Expected Goals transformed the way we discuss football. It gave us a shared vocabulary for chance quality, a method for separating process from outcome, and a tool that reshaped everything from pre-match previews to post-match analysis. Yet xG carries a fundamental blind spot. It estimates the likelihood of a shot becoming a goal based on where it was taken and the conditions surrounding it, but it says nothing about what happened once the ball left the player’s foot. A strike from six yards might register an xG of 0.6, but if it rolls gently into the goalkeeper’s arms, was it ever truly a 60% opportunity? The answer is no, and the next generation of metrics was built to fix that flaw. For anyone who takes soccer predictions seriously, this shift is worth understanding.
The clearest illustration of xG’s weakness comes from a single famous moment. Daniel Sturridge’s late equaliser for Liverpool against Chelsea in 2018 had an xG value of just 0.03. He was more than 27 metres from goal, hemmed in by defenders, and the opportunity was objectively poor. Yet the ball flew into the top corner, leaving the goalkeeper helpless. The pre-shot model said 3%. The result said goal. That gap between number and reality is exactly where modern metrics operate. When Sturridge struck the ball, the chance was low quality, but the execution was extraordinary. Traditional xG cannot capture that difference, and that failure has real consequences for how we judge players, teams, and matches.
The Limits of Pre-Shot Models
xG is a pre-shot model. It assesses the situation before contact: distance to goal, angle, defensive pressure, body part, and the type of pass that created the chance. This is valuable for measuring chance creation, but it merges two distinct skills. A forward who repeatedly finds good positions will build high xG totals even with average finishing. A goalkeeper facing a stream of low-quality efforts will appear busy without necessarily performing well. The metric captures the opportunity, not the execution.
Stats Perform created Expected Goals on Target (xGOT) to close this gap. xGOT is a post-shot model. It retains the original xG of the shot but adds a vital new input: the goalmouth location where the ball ended up. A shot that finishes in the bottom corner earns a higher xGOT than one struck straight at the goalkeeper, even when the pre-shot xG was identical. This may sound like a minor tweak. It changes everything.
Pre-xG, PSxG, and the Bayesian Shift
Academic work has pushed the idea further. A 2026 study in the International Journal of Performance Analysis in Sport combined Pre-shot Expected Goals (Pre-xG) and Post-shot Expected Goals (PSxG) inside a Bayesian probabilistic framework. The researchers tested their model on a Real Madrid versus Barcelona match from the 2024-25 season, and the findings were dramatic. With Pre-xG, Barcelona’s win probability stood at 65.51%. Once PSxG was applied, that figure climbed to 91.90%. The most likely scoreline moved from 1-2 to 0-3.
This is no small statistical detail. It is a fundamental rethink of what a shot represents. Pre-xG asks how good the chance was. PSxG asks how good the shot was. A team can craft excellent chances and waste them. A goalkeeper can face poor chances and still concede. Separating these questions produces a much truer picture of events on the pitch. For analysts building predictive models, the message is clear: relying on xG alone leaves valuable information unused.
The Goalkeeper Revolution
The most practical use of post-shot metrics may be in evaluating goalkeepers. Traditional save percentage is a crude tool. A goalkeeper who faces ten shots from 30 yards will save almost all of them, while one who faces three shots from six yards might concede twice. Judging them by saves made reveals nothing meaningful about their ability.
xGOT offers a solution. By comparing the xGOT value of shots faced against the goals actually conceded, analysts can calculate “goals prevented”. A goalkeeper whose xGOT conceded is much higher than their actual goals conceded is performing above expectation. This metric separates goalkeepers who make high-quality saves from those whose save counts are inflated by easy, low-quality shots. It is a far more honest measure of the position, and it is already in use at the elite level.
What This Means for Prediction
Integrating post-shot metrics into predictive models is still developing, but the direction is unmistakable. Research from the WorldCupArena project has started evaluating language models and deep-research agents on their ability to forecast not only match results but specific match statistics, player events, and tactical patterns. The benchmark asks models to predict not just who wins, but how the game unfolds. Post-shot metrics supply the granular data that makes such forecasts possible.
The practical effects spread outward. Picture a hypothetical Europa League qualifying tie where a side dominates possession and builds 2.5 xG but loses 2-0. Pre-shot analysis would call them unlucky. Post-shot analysis might show their shots were poorly placed, the goalkeeper was never seriously tested, and the xG total was inflated by speculative efforts from distance. A model using PSxG would lower its assessment of that team’s attacking quality, and its forecast for the next match would reflect that correction. Likewise, a team that concedes high PSxG yet keeps clean sheets is relying on luck, and a well-calibrated model will expect regression.
The same logic applies at the top of the game. Consider the race for a Champions League place, where fine margins decide millions in revenue and prestige. A club that consistently underperforms its PSxG at both ends of the pitch is likely to slide down the table, while one that outperforms it may be riding a wave that will eventually break. Predictive models that incorporate post-shot data can flag these trends before they show up in the league standings, giving analysts and bettors a genuine edge.
The Road Ahead
The metrics arms race shows no sign of slowing. Researchers are now exploring Post-Shot-On-the-Line Expected Goals (PSOLxG), which models the exact instant the ball reaches the goal line, factoring in shot velocity, trajectory, and the goalkeeper’s position. This is the logical endpoint of the post-shot revolution: assessing the shot at the precise moment before it either becomes a goal or does not.
None of this makes xG obsolete. Pre-shot metrics remain essential for measuring chance creation and tactical effectiveness. But they are incomplete alone. The next generation of football analytics blends pre-shot and post-shot models, asking not only what opportunities were created but what was done with them. The teams and analysts who embrace this fuller picture will hold a real advantage. Those who cling to xG alone will keep misreading matches, misjudging players, and missing the signal in the noise.
The ball leaves the boot. What happens next is no longer a mystery.