Expected goals changed the way football is analyzed, and it is now changing the way football is bet. The metric — xG for short — assigns a probability to every shot taken in a match, based on the location, the type of assist, the body part used, the game state, and dozens of other variables. A shot from six yards out after a cutback across the box might be worth 0.45 xG. A header from the edge of the area off a deep cross might be worth 0.03 xG. Sum all the shot values for a team in a match, and you get the team’s expected goals — a number that represents how many goals they “should have” scored given the quality of chances they created.
For football analysts, xG is a better measure of team quality than actual goals scored, because it strips out the noise of finishing variance. A team that creates 2.5 xG per match but scores only 1.5 goals is underperforming its chances — their finishing will likely regress toward the mean, and their future results will look better than their recent results. A team that creates 0.8 xG per match but scores 1.5 goals is overperforming — their finishing is unsustainably good, and their results will likely worsen.
For bettors, this reversion-to-the-mean property is where the edge lives. If you can identify teams whose xG significantly diverges from their actual output, you can bet on the correction before the bookmakers fully adjust their lines. The bookmakers are not blind to xG — every major sportsbook has access to xG data — but the market does not always price the reversion efficiently, especially in lower leagues and early-season markets where sample sizes are small and public perception lags the data.
From xG to Betting Odds: The Conversion Problem
Having xG data is one thing. Converting it into actionable betting odds is another. The raw xG numbers tell you how many goals a team is expected to score and concede, but they do not directly tell you the probability of specific match outcomes — which is what you need to compare against bookmaker prices.
The standard conversion method uses a Poisson distribution. If Team A has an expected goal output of 1.8 and Team B has an expected output of 1.1, the Poisson model calculates the probability of every possible scoreline — 0-0, 1-0, 0-1, 1-1, 2-0, and so on — by treating each team’s goal-scoring as an independent Poisson process. The probabilities of all scorelines sum to the probabilities of match outcomes: home win, draw, away win.
The Poisson model is not perfect. It assumes independence between the two teams’ scoring, which is not strictly true — a team that goes ahead tends to sit deeper and concede fewer chances, while a team that falls behind pushes forward and creates more but lower-quality opportunities. It also assumes that goals are independent events within a match, ignoring momentum effects and the tendency for goals to cluster in certain phases of play. More sophisticated models — bivariate Poisson, Dixon-Coles, and various machine learning approaches — address these limitations to varying degrees.
For most bettors, the basic Poisson conversion is accurate enough to identify large discrepancies between model-implied odds and bookmaker odds. The edge from using any xG-to-probability model, even a simple one, over not using one at all is much larger than the edge from using a sophisticated model over a simple one. The first step matters most. Convert xG values into correct score probabilities with an expected goals calculator for betting — it simulates thousands of match outcomes using the Poisson framework to show the probability of every scoreline and the implied odds for match result, over/under, and both-teams-to-score markets.
Where xG-Based Betting Works Best
Not all markets respond equally well to xG analysis. The markets where xG provides the most edge are those where the bookmaker’s line is most influenced by actual results (which are noisy) rather than underlying chance quality (which xG measures).
Match result markets (1X2) are the most natural application. If a team’s xG profile suggests they should be winning 55% of matches but their actual win rate is 40% over the last 10 games, the bookmaker may still be pricing them closer to a 40% team — especially in early-season markets or lower leagues where the book relies more heavily on recent results than on advanced metrics. The xG-based bettor sees a team priced at 2.50 (40% implied) that their model says should be priced at 1.82 (55% implied) — a massive edge if the model is right.
Over/under goals markets respond well to team-level xG but require adjustments for opponent strength and game context. A team that creates 2.2 xG per match against an average defense might create only 1.4 xG against a top-five defense. Naively using season-average xG without adjusting for opponent quality is the single most common mistake in xG-based betting, and it produces bets that look profitable on paper but are systematically misprediced.
Correct score markets are where the Poisson model shines brightest, because the bookmaker’s margin on correct score bets is typically 10-15% — much higher than on match result or over/under. The wider margin means the bookmaker’s prices are less precise, and a bettor with a good xG-to-probability model can find edges that simply do not exist in the tighter match result markets. Correct score betting is higher variance, but the edge per bet is larger, and the Kelly-optimal stakes are correspondingly larger.
Building an xG Model: What You Actually Need
Building a functional xG model from scratch requires shot-level data with location coordinates, outcome labels (goal, saved, blocked, missed), and ideally additional features like body part, assist type, game state, and pre-shot player positioning. This data is available from several providers — StatsBomb, Opta, Understat, FBref — at varying price points and levels of detail.
The modeling approach can be as simple as a logistic regression on shot distance and angle, which captures roughly 70% of the variance in shot outcomes, or as complex as a gradient-boosted ensemble that incorporates 50+ features and captures 85%+ of the variance. The diminishing returns from model complexity are real — the jump from no model to a simple model is enormous, while the jump from a simple model to a complex one is modest.
For bettors who do not want to build their own model, using pre-computed xG values from public sources like Understat or FBref and converting them to odds via the Poisson framework is a pragmatic shortcut. The pre-computed values are not as accurate as a custom model tuned to the specific league and season, but they are vastly better than using no xG data at all. The key is to use the same data source consistently and to track the accuracy of the resulting predictions over a meaningful sample — at least 100 matches before drawing conclusions about the model’s edge.
The Pitfalls of xG-Based Betting
The most dangerous pitfall is treating xG as a certainty rather than a probability. A team with 2.5 xG in a match did not “deserve” to score 2.5 goals — they created chances worth 2.5 expected goals, and the actual number of goals follows a distribution around that expectation. Variance is real, and xG does not eliminate it. It quantifies it, which is useful, but quantifying variance and eliminating it are different things.
The second pitfall is using small samples. A single match’s xG can be highly noisy — a team can generate 3.0 xG in one match and 0.5 in the next against similar opposition. Season-long xG is much more stable and much more predictive of future performance. Bettors who react to a single match’s xG — “they had 3.0 xG, they’ll dominate next week too” — are making a sample-size error that is just as damaging as ignoring xG entirely.
The third pitfall is ignoring model limitations. Standard xG models do not capture set-piece quality, individual finishing skill above or below average, or tactical adjustments that teams make in response to specific opponents. A team with a world-class striker will systematically outperform their xG because the model does not account for his above-average conversion rate. A team that relies heavily on set pieces will also outperform, because most xG models undervalue set-piece chances. These are not bugs in xG — they are known limitations that the bettor needs to account for manually.
The Integration of xG Into Bookmaker Pricing
Bookmakers in 2026 are far more xG-aware than they were five years ago. Every major sportsbook now incorporates some form of expected goals data into their pricing models. This means that the easy edges — teams with massive xG divergence from actual goals over long samples — are largely priced into the major markets at major books.
Where edges remain is in the speed of adjustment (the bookmaker updates weekly, but xG changes match by match), in lower-tier leagues (where the bookmaker has less data and relies more on results), and in specific market types (correct score, halftime result, period-specific over/under) where the Poisson conversion from xG is more precise than the bookmaker’s pricing model.
The evolution of xG in betting is following the same pattern as every other analytical tool in financial markets: early adopters gained large edges, those edges attracted competition, the competition drove the edges down, and the remaining edges are smaller but still accessible to bettors who combine good data with good execution. The tool has not stopped working. It has stopped being a shortcut, and has become what it always should have been — one component of a disciplined analytical process, not a magic formula.
