Football models lean on four kinds of data: expected goals (xG), recent form, performance indicators (possession, shots, duels) and market odds. The limit comes from the sport itself. Goals are rare, and research shows that the fewer decisive events a sport has, the more chance decides the result.
What models look at in football
The same four families of data turn up in football research again and again. AI didn't invent any of them. What models add is the ability to combine them across huge volumes.
| Family | What it captures | Where it falls short |
|---|---|---|
| Expected goals (xG) | The quality of the chances created and conceded | Says nothing about how well a given squad finishes |
| Recent form | Momentum over the last few match weeks | Depends on how strong the opponents were |
| Performance indicators | Possession, shots, duels, pressing | The link with the result varies |
| Market odds | The information bookmakers have already pooled | Ties the model to the consensus it's meant to judge |
Look at that last row. Plenty of studies feed odds into the model, because odds condense information that's hard to rebuild any other way. The price ends up as both a benchmark and an input, so a gap between the model and the price tells you less than it seems.
This page sticks to football: xG, low scoring, the weight of the draw. For AI applied to predictions across every sport, head to AI and sports predictions.
xG, and what it can't tell you
Expected goals estimate the chance that a shot becomes a goal, based on where it's taken, how the move built up and the context. Add them up over a match and you get a measure of attacking and defensive performance that doesn't depend on the final score.
What's it for? Mostly to correct how you read a result. A team that wins 1-0 with 0.4 xG against 2.1 hasn't played the same match as a team that wins 1-0 with 2.3 against 0.5.
The limits are just as clear. xG is an average built on a long history. It knows nothing about one player's finishing. And over a single match, the gap between expected goals and goals scored stays wide.
Why football resists modelling
Football has few decisive events, so chance weighs heavily by construction.
Bunker and Susnjak wrote a review that pulls together more than twenty years of work on predicting team-sport results. They're clear on this point. Two findings, checked against the text of the review:
- football is the most studied sport in the whole body of work, yet the best accuracy recorded for it is 78%, which puts it only fifth among the sports compared;
- some higher-ranked sports, such as rugby union, owe part of their lead to less balanced competitions. The review stresses that how deep the competition is matters at least as much as the sport itself.
The review also cites work suggesting that pure chance can decide the outcome of up to 18% of the matches in a season. Nearly one in five.
A word of caution. These figures come from research on predicting results. They aren't a success rate you can apply to a bet. A bet is judged against a price and its margin, not against a share of right answers.
Responsible gambling: a model doesn't remove the risk. Sports betting carries a risk of financial loss. Set yourself a budget, don't chase your losses, and use the deposit limits and self-exclusion tools available. Learn more about responsible gambling.
The draw is a special case
Football is one of the few big sports with a third outcome, and it's the hardest one to model.
A draw has no statistical signature of its own. It doesn't signal dominance or a lopsided match. It's a balance, and a balance can come from very different set-ups. Several studies in the literature sidestep the problem by working with two classes instead of three. The accuracy on show goes up mechanically, yet nobody understands the match any better.
So when a tool quotes you a figure, check how many outcomes it covers. Accuracy over two outcomes and accuracy over three can't be compared.
What AI brings to football, and where it stops
What does it do well? It works through whole seasons of data, compares indicators from one match to the next on the same footing, and follows what the odds are doing around the clock. It can also estimate a probability for each outcome, with an error margin that stays wide in a low-scoring sport. The result of a match is something it doesn't know. The exact score, even less.
That last limit shapes how OddScore presents things. The starting point is odds from several bookmakers with the margin removed, and how they move. The 0 to 100 signal and the AI reading sum up what the market says. They don't say how a match will end. In a sport this ruled by chance, I'd rather spell that out than leave you to guess.
This page stays on the data and models side. For the human method of analysing a match (1X2 market, weight of the draw, fixture list), go to football prediction. The markets the platform actually tracks are covered in football odds.
Back to the AI and sports predictions guide →
Back to the complete sports predictions guide →
Put it into practice
Apply this guide to real odds.
See a market reading
A real match with its odds, movements and 0-100 signal.
Open the demoMargin calculator
Measure a bookmaker's margin on any market.
Calculate the marginDig into the market
Odds movements are only part of the story. Here are the next topics to read.
AI and sports predictions
What AI can do with data, and why it never turns into a certainty.
Understand the role of AIFootball prediction: how to analyse a match
The 1X2 market, how much the draw weighs, form and the fixture list.
See the football guidePoisson and xG models in football
How a statistical model turns xG into score probabilities.
See the modelsFootball odds: the markets OddScore tracks
1X2, over/under goals and the competitions covered.
See the marketsSports predictions: analyse before you bet
The full guide: data, odds, reliability, bias and where AI fits.
Read the guideThe limits of AI predictions
Patchy data, black boxes, and too much trust in a number that looks precise.
See the limitsFrequently asked questions
Can an AI model predict an exact score?
Not realistically. At best a model gives a spread of possible scores, and even the most likely one only gets a small slice of it.
What is xG and what is it for?
Expected goals measure the quality of the chances a team creates, not the goals it scores. It helps you tell a strong performance from a result squeezed out of two or three chances.
Why is football hard to model?
Few decisive things happen in a match. Bunker and Susnjak's review (2022) notes that low-scoring sports leave more room for chance, which caps how accurate models can get.
Are odds used as an input for models?
Often, yes. Several studies in the literature use them as a variable because they pack in so much information.
Does OddScore give tips?
No. OddScore compares odds from several bookmakers, strips out the margin and tracks how they move. The AI reading sums up what the market says, such as the side it favours or the line it expects. It isn't a bet to follow, and OddScore gives no staking advice.
Sources & methodology
Three things sit behind this page: an academic review of machine learning for predicting team-sport results, a recent systematic review of machine learning in sports betting, and the Gambling Commission and BeGambleAware for the UK side of safer gambling.
- Quote accuracy figures as the review publishes them, sport by sport.
- Never present an academic model's accuracy as a success rate you can apply to a bet.
- Keep estimating a probability separate from picking an outcome.
- Label work that hasn't been peer reviewed as preprints.
- Bunker, R. & Susnjak, T. (2022). The Application of Machine Learning Techniques for Predicting Match Results in Team Sport: A Review. Journal of Artificial Intelligence Research, 73, 1285-1322.
- Galekwa, R. M., Tshimula, J. M., Tajeuna, E. G. & Kyandoghere, K. (2024). A Systematic Review of Machine Learning in Sports Betting: Techniques, Challenges, and Future Directions. Preprint arXiv 2410.21484 (not peer-reviewed).
- Gambling Commission: Public and players
- BeGambleAware