AI Football Predictions

We built our prediction engine to look at the same things a good analyst would look at before a match, just faster and across a lot more fixtures than one person could cover. For every game, we feed the model recent form for both sides, head-to-head history, player availability where it’s known, and match context like competition stage and home advantage.

From there, the model produces three things: a 1X2 probability split (home win, draw, away win), a predicted correct score, and calls on secondary markets including both teams to score and over/under 2.5 goals. We generate these automatically ahead of kickoff, and if something changes close to the match, like a late team news update, we refresh the numbers rather than leave a stale prediction up.

We’re upfront that this is a statistical model, not a crystal ball. It’s built to spot patterns across thousands of matches, not to guarantee any single result.

Our model runs at 85% prediction accuracy across our football coverage. That number comes from checking every prediction we’ve published against the actual final result, not just the ones that happened to land.

We don’t hide the misses. Every report in this archive shows what we predicted next to what actually happened, so instead of asking you to trust an accuracy figure on faith, we’re leaving the full track record on the page for you to check yourself. If you scroll back through older fixtures, you’ll see calls we got right and calls we got wrong, side by side.

One thing worth being clear on: 85% is measured against our own scoring method (see “How We Track Our Own Accuracy” below), and accuracy naturally varies by league, by how settled a team’s form is, and by how far out from kickoff we’re predicting.

Each report gives you three probabilities: the chance of a home win, a draw, and an away win, based on everything the model weighed for that fixture. These add up to 100%, but they’re likelihoods, not promises, a 70% favourite still loses roughly three times in ten.

The predicted correct score is the single most likely scoreline our model landed on, not necessarily the “safest” prediction. Once the match has been played, we show the actual final score right alongside it, so you can see how close we got without having to dig for the result elsewhere.

If a match has secondary market calls (BTTS, over/under, and so on), those sit underneath the headline probabilities in the same report. We keep the layout consistent across every fixture so you’re not relearning how to read it match to match.

We cover the world’s top leagues in depth, Premier League, LaLiga, Serie A, Bundesliga and Ligue 1 among them, plus 1,200+ more competitions worldwide. That range is deliberate: we didn’t want this to be another site that only predicts the five or six leagues everyone already covers.

If you follow a smaller league that rarely gets forecast anywhere else, there’s a decent chance we’ve got it. Use the league filter above to jump straight to the competition you care about instead of scrolling the full archive. We’re also building out dedicated pages per league, now covering the full Big Five: LaLiga, Premier League, Serie A, Bundesliga and Ligue 1, with more leagues to follow.

We’d rather tell you where our model’s limits are than let you assume it’s more precise than it is. A few things it doesn’t account for well: last-minute team news that breaks after we’ve generated a prediction, refereeing decisions and VAR calls that swing individual matches, and motivation factors that don’t show up in form data, a mid-table side with nothing to play for versus one battling relegation, for example.

The model also weighs historical patterns more heavily than it weighs a single manager change or a new signing’s first few weeks, simply because there isn’t enough data yet for those situations to be reliable signals. Early in a season, or right after a big squad shake-up, treat our predictions as a rougher guide than usual.

None of this means the numbers are wrong, it means they’re probabilities built from what’s measurable, and football has always had room left over for the immeasurable.

Every prediction we publish gets logged before kickoff, so we’re never able to quietly revise a call after the fact. Once a match finishes, we mark the prediction against the real result and roll it into our running accuracy figure.

We count a correct score prediction as a hit only when we get the exact scoreline right, not just the winning side, and we track 1X2 accuracy separately from correct-score accuracy since they’re different levels of difficulty to get right. That’s part of why you’ll see us quote 85% for overall directional accuracy while individual markets can run higher or lower.

We publish the archive itself as the evidence. If you want to check our working on any league or stretch of matches, the reports are all there to look through.