How FC Kickoff works

We forecast football matches, publish every forecast before kick-off, and score ourselves in public against the betting market. Here is exactly how.

The forecasts

The FC Kickoff Model is a combination of our staff's expertise and our AI prediction model, which is based on a wide range of statistics and variables. For every match it reads the same information a careful analyst would: the league table and recent form, team statistics, injuries and suspensions, recent transfers, how much rest each team has had, the weather at the stadium, and the latest team news. It then gives a probability for a home win, a draw and an away win.

The model never sees bookmaker odds. Its forecast is its own view, and that is what makes the comparison with the market meaningful.

Forecasts are made up to three days ahead. A final version, which includes the latest team news, is made in the last 48 hours before kick-off. The version closest to kick-off is the one that counts.

How we score

• Correct pick. Our pick is the outcome with the highest percentage. If two outcomes have the same percentage, the pick is the home win, then the away win, then the draw. It is correct if that outcome happens after 90 minutes. Extra time and penalties do not count.
• Brier score. A pick only says who we think wins. The Brier score also checks how good the percentages were. A team we gave 70% that loses hurts more than a team we gave 40%. Lower is better. Guessing 33% every time scores 0.667.
• Same matches for both. We compare the model and the market only on matches for which both have a forecast. We also show the model's score on all matches.

See the current numbers on the scoreboard. They are calculated from every finished match, and we do not remove the ones we got wrong.

The market

"The market" is what bookmakers expect. For each match we take the win, draw and away odds of up to 12 bookmakers, average them, and convert them into percentages with the bookmakers' margin removed. See which bookmakers.

Adjusted forecasts

Sometimes the editor disagrees with the model, for example when important news arrives late. Before kick-off the editor can change the percentages and the score. Such a forecast is labelled Adjusted, and the original model forecast is shown next to it. Nothing can be changed after kick-off. The scoreboard always scores the model itself, so adjustments never improve or hurt the model's record.

Visitor votes

You can pick an outcome yourself on each forecast. We use one anonymous cookie to remember your votes. It is a game, not data: the numbers are small and anyone can vote.

What to keep in mind

• A forecast is a probability, not a promise. A 60% forecast should be wrong four times in ten.
• The market currently scores better than our model. We show that because you should be able to check us.
• The sample is still small, so the numbers move from week to week.
• An AI model can miss things, such as a late injury or a lineup surprise.

Responsible gambling

Not in this list? Contact the gambling help service in your own country. We checked these services in September 2026. If something has changed, tell us and we will update it.