Expected Goals Explained: The Metric Changing Football

If you have watched a football broadcast in the last few years, you have almost certainly seen the term "xG" flash across the screen. Expected Goals has gone from a niche analytics concept to one of the most discussed metrics in football. But what does it actually measure, why should you care, and how does it help predict future results?

What Expected Goals Actually Measures

At its simplest, Expected Goals assigns a probability to every shot taken in a football match. That probability represents the likelihood of that shot resulting in a goal, based on thousands of historically similar shots.

A penalty has an xG of roughly 0.76 -about 76% of penalties are scored. A shot from 30 yards with a defender in the way might have an xG of 0.03. A one-on-one with the goalkeeper from six yards could be 0.45. Add up all the xG values from a team's shots in a match and you get their total Expected Goals for that game.

In the example above, Team A were the better side by almost any measure. They created chances worth 2.3 expected goals but scored none. Team B were clinical -or lucky, depending on your perspective -converting a single low-quality chance. Over a full season, Team A's approach will win more often. Expected Goals reveals the underlying quality that the scoreboard sometimes hides.

What Goes Into an xG Model

The probability assigned to each shot depends on several factors. The most important ones include the distance from goal (closer shots are more likely to be scored), the angle to goal (a central position gives a bigger target than a tight angle), the body part used (headed shots have a lower conversion rate than shots with the foot), and the type of assist (through balls and crosses create different shot qualities).

More advanced xG models also incorporate the positions of defenders between the shooter and the goal, whether the shot was taken first-time or after a dribble, the speed of play leading up to the shot, and the goalkeeper's position. Each additional factor adds predictive accuracy but also complexity.

Why xG Is Better Than Goals for Prediction

This is the crucial insight for anyone interested in football prediction: actual goals scored are a noisy, unreliable indicator of future performance. Expected Goals are a much more stable and predictive metric.

Consider two teams that have each played ten matches. Team A has scored 18 goals from 12.0 xG -they have been over-performing their chances. Team B has scored 9 goals from 14.5 xG -they have been under-performing. A naive prediction model using actual goals would rate Team A much higher. But an xG-aware model knows that Team A's scoring rate is unsustainably high and Team B's is unsustainably low. Over the next ten matches, Team B is likely to outscore Team A.

xG and Score Forecasting

For forecasting strategy, xG data is invaluable because it helps identify teams whose actual results do not reflect their true quality.

A team on a three-game losing streak might look weak, but if their xG data shows they have been creating 2.0+ expected goals per game while conceding only 0.8, they are likely to bounce back soon. Conversely, a team on a winning run built on low-xG performances is vulnerable to a correction. Our Bayesian models incorporate this xG data to produce more accurate score predictions than models based on results alone.

This is particularly useful early in the season when actual results can be heavily influenced by small sample sizes. After five matches, a team's xG data tells you more about their true level than their points total does.

The Limits of xG

Expected Goals is powerful but not perfect. It does not account for individual player quality within a position -a penalty taken by a specialist penalty taker has the same base xG as one taken by a nervous defender, even though the conversion rates differ. Some strikers consistently out-perform their xG because they are genuinely elite finishers. Others consistently under-perform because their technique is poor under pressure.

xG also says nothing about events that prevent shots from happening in the first place: tactical discipline, pressing intensity, possession control, and set-piece defending. A team that concedes few shots might have a low xG-against not because they are lucky but because their defensive structure is excellent. These factors require additional models to capture.

xG vs xPts: A Related Metric

Expected Points (xPts) extends the xG concept from individual shots to match outcomes. Using the xG totals for both teams in a match, you can simulate thousands of possible outcomes and calculate the expected points each team "deserved" based on the quality of chances created.

Over a season, the gap between actual points and expected points often corrects itself. Teams significantly above their xPts tend to drop, while those below tend to rise. This makes xPts a useful input for predicting which teams will improve or decline in the second half of a season.

How We Use xG in Our Models

Our prediction engine uses xG in several ways. It feeds into the attack and defence ratings that drive our Bayesian model, giving a more stable estimate of team strength than goals alone. It helps identify regression candidates -teams whose results are likely to shift. And it informs our confidence scores: predictions for matches where xG and actual performance align carry higher confidence than those where the two diverge.

Combined with market data, form analysis, and historical head-to-head records, xG is one of the most important inputs to our 12-model ensemble. It is not the only metric that matters, but it is arguably the most important single innovation in football analytics of the last decade.