xG in ten minutes for the data-curious football fan

You've seen the term “xG” on Match of the Day, in The Athletic, on Twitter. Expected goals. It's become one of those football terms that everyone uses and few people pause to define. Here's the ten-minute working version — what xG is, what it isn't, and how to read it without getting fooled.

The core idea

Every shot in a football match has some probability of becoming a goal. A penalty: roughly 0.76. A 35-yard speculative effort: maybe 0.02. A close-range header from a corner: depends on whether it's contested, the angle, the keeper's position, but ballpark 0.20.

xG models assign each shot a probability based on its characteristics (location, angle, body part, defensive pressure, pass type leading to the shot, etc.). Sum the xG of all of a team's shots and you have the team's xG for that match: a number that says “based on the chances they created, on an average day, they would have scored this much.”

That number is almost always more stable, match-to-match, than the actual goals scored. Goals are bursty; the underlying chance creation usually isn't.

Why this is useful

Football has a lot of noise. A 1-0 result could mean one team utterly dominated and converted one of fifteen chances; it could mean an even game where one team got a lucky deflection. The scoreline tells you nothing about which. xG tells you a lot.

Three of the most useful patterns xG reveals:

• Over-performance: a team scoring well above their xG rolling average is converting at an unsustainable rate. Their luck tends to regress.
• Under-performance: opposite. A team consistently creating chances but not converting is almost always a positive regression candidate (until you find out their striker has been replaced by a chair).
• Defensive shape: xG-against (xGA) is the same idea mirrored. A team with low xGA is keeping opponents away from the box. High xGA paired with low goals-against often points to a goalkeeper on a hot streak.

What xG is NOT

Three honest caveats.

It's not a verdict on the result.A team can have higher xG and lose 1-0. That doesn't mean the result was wrong — it just means the lower-xG team converted their chance and the higher-xG team didn't. xG models the chance, not the conversion; finishing skill, goalkeeper performance, and luck all live in the gap.

It's not a perfect model.Different xG models disagree on the same shot. Opta, StatsBomb, Understat, and the various proprietary models each use different feature sets. The right way to read xG is as a directional signal, not a precise number. “Arsenal 2.1 xG vs Chelsea 1.3 xG” means Arsenal probably had the better of it. It doesn't mean Arsenal “deserved” to score 2.1 goals.

Single-match xG is noisy.xG is a tool for understanding patterns over time. A single match's xG can be misleading — one team might get a low xG by registering twenty speculative long-range efforts, another might get a low xG by parking the bus and conceding nothing dangerous. Always pair single-match xG with context.

How MatchMind uses it

We pull team-level xG totals from Sportmonks for every Big-5 fixture and use them as features in our 1×2 probability model. Specifically: rolling-window xG averages, xG-for vs xG-against differentials, and home/away xG splits all feed into the calibrated probability you see on the match detail page.

We also surface raw team xG on every match card and on the match-detail Goal Analysis panel, so you can compare the model's output to the underlying inputs. If the probability looks weird, the xG numbers usually tell you why.

What we don't have (yet)

Per-shot xG. Shot maps. xG timeline charts (cumulative xG by minute). These exist on sites like Understat and Opta, but they require shot-level event data that Sportmonks doesn't expose at any plan tier. If we ever add them, it'll be because we've added a second data source. For now, team totals are what we have, and they're enough for the model.

The bottom line

xG is the football statistician's most useful tool. It's not magic, it's not perfect, and it's not a betting signal. It's a way of summarising the quality of the chances created in a match into a single number that's more informative than the scoreline alone.

Read it as a directional signal, pair it with context, and treat single-match values with appropriate skepticism. The numbers MatchMind publishes are honest team totals — that's the level of detail Sportmonks gives us, and we don't want to invent precision we haven't earned.