How Sportly AI works
No magic: what the model looks at, where the percentages come from and how we keep ourselves honest.
What data does the model use?
The model looks at what actually moves a match: recent results and form for each side, head-to-head record, home and away performance, goals scored and conceded, expected goals (xG), corners and cards.
We don't feed it rumours or "insider tips". When some data for a match is missing, the model works with what it has — and the site simply hides that block.
How is an outcome probability calculated?
For every match the model estimates how many goals each team is likely to score and turns that into the chances of a home win, a draw and an away win. Together they add up to 100%.
So "60% home" doesn't mean the home side will win. It means that in ten similar matches they'd win about six.
What does "AI confidence" mean?
Confidence tells you how clearly the model sees a favourite. When the teams are close, confidence is low even if one outcome is technically ahead.
High confidence isn't a guarantee. Football is a low-scoring game and upsets happen all the time.
How do we measure prediction accuracy?
We only count finished matches with a known result and check whether the AI's main pick landed. Matches without a result don't count either way.
How we publish a prediction. If the main outcome's probability in the prediction snapshot is 55% or higher, we show the result prediction. If it is lower, the match is evenly balanced and we show the double chance instead — the two most likely outcomes (1X, X2 or 12) and their combined probability. Friendlies get a total-goals call only. Result accuracy counts only matches where we showed a result prediction; friendlies are left out of result and double-chance accuracy. A market appears in the table once it has at least 100 finished matches.
Friendlies are shown with a total-goals call but no result prediction: line-ups and motivation are unpredictable. They are left out of result and double-chance accuracy but counted for totals. Why.
Markets where the model's accuracy is below the baseline (50% for binary markets) are not published on the site and are being improved. Currently being improved: Both teams to score, Total corners, First-half goals, Handicap, Player to be booked, Red card, Total yellow cards.
Sportly AI calls the match result right 67% of the time, and rules out the losing option 77% of the time (last 30 days).
| Market | Accuracy | Matches |
|---|---|---|
| double chance | 77% | 448 |
| match result (1X2) | 67% | 181 |
| total 2.5 | 56% | 368 |
Calibration: accuracy by confidence level
56% On high-confidence picks, Sportly AI is right 56% of the time (325 matches over 30 days).
Accuracy by AI confidence level
| High (70% and above) | 325 matches | 56% |
| Medium (55–69%) | 73 matches | 48% |
| Low (below 55%) | 50 matches | 42% |
| All picks | 448 matches | 53% vs 33% for picking an outcome at random |
By competition
| UEFA Nations League | 22/35 | 63% |
| Taca de Portugal | 13/21 | 62% |
| Spanish LALIGA | 8/12 | 67% |
| Dutch Eredivisie | 8/12 | 67% |
| Saudi Pro League | 6/11 | 55% |
| Spanish Copa del Rey | 6/10 | 60% |