Allianz Arena, Munich, April 2024. The 78th minute of the Champions League quarterfinal between Bayern Munich and Arsenal. A right-foot effort from Bukayo Saka, from twelve yards out, traveling at roughly 87 kilometers per hour into the bottom-right corner of the goal, off a deflection that meant Manuel Neuer had less than 0.4 seconds from the moment the ball changed direction to the moment it crossed the line. Neuer is fully extended, gloves first, fingertips brushing the leather of the ball as it kisses the post and stays out. The save, by the post-shot expected goals model running on StatsBomb’s data feed, was worth +0.31 expected goals — the highest single-action goalkeeping value in any Champions League match that week. The broadcast credits Neuer for “a great save.” The traditional goalkeeping statistics will credit him for “one save.” The post-shot xG model is, in 2026, the only public framework that captures what just happened with anything resembling precision.
Goalkeeping is, in 2026, the most underrated corner of football analytics in mainstream public coverage. The traditional metrics — saves, save percentage, clean sheets — were designed in an era when football had no measurement of shot quality at all, and they consequently treat all saves as equivalent regardless of difficulty. A keeper who makes a routine grab from twenty yards out is credited with the same “save” as one who tips a top-corner volley over the bar. The modern post-shot expected goals framework, paired with goalkeeper-specific extensions like xG on target (xGOT) and goals prevented above expected, has resolved this measurement problem inside clubs and at commercial data providers. The public conversation about goalkeeping has barely caught up. Mainstream coverage still treats keeper rankings as a list of save percentages and clean sheet totals, when the underlying analytical reality is that those metrics are nearly useless without shot-quality context.
I have been writing about football analytics from London since 2014, with growing focus on goalkeeping as a measurable craft rather than a mystified one, and the framework that has changed my own reading of the position most measurably is the one this article is going to unpack. Goalkeeping beyond saves — what post-shot xG actually measures, how the modern goalkeeping toolkit works, where the math breaks, and how to read a keeper’s season without reverting to clean-sheet illusions, is the subject of this article.
The origin: where goalkeeper analytics came from
Goalkeeper performance analysis, in its modern statistical form, dates to the mid-2010s when the broader xG framework was being established. The original xG models measured the probability that a shot would result in a goal given its location, angle, and basic context — but the model was a pre-shot framework. It described the chance the shooter had created. It did not say anything about what the keeper did with that shot.
The first serious public-facing goalkeeper analytics came from StatsBomb‘s introduction of post-shot expected goals (sometimes called xGOT, for expected goals on target) in the late 2010s. The post-shot model, conditional on a shot being on target, estimated the probability that the shot would beat a league-average goalkeeper given the shot’s placement, velocity, and the keeper’s pre-shot position. A shot placed perfectly into the top corner has a post-shot xG near 0.85; a shot that hits the keeper in the chest has a post-shot xG near 0.05.
The keeper’s performance is then measured by goals prevented above expected — the difference between the post-shot xG faced and the actual goals conceded. A keeper who faces 25 expected goals worth of on-target shots and concedes 20 has prevented 5 goals above expectation, an elite season-long performance. A keeper who concedes 30 against the same expected total is performing 5 below expectation, a struggling season.
By 2023, the post-shot framework was the dominant public-facing keeper analytics. FBref publishes goalkeeper-specific xGOT and goals-prevented-above-expected for top European leagues. InStat and Opta provide competing commercial versions. StatsBomb’s commercial product includes the deepest goalkeeper-specific dataset, with shot-by-shot expected outcomes and post-shot placement tagging.
How post-shot xG works: in plain language
The basic mechanic conditions on a shot being on target — meaning the shot, if untouched, would go in the goal — and then estimates how likely the shot is to beat a league-average keeper. The inputs include shot location, shot velocity, shot placement within the goal frame (corners are harder; central shots are easier), the pre-shot position of the keeper, and the time the keeper has to react.
The output is a probability between 0 and 1. A perfect top-corner shot, struck hard and traveling unimpeded toward the post, can post a post-shot xG of 0.90 or higher — meaning a league-average keeper saves it about one time in ten. A shot at the keeper’s body posts a post-shot xG closer to 0.05 — saved 95% of the time. The model captures, with reasonable precision, how difficult each on-target shot actually was to save.
Aggregated across a keeper’s season, the post-shot xG total represents the expected goals against if the keeper performed at league average on every shot they faced. The keeper’s actual goals conceded is the realized outcome. The difference — usually expressed as “goals prevented above expectation” or “PGA” — is the keeper’s contribution above baseline.
An elite keeper season, by this metric, prevents 6-10 goals above expectation across the campaign. A struggling season can have a keeper performing 5+ goals below expectation. The variance between elite and poor keeper performance, on this metric, is large enough that the framework genuinely separates the top tier from the rest.
The critical component: separating keeper performance from chance quality faced
The single most important insight from the post-shot framework is that keeper performance and the chance quality faced are separate measurements that must be analyzed separately. A keeper conceding 35 goals in a season is not, by itself, a useful number. The same total could come from facing 50 expected goals worth of chances (in which case the keeper is performing 15 goals below expectation, a disaster) or from facing 25 expected goals (in which case the keeper has prevented 10 goals above expectation, an elite season).
The traditional clean sheet metric is the worst version of this problem. A keeper with 15 clean sheets in a season may be facing a Manchester City-quality defense in front of them, where the keeper barely has to make a save in many of those clean-sheet games. Another keeper with 5 clean sheets may be facing twice as many on-target shots and saving the majority of them. The clean sheet count rewards the team’s overall defensive structure, not the keeper specifically.
The Allianz Arena Neuer save that opened this article is, in this framework, a +0.31 expected goal contribution from the keeper. A keeper who routinely makes those saves — who consistently outperforms the post-shot expectation — is doing measurable work that the clean sheet column will never capture.
Goalkeeper metrics vs the alternatives: a comparison
| Metric | What it measures | Strengths | Weaknesses |
|---|---|---|---|
| Save percentage | %% of on-target shots saved | Simple, widely cited | Doesn’t account for shot difficulty |
| Clean sheets | Games conceding 0 goals | Counting stat, audience-friendly | Rewards team defense, not keeper |
| Goals prevented above expected (PGA) | Actual goals minus post-shot xG | Cleanest individual keeper measure | Requires post-shot xG infrastructure |
| Goals conceded per match | Total goals against per 90 minutes | Direct outcome | Confounded by team and league context |
| Distribution metrics (passes, pass completion %%) | Keeper’s ball-playing ability | Captures modern keeper role | Separate from shot-stopping evaluation |
The honest evaluation uses post-shot xG as the primary shot-stopping measure, supplemented with distribution metrics for the keeper’s ball-playing contribution. Save percentage and clean sheets are noise without the underlying context.
What the data needs: inputs
Post-shot xG calculation requires shot-by-shot event data with placement information (which corner of the goal the shot was aimed at), shot velocity (when available from optical tracking), and pre-shot keeper positioning. The most sophisticated versions also include defensive context (was the keeper unsighted, was the shot deflected before reaching the keeper).
StatsBomb‘s commercial product is the most-cited source for full post-shot detail. FBref publishes derived versions for public access, including season-aggregated post-shot xG and goals prevented above expected for top European leagues. Opta and InStat provide competing commercial versions used by clubs.
Building the analysis: a working framework
- Pull the keeper’s season-long post-shot xG faced and actual goals conceded. Calculate the difference (goals prevented above expectation).
- Compare to season-long context: how many on-target shots did the keeper face, what was the average shot quality.
- Identify specific patterns: is the keeper performing better against headers vs feet, against close-range vs long-range, against high vs low shots?
- Cross-reference with distribution metrics: pass completion percentage, long-pass accuracy, sweep success rate. The modern keeper’s value includes ball-playing.
- Compare across multiple seasons: season-to-season variance in PGA is real but stabilizes around the keeper’s true level over 2-3 seasons.
Where this gets weird: common mistakes
Single-season overconfidence. A keeper’s PGA over one season has substantial variance. The keeper’s underlying skill is better captured by 2-3 seasons of data than by single-season swings.
Defensive-line confounds. A keeper playing behind a high defensive line faces different shots than one behind a deep block. The shots-faced volume and the typical shot quality differ. The framework partially accounts for this but doesn’t fully isolate keeper skill from team context.
The “lucky save” interpretation. A keeper who makes a +0.85 post-shot save (one in ten chance of saving) gets credit for a brilliant save. Mainstream coverage sometimes treats this as fortune rather than skill. The post-shot model, structured carefully, says it’s exactly the kind of save the elite keepers make at higher rates than the league baseline. The “luck” framing under-credits the position.
Distribution evaluation in isolation. A keeper with elite pass completion is a different player than one whose primary skill is shot-stopping. The two are partially independent. A keeper who can’t pass but stops shots elite is valuable to a team that wants direct play; a keeper who passes well but is average shot-stopping fits a possession-based system. Neither profile is “better” — they serve different tactical needs.
Set-piece distinction. A keeper’s performance on open-play shots and set-piece shots can be very different. Some keepers are elite at one and average at the other. Aggregated PGA can obscure this. The granular version separates the two.
When goalkeeper analytics shines: use cases
Cross-team comparison. Comparing keepers on different teams is the cleanest application of post-shot xG. The team-context confounds that distort save percentage and clean sheet totals are addressed by the framework.
Transfer evaluation. A club signing a new keeper wants to know if the keeper will improve their team’s goals conceded. The post-shot xG history of the player, combined with their distribution metrics, produces a more reliable projection than save percentage alone.
Position-specific scouting. A team that needs a sweeper-keeper (one comfortable playing outside the penalty area) versus a traditional shot-stopper-only is evaluating different profiles. The metrics, used together, support specific role identification.
Long-term performance tracking. A keeper’s career arc — when they peaked, when they started declining — is much more visible through PGA than through traditional metrics. The framework can identify keepers who are quietly elite but on poor defensive teams (the inverse of clean sheet inflation).
A working example: Emiliano Martínez and the 2022 World Cup
Argentina’s Emiliano Martínez and the 2022 World Cup is the cleanest recent keeper-analytics case study. Martínez prevented an estimated 4.2 goals above expectation across the tournament, anchoring the Argentina defense through a series of high-stakes matches. The most-cited individual contribution was the final save in the World Cup final shootout against France, where Martínez’s decision-making, body positioning, and reaction time produced a +0.65 post-shot expected-save contribution in a single moment.
The tournament-long context is more interesting. Argentina’s defense was not statistically elite by team-level expected goals against — the side conceded substantial on-target shot volume across the seven matches. Martínez’s performance compressed the actual goals against well below what an average keeper would have allowed. The tournament’s narrative credited multiple players for Argentina’s success; the post-shot xG framework specifically credited Martínez for a measurable share of the title.
The retrospective writing about the 2022 World Cup, in mainstream coverage, primarily emphasized Lionel Messi’s scoring and Argentina’s emotional run. The Martínez contribution was treated as a single highlight (the final shootout) rather than a tournament-long pattern. The post-shot framework, applied carefully, tells a more complete story.
The limits: what goalkeeper analytics cannot tell you
Post-shot xG cannot predict individual saves. The framework gives probability estimates over series of shots. Single saves can swing in any direction relative to expectation.
Post-shot xG cannot model penalty situations cleanly. Penalties are technically shots but have substantially different dynamics (no chance for keeper movement until the kick, specific psychological elements). Most rigorous keeper analytics report penalty performance separately.
Post-shot xG cannot capture command-of-area performance. A keeper who claims crosses confidently, organizes the defensive line, and dominates set-piece situations contributes value that the shot-by-shot framework doesn’t fully measure. Separate command-of-area charting exists at commercial providers; the public versions usually don’t include it.
Post-shot xG cannot replace film evaluation for the harder keeper-evaluation questions — handling under pressure, communication, decision-making on rushed situations, distribution under press. The metrics are the analytical foundation. The film fills the gaps.
One additional limit: post-shot xG models vary slightly across providers. StatsBomb’s, Opta’s, and FBref’s derived versions can produce somewhat different season-long numbers for the same keeper. The careful writing names the source and avoids treating any single number as definitive.
Frequently asked questions
What is a “good” goals prevented above expected season?
Elite season-long PGA is +6 to +10 goals across a typical league campaign. A solid season is +2 to +5. A struggling season is below zero, with anything beyond -3 indicating significant underperformance. The Premier League’s elite keepers consistently produce in the +5 to +8 range.
How does the post-shot model handle deflections?
Most models include a deflection flag and adjust the post-shot expected goal accordingly. A shot that’s deflected before reaching the keeper presents a different probability of saving than an undeflected one. The major commercial providers handle this; the public versions vary in their handling.
Are save percentages still useful?
Largely as historical reference. For modern keeper evaluation, save percentage without underlying shot-quality context is misleading. Two keepers with identical save percentages can be performing very differently if they’re facing different shot quality.
Where can I see public keeper data?
FBref publishes goalkeeper-specific xGOT and goals prevented above expected for top European leagues. StatsBomb’s commercial product is the gold standard for shot-by-shot detail. Opta and InStat have competing commercial offerings used by clubs.
Sources and further reading
- StatsBomb’s post-shot xG explainer — the most accessible introduction to the framework.
- FBref — public-facing goalkeeper statistics including PGA for top European leagues.
- The Athletic football coverage — Tom Worville and others writing keeper-specific analytical pieces.
- Understat — xG-based shot data, useful as cross-check on commercial providers.
- Tifo Football — video analytical content frequently surfacing post-shot xG context for high-profile keeper performances.
The Neuer save at the Allianz that opened this article — fingertip extension on a 0.4-second reaction window, +0.31 post-shot expected goal value — was the kind of moment the post-shot xG framework was built to credit. Traditional metrics would have recorded “one save.” The corrected framework records something closer to the truth: a goalkeeper, at his best, doing work that a league-average keeper saves perhaps one time in three. For the broader analytical frame on reading football matches honestly, our guide to expected goals is the natural foundational read.
