Expected Goals (xG) Explained

Expected goals assigns each shot a probability of becoming a goal based on the characteristics of the chance.

Expected goals, universally abbreviated as xG, has become the foundational statistic underpinning most serious football analysis and betting models over the past decade. Understanding what it actually measures, and just as importantly what it does not measure, is essential before relying on it for betting decisions.

What xG actually measures

Every shot in a football match is assigned an xG value between 0 and 1, representing the probability that a shot with those specific characteristics results in a goal, based on analysis of thousands of historical shots with similar characteristics. A tap-in from close range might carry an xG of 0.8, while a speculative long-range effort might carry an xG of 0.03.

Summing the xG values of every shot a team takes across a match produces that team's total xG for the game, a single number representing the quality of chances created, independent of whether those specific chances were actually converted into real goals on the day.

The inputs behind an xG model

The specific factors that typically drive a shot's xG value include the distance from goal, the angle to the goal, whether the shot was taken with a foot or head, how many defenders were positioned between the shooter and the goal, the type of pass that created the chance, and whether the shot followed a particular passage of play such as a fast break or a set piece.

More sophisticated xG models incorporate additional context such as player positioning data and shot placement, producing increasingly granular chance-quality estimates, though even relatively simple xG models built on the core factors above tend to capture the majority of the useful signal for most betting applications.

Why xG matters more than actual goals scored

Actual goals scored in any single match are heavily influenced by finishing quality and simple randomness: a team can create several excellent chances and convert none of them, or score from their only meaningful opportunity of the match, both of which say relatively little about the underlying quality of performance compared to what xG captures.

Over a larger sample of matches, a team's actual goal output tends to converge toward what their xG would predict, following the same regression to the mean principle covered in this site's Trading Academy, which is precisely why xG is considered a more reliable predictor of future performance than recent actual scoring results alone.

Using xG to build betting models

A common application of xG in betting is building a team's expected attacking and defensive strength directly from their xG for and xG against figures, rather than from raw goals scored and conceded, since xG-based ratings tend to be less distorted by short-term finishing variance and therefore more stable and predictive over a rolling sample of recent matches.

These xG-based ratings can then feed directly into the kind of power rating and scoreline probability models described elsewhere on this site, offering a more robust foundation than raw scoring statistics alone for markets including BTTS, over/under goals, and correct score.

The limitations of xG

xG is a genuinely useful tool but not a complete description of match quality; it generally does not fully capture factors such as exceptional individual finishing skill, goalkeeper quality in stopping specific types of shots, or broader tactical context beyond the shot itself, meaning two teams with identical xG figures can still differ meaningfully in their actual underlying quality.

Different publicly available xG models also use somewhat different underlying methodologies and input weightings, meaning the specific xG figure quoted for the same match can vary noticeably between providers, which is worth keeping in mind when comparing xG figures sourced from different data providers.

xG and small sample sizes

Like most underlying statistics in football, xG is considerably more reliable and predictive over a larger sample of matches than over just a handful of recent games, since even a well-calibrated xG model cannot fully account for genuine short-term variance in a small number of matches.

A team's xG figures from their last three or four matches should generally be treated with real caution as a predictor of future performance, while a rolling sample covering ten or more matches tends to offer a considerably more stable and reliable read on a team's genuine underlying attacking and defensive quality.

Comparing xG with actual results to spot value

One practical betting application of xG is identifying teams whose actual results have diverged significantly from what their underlying xG would predict, whether over- or under-performing their chance quality, since this divergence often signals an unsustainable run of finishing luck likely to regress toward the underlying xG level over time.

A team significantly outperforming its xG through exceptional recent finishing is a candidate for regression toward more modest actual results going forward, while a team significantly underperforming its xG despite creating good chances may represent value going forward as that underperformance is statistically likely to correct.

xG for individual players versus teams

Beyond team-level analysis, xG is also calculated for individual players, providing a way to assess a striker's chance quality and finishing ability separately, comparing their actual goals scored against their own personal xG total to judge whether they are over- or under-performing their underlying chance quality on an individual level.

Player-level xG data is particularly useful for player prop betting markets, such as anytime goalscorer or shots on target props, offering a more rigorous foundation than simply looking at a player's raw recent goal tally, which can be heavily skewed by a small number of matches.

Practical tools for accessing xG data

Several publicly available websites and data providers publish match and player-level xG figures for most major leagues, offering a reasonable starting point for bettors without access to proprietary data or the technical resources to build a full custom shot-quality model from raw match data.

As with any publicly available statistic, the fact that xG data is widely accessible to the broader betting market, including sportsbooks themselves, means it should be treated as a foundational input into a wider analytical process rather than a standalone source of guaranteed edge on its own.

Building xG into a wider football betting process

xG works best when combined with other contextual information covered throughout this Football Academy section, including confirmed lineups, home and away splits, and situational factors such as rest and fixture congestion, rather than being relied upon as a single, standalone metric in isolation.

Tracking how well xG-based model predictions have performed against actual results and market prices over a large sample, following the record keeping principles described in this site's Trading Academy, is the most reliable way to confirm whether a specific xG-based approach is genuinely adding value.

Non-shot xG and possession-based metrics

Beyond traditional shot-based xG, some more advanced models incorporate non-shot expected goals or possession-value metrics, which attempt to assign a value to every touch of the ball, not just shots, capturing build-up play quality that traditional xG, focused purely on the shot itself, does not directly measure.

These more advanced metrics remain less widely and freely available than standard xG, but they represent a genuine area of ongoing development in football analytics, worth being aware of even for bettors who primarily rely on simpler, more accessible shot-based xG figures for most of their practical modelling work.

xG as a common language across the betting industry

Because xG has become so widely adopted, it now functions as something of a common language between bettors, analysts, and even sportsbooks themselves, all of whom use broadly similar underlying concepts when discussing chance quality, even where the specific numerical models differ in their precise methodology.

This shared vocabulary makes xG a genuinely useful starting point for a bettor building their own football model, since it connects directly to a large and growing body of publicly available analysis, historical data, and ongoing methodological refinement happening across the wider football analytics community.

Example shot xG values by chance type

The table below shows illustrative xG values for a range of common chance types, based on typical historical conversion rates for similar shots.

These ranges illustrate why xG treats shots so differently based on their characteristics; a match with several close-range chances can carry a high total xG even if none are actually converted, reflecting the genuine quality of the chances created rather than the eventual scoreline.

Key takeaways

• xG assigns each shot a probability of becoming a goal based on its specific characteristics.
• Total match xG measures chance quality created, independent of whether chances were actually converted.
• Actual goals converge toward xG over a larger sample, making xG more predictive than recent scoring alone.
• xG does not fully capture finishing skill, goalkeeper quality, or all tactical context.
• xG is most valuable combined with lineup news, home/away splits and other contextual factors, not in isolation.