How accurate are Footlab AI's football predictions?

Live numbers

All competitions combined, friendlies included. Random guessing would give about 33%.

Does Footlab AI beat the bookmakers?

The table compares, on the 1,421 served matches with known pre-match odds, how often Footlab AI and the bookmakers' favourite were right, and what a €1 stake on every pick would have returned at the average market odds. When odds exist for a match, the displayed 1X2 probability combines the Footlab AI model with the consensus of published odds, which carries most of the weight, because that is what proved most accurate: Footlab AI does not try to beat the market, it builds on it.

Wider measurement of October 5, 2026, over 19,583 matches with odds, all competitions combined: Footlab AI 52.7 %, bookmakers' favourite 52.8 %, the same pick in 98.5 % of cases. The model alone, without the odds, scored 50.5 %.

Can you make money betting on Footlab AI's picks?

No. A €1 stake on every Footlab AI pick would have lost 5.6 % at the bookmakers' average odds and 2.3 % at the best market odds, almost exactly like betting on their favourite (5.4 % loss): the bookmakers' margin, about 9.7 % per match, absorbs everything (measured on October 5, 2026 over 19,583 matches). Closing line value (CLV) is not measured: Footlab AI only records an average price, usually the day before the match. Footlab AI is built to understand a match, not to beat the market, and gives no betting advice.

Why does Footlab AI so rarely call a draw?

Because a draw is rarely the most likely outcome: its probability seldom exceeds 30%. Footlab AI picks the draw as the most likely outcome in only 0.2 % of matches, just like the bookmakers. The hit rate therefore starts with the draws, about 23.9 % of matches, missed in advance; excluding draws, the outcome shown comes true in 69.1 % of cases. The draw probabilities themselves are accurate: 24.8 % announced on average for 23.9 % observed (measured on October 5, 2026 over 19,583 matches).

How does Footlab AI measure the accuracy of its predictions?

Footlab AI records the last card served before kick-off, then compares it with the official result of the match no earlier than three hours after kick-off. Nothing is sorted, nothing is deleted. Separately, the model alone was tested once on matches played after its calibration period, to check that it does not simply recite the past.

Are the probabilities announced by Footlab AI calibrated?

The table compares, in bands of 10 points, the probability announced by the model with the frequency actually observed, across 1,718 evaluated predictions. A well-calibrated model aligns the two columns; a gap within one band remains expected while that band's sample is small, hence the sample column.

Does accuracy depend on the level gap between the two teams?

Yes, and the difference is clear: the larger the world Elo gap, the more reliable the prediction, and a match between clubs from two different leagues remains harder to predict than a domestic one. Each row states the number of evaluated predictions it rests on.

What is Footlab AI's accuracy by competition?

The table only lists the 12 competitions with at least ten evaluated predictions: below that threshold, a rate measures nothing. These rates rest on much smaller samples than the overall figure and are comparable neither with each other nor with that figure.

Can Footlab AI's past predictions be reviewed, including the wrong ones?

Yes: here are the last 30 evaluated predictions, in chronological order, with the probability announced before kick-off, the final score and the verdict, hit or miss. The list is not filtered on correct calls and no prediction is removed after the fact.

Announced probability = the probability the model gave to the outcome it deemed most likely, at the time of the last card served before kick-off.

Frequently asked questions about Footlab AI's accuracy