How post-shot expected goals measures placement and finishing

Expected goals on target is the metric that answered a complaint about its predecessor. Standard xG scores an attempt before it is struck, so a shot rifled into the top corner and a scuffed effort straight at the goalkeeper carry the same value.

Post-shot models add the missing information. They describe the probability that a shot on target becomes a goal given where it crosses the line, which makes the metric a measure of finishing and placement rather than of chance creation.

Why a second model was needed

The original xG has no way to separate a good chance from a good finish. Two players can receive identical service, and only one converts; the model gives them the same credit for the opportunity and no credit for what they did with it.

Post-shot xG closes that gap by moving the moment of evaluation later. The attempt is only scored once it is on target, and the coordinates used are those of the ball at the goalline rather than those of the shooter's feet.

The two questions are not rivals. One describes the situation the shooter inherited, the other describes what the shooter did with it, and a full reading of a chance needs both numbers side by side.

Placement becomes the input

The central question changes from where the shot was taken to where it would have entered. Height and lateral position at the goalline describe how much of the target the shooter actually used, and the model converts that description into a probability.

Placement near a corner is worth more than placement near the middle, and a shot lifted into the top third is worth more than one rolled along the ground into the same lateral position. The geometry of the goal mouth does most of the work.

The model does not need to know how the ball was struck. A curled effort and a driven one that cross the line in the same place are scored identically, because the geometry of entry is all the model is given.

The goalkeeper is part of the model

A shot on target is only meaningful against a goalkeeper, so post-shot models record where the keeper stands and how the body is set when the ball is struck. A finish into the space the keeper has left is valued differently from the same finish into the space the keeper covers.

This is the most argued part of the metric. Including the goalkeeper means the number blends finishing with the keeper's positioning, and a goalkeeper who is consistently well placed will suppress the values of everyone who shoots at them.

Some implementations go further and model the keeper's reach as a probability surface. Others stay with a simpler positional term. The difference is invisible in the final figure and real in the comparison.

Timing matters as well. The keeper is recorded at the moment of the strike rather than at the moment the ball arrives, so an early shot past a keeper still moving is rated more generously than one struck straight at a set position.

What the difference between xG and xGOT tells you

The two numbers describe different halves of the same event. xG asks how good the chance was when the ball arrived; xGOT asks how good the finish was once the ball was on target.

A player whose xGOT runs ahead of xG is placing shots into areas the model rates highly, which is the closest thing the data has to a finishing skill signal. The reverse gap suggests placement into the middle of the goal, or a keeper who reads the shot early.

Neither gap is permanent. Finishing runs are short, and a modest difference over a handful of matches is noise rather than evidence of technique.

Aggregated over a season the pair becomes a diagnostic. A team whose post-shot totals track its expected goals is finishing as the chances suggested, while one that runs ahead is converting placement into value rather than generating better situations.

Limits of post-shot models

The most important limit is what the model never sees. Attempts that miss the target and attempts that are blocked are excluded, so xGOT totals are computed over a smaller and friendlier set of events than xG totals.

Deflected attempts that end on target are scored as if they had been struck that way. The model has no way to discount the luck, and a shanked effort that loops into the corner is recorded as an excellent finish.

• Only attempts on target are scored, so totals are not comparable with xG.
• Blocked shots disappear from the sample entirely.
• Goalkeeper positioning is baked in, which mixes two skills in one figure.
• The model cannot tell a deliberate placement from a lucky deflection.
• Small samples of on-target attempts are extremely noisy.
• Definitions of what counts as on target differ between providers.

Reading xGOT correctly

Treat the metric as a description of what happened to the shot rather than as a verdict on the shooter. It is most useful when read beside xG, because the pair separates the chance from the finish.

And read it over a run of matches. A single post-shot value can be produced by a deflection as easily as by a precise finish, and the model has no way to tell you which one it was.