Step 1: Rate every team
The model rates each team's attack and defence using results from dozens of competitions. Scoring three against a strong defence counts for more than scoring three against a weak one, and recent matches count for more than old ones. Because teams are linked through shared opponents, a Champions League tie can compare clubs from different countries, and a qualifier can draw on a nation's other matches.
Step 2: Turn ratings into probabilities
The ratings give each side's expected goals for the match. The Poisson distribution, which describes how often a rare event happens in a fixed time, converts those into the probability of every plausible scoreline. A Dixon-Coles adjustment corrects the plain model's tendency to under-predict low-scoring draws, and a calibration step keeps the stated percentages honest.
Step 3: Check against the betting market
Bookmaker odds are among the most accurate public forecasts in football, because they react to injuries, line-ups and team news that our ratings cannot see. When prices are available, usually in the last week before kickoff, we remove the bookmaker's margin and blend them into the forecast. Each match page says when this has happened.
Data Sources
Fixtures, results, odds and final scores come from a professional football data provider. The written match analysis is generated by AI after the forecast is made, so it explains the numbers rather than setting them. If a team has too little history to rate, such as in some youth tournaments, the model uses form averages or an AI estimate of expected goals, and the match page says so.
The ratings themselves do not see injuries, managerial changes or weather. The market blend picks up some of that, but only when odds exist.
From Probability to Decision
Not every prediction is a recommendation. Each match is graded as one of three decisions:
- Recommended Pick: confidence of at least 58%, a gap of at least 12% to the next outcome, a draw probability of no more than 30%, and a model edge of at least 0.52.
- Watch Only: there is a signal (confidence of at least 50% and a gap of at least 8%), but it falls short of a recommendation.
- No Pick: the probabilities are too close to call.
How we test it
We tested the model on past matches it had never seen. Its settings were fitted on matches up to July 2024, then it predicted August 2024 to November 2026, using only information available before each game. This is a simulation, kept separate from our live verified track record, and past results do not guarantee future ones.
On the 4,370 matches where both could be compared, it picked the right result 51.4% of the time, against 47.3% for our previous model. For comparison, always picking the home team gets about 45%, and bookmakers' final odds typically manage 53 to 55%.
| Competition | Matches | Previous model | Current model |
|---|---|---|---|
| Top 5 European leagues | 3,051 | 48.0% | 52.1% |
| UEFA club competitions | 343 | 51.9% | 57.7% |
| Major League Soccer | 976 | 43.6% | 47.0% |
| International football | 378 | 51.3% | 55.3% |
The stronger the call, the more often it came in
| Type of pick (club football) | Matches | Share of games | Correct |
|---|---|---|---|
| Recommended Pick | 1,098 | 17.1% | 73.3% |
| Watch Only | 1,946 | 30.2% | 56.6% |
| No Pick | 3,395 | 52.7% | 42.9% |
When it says 65%, does it mean 65%?
Mostly, yes. This is what makes the percentages useful: they can be read at face value.
| Stated confidence | Matches | Average stated | Actually correct |
|---|---|---|---|
| 30-40% | 1,080 | 37.8% | 35.6% |
| 40-50% | 2,315 | 44.8% | 46.2% |
| 50-60% | 1,609 | 54.5% | 55.6% |
| 60-70% | 924 | 64.3% | 65.5% |
| 70-80% | 391 | 74.2% | 78.0% |
| 80%+ | 120 | 84.5% | 85.8% |
Important Disclaimers
While our model provides a statistically sound baseline for match outcomes, football is inherently unpredictable. Red cards, controversial decisions, and sheer luck cannot be perfectly modelled. All percentages and predictions provided by UTABIRI are purely informational.
No outcome is guaranteed. We strongly advise applying your own knowledge and behaving responsibly when using these predictions.