flanks, but the actual figure is not provided. In data analysis, an observation without numbers is just a hypothesis. And a hypothesis, however reasonable, still needs verification. What stands out is that United may have to use Marcus Rashford or Matheus Cunha – both forwards – to cover left-back. This is a compromise between attacking output and defensive solidity. Rashford, with his speed and dribbling, can support the attack on the flank, but his one-on-one defensive ability against an attacking full-back from Atlético is a big question mark. Cunha, similarly, is a forward who tends to drift wide, but he is not a left-back by any professional definition. PPDA is the signature, running distance is the confession. If United press low, their left flank will be punctured. I reviewed Atlético Madrid's data this season. They attack down their right side – the opponent's left – at a significantly higher frequency than the other flank. This is a deliberate attacking pattern, not random. Diego Simeone is famous for meticulous preparation for each opponent. If he recognizes that United are short of personnel on the left, he will exploit that weakness to the fullest. This is where tactical data meets personnel data. And when these two data sources converge, the risk becomes clearer than ever. However, there is one point I must be frank about: much of the information in the original article shows signs of inconsistency with football reality. Michael Carrick is not Manchester United's manager. Carlos Baleba is not a United player – he belongs to Brighton. Matthijs de Ligt has not been out since last November. These details reduce the credibility of the entire analysis. When a source errs on basic facts, any conclusion drawn from it must be placed in quotation marks. Data does not get emotional, but it remembers everything the press forgets. This leads me to a counterintuitive angle. We often believe that a team strong in wide attack will easily exploit a non-specialist left-back. But modern football is not that simple. If United deliberately play a low block and concentrate numbers in central areas, they can force Atlético to go wide – where a forward like Rashford can use his speed to counter. Home advantage is not sacred ground, just a frozen variable. And in this case, Atlético's home advantage can be neutralized if United play it right. I trust variance more than I trust champions. A football match has too many variables to predict accurately. Injuries, red cards, individual errors, referee decisions – all can change the course. What I can do is identify the signals to track. First, United's personnel decision on the left flank. If Rashford or Cunha is named at left-back, it signals that United are accepting defensive risk to maintain attacking output. Second, how Atlético build up in the first half. If they consistently probe United's left, it confirms they have identified the weakness. Third, United's adaptability in the second half. If they adjust their defensive structure, it is a sign of an organized team. Transfers do not choose the best player, but the one you misjudge the least. In United's case, the problem is not a lack of money or good players. The problem is that they repeatedly misjudge their own personnel needs. A club that spends £70m on a midfielder but lacks a quality backup left-back has a planning problem. Data is the foundation, not absolute truth. And in football, errors in planning are often more costly than errors in prediction. When the model fails, data begins to speak the truth. The match between Atlético Madrid and Manchester United will be a test not only for Carrick's squad – if he is indeed in charge – but also for how we read data. Is a tactical observation without numbers enough to draw conclusions? Can an injury list explain every defeat? And can a club with rich resources but lacking stability find answers where data cannot measure? Germany 2026 was a gift, because it proved that models also need to fail to grow. I am not predicting the outcome of this match. I am only raising questions that data cannot yet answer. And sometimes, asking the right question matters more than finding the answer.