You're watching a Premier League match with a mate. Full time: the home side wins 2-1. Your mate is convinced they "dominated": he saw two goals, case closed. You check your phone and see an xG of 1.2 against 2.1: the visitors actually created more, but missed three one-on-ones. Same match, opposite conclusions. Who's right?
That gap sits at the heart of the data revolution in football analysis. On one side, traditional predictions: bookmaker odds, instinct shaped by highlights, overrated head-to-heads. On the other, xG models: every shot rated by its probability of becoming a goal, added up over several matches and compared between the two teams. The two worlds don't say the same thing. Here's why, with worked numbers.
What is a "traditional" prediction?
For a long time, almost everyone predicted matches this way, and plenty still do, including tipsters with big followings. The traditional approach rests on three shaky pillars.
1. The bookmaker's implied odds
Odds of 1.80 on a home win mean an implied probability of 55.6% (1 ÷ 1.80). Many tipsters start there, adjust a little and think they have an opinion. But bookmaker odds include a margin. Take a match priced 1.80 / 3.60 / 4.50:
| Outcome | Odds | Implied probability | Without the margin |
|---|---|---|---|
| Home win | 1.80 | 55.6% | 52.6% |
| Draw | 3.60 | 27.8% | 26.3% |
| Away win | 4.50 | 22.2% | 21.1% |
| Total | 105.6% | 100% |
The extra 5.6% is the bookmaker's cut (the bookmaker margin). Odds are a price designed to balance the book and protect the operator, not a neutral truth about the match. More on reading them in implied odds.
2. Gut feeling and pundit talk
"They're flying at home at the moment." "The dressing room has turned since the manager's outburst." "It's a derby, it'll end level." These lines are storytelling, not prediction. They sound like analysis because they use football vocabulary, but there's no measurable signal behind them.
3. The usual biases
We've written a full guide to the 7 common mistakes in football predictions, and traditional predicting tends to pile them up: recency bias, overrated head-to-heads, forgotten injuries, overreacting to the last result.
Traditional prediction isn't worthless. It picks up real signal, because trained eyes do see things. But it hits a ceiling quickly.
What xG really is
xG means expected goals: the probability that a given shot becomes a goal, calculated from the shot's characteristics: distance, angle, body part, the type of pass before it (cross, cut-back, through ball), the goalkeeper's position, how many defenders are in the way, the speed of the move.
A close-range shot straight after a one-two might be worth around 0.7 xG: it goes in about 7 times out of 10. A 30-yard strike from a tight angle might be worth 0.03 xG. Added up over a match, xG measures the real quality of the chances created, whether or not they went in.
For the maths and the history, read our complete xG explainer. In short: it has become the reference metric in professional football analysis and feeds most serious predictive models.
Why xG predicts better than the raw score
Over a single match, the score is noisy. A shot off the post, a world-class save, an own goal: rare events that can flip a 1-0 into a 0-1 without saying much about which team was better.
xG measures the underlying quality of play. Added up over several matches, it gets much closer to a team's true level than results do. A team that keeps winning while creating less than its opponents is living on borrowed time; a team that keeps losing while creating more is usually due a better run.
Three cases where xG beats the traditional reading
These are hypothetical cases, built to illustrate the logic, not real matches.
Case 1: the 2-1 that hides 1.8 vs 2.4
Imagine a league match that ends home side 2, visitors 1. The traditional tipster files the home side under "knows how to win" and expects the visitors to lose again next week. But the numbers say something else:
| Indicator | Home side | Visitors |
|---|---|---|
| Shots on target | 4 | 8 |
| Total xG | 1.8 | 2.4 |
| Big chances (xG > 0.3) | 3 | 6 |
| Goals minus xG | +0.2 | −1.4 |
On the quality of chances, the visitors deserved more. The home side won thanks to an inspired goalkeeper and clinical finishing. That can happen once; it's much harder to repeat week after week. The xG reader sees the pattern; the traditional reader gets fooled.
Case 2: the 0-0 that wasn't really a 0-0
A home side draws 0-0. The morning-after verdict: "boring game, the visitors defended brilliantly". But the xG tells another story:
- Home side: 2.8 xG, 18 shots, 7 on target, 4 big chances missed.
- Visitors: 0.4 xG, 3 shots, 1 on target, no big chance.
The home side completely dominated and missed an unusual number of chances. Roughly speaking, a team creating 2.8 xG fails to score only about 6 times in 100, and a 0-0 with those numbers happens about 4 times in 100. The traditional reader thinks the team is "stuttering"; the xG reader expects their finishing to return to normal. Bear in mind that bookmakers and markets watch xG too, so this kind of reading is less of a secret than it used to be.
Case 3: the knockout tie
In a two-legged tie, the traditional reading leans on reputation and the first-leg score. The xG reading asks a different question: over the first leg and the recent matches, which side created better chances? A team that won the first leg 1-0 while being clearly outplayed on xG is more fragile than the scoreline suggests. That doesn't mean it will go out; it means the tie is more open than the aggregate score makes it look.
The xG traps: what it doesn't tell you
xG is not a magic wand. You need to know where it fails.
1. Sample size
Over a single match, xG is almost as noisy as the score. A 0.4 xG shot that goes in is worth 1 real goal but 0.4 xG. Wait for around ten matches before trusting a team's xG trend. Below that, it's a clue, not a verdict.
2. xG per shot vs total xG
Many analysts publish xG per shot to compare the "average quality" of chances. Useful, but it ignores volume. A team with 8 shots at 0.18 xG per shot (1.44 xG in total) is more dangerous than a team with 3 shots at 0.25 (0.75 xG in total). Always look at both.
3. Special contexts
- Heavily rotated cup line-ups: a cup match with eight changes doesn't carry the same predictive value as a league match.
- End-of-season dead rubbers: teams with nothing to play for can produce flattering xG without real intensity.
- Red cards: a team playing against ten men for an hour piles up xG that says little about the next match.
4. Set pieces
Standard xG slightly undervalues the edge of teams that are genuinely strong at set pieces: the quality of their delivery and routines isn't fully captured by the shot alone. Many models now track set-piece xG separately to correct for this.
5. Regression to the mean works both ways
A team that outscores its xG over three matches is probably riding its luck and will come back down. But it may also have a genuinely elite finisher who consistently beats the average. Telling luck from talent takes a lot more data than a handful of matches.
How Elofoot combines both worlds
We don't think you should throw out every human insight and go 100% xG. The best reading combines them: xG for the quality of play, context for what the numbers can't see yet (absences, line-ups, the fixture list).
That's how an Elofoot prediction is built. A statistical model estimates each team's expected goals from their xG, form, standings and home advantage, turns them into probabilities with the Poisson distribution, and blends in a real-money market signal on big matches where one is available. The AI then explains the scenario in plain English, without ever producing a number itself. The result:
- Coherent probabilities: the three 1X2 outcomes always total exactly 100%.
- A written scenario: the most likely score and how the match should unfold, with the key factors (form, xG, absences).
- Honesty about uncertainty: open matches look open, and missing data is not made up.
The full chain is in how Elofoot works, and why a scenario says more than a flat 1X2 is explained in football predictions vs 1X2 odds.
Summary
| Approach | Strength | Weakness |
|---|---|---|
| Traditional prediction | Human reading, tactical intuition | Cognitive biases, quick ceiling |
| xG model alone | Measures chance quality, stable over time | Blind to special contexts, set pieces |
| Combined model (Elofoot) | Coherent probabilities + written scenario | Still needs reading: not a "miracle tip" |
Going further
Want to try it on a real match, with xG, a written scenario and the probable line-ups? Public predictions are on AI football predictions, including the Premier League predictions, and your first full prediction is free when you sign up, no card needed.
Responsible gambling. Elofoot's predictions are statistical estimates, not certainties or betting advice. Betting is for adults only (18+ in the UK and most countries, older in some) and carries real risks: debt, dependency, isolation. Never stake money you need and never chase losses. If gambling stops being fun, talk to someone you trust or a local support service. In Great Britain, the National Gambling Helpline (run by GamCare) is free and open 24/7 on 0808 8020 133.
