How our predictions work

1. Expected goals for every match

For each team we measure how many goals they score and concede per game this season, at home and away, and compare that with the league average. That gives an attacking and a defensive strength for each side. Multiplying them out gives the number of goals each team is expected to score in this match.

2. From goals to chances

Goals in football follow a well-known pattern (a Poisson distribution). From each team's expected goals we work out the chance of every scoreline from 0–0 to 10–10. Plain Poisson underrates low-scoring draws, so we apply the Dixon-Coles correction to 0–0, 1–0, 0–1 and 1–1. Adding the scorelines up gives the chance of a home win, a draw, an away win, the most likely scores, over 2.5 goals and both teams scoring.

3. Blending with the market

A goals-only model can't see everything: injuries, rotation or a new manager. The bookmakers' prices do reflect those things, so we blend our chances half and half with the bookmakers' own view, with their margin removed. That keeps tips grounded, and a tip only counts as value when the best price still beats that blended view.

4. Goalscorers, corners and cards

Each player's share of his team's goals this season, scaled by minutes played, gives his chance to score if he plays. Corners and yellow cards come from each team's recent match statistics, for and against, so they appear once a team has played a few games.

5. One tip, across markets

For every match we compare our chance with the best price in each market: match result, double chance, over or under 2.5 goals, both teams to score and anytime goalscorer. The tip is the one with the biggest edge, leaning towards likelier outcomes. We never tip prices under 1.30 or chances under 25%. When nothing is value, we show the most likely tip that still pays a sensible price, and say so.

6. Form, head to head and team news

Tap any match to see each team's last five results, league position and goals per game, the last five meetings between them, and reported injuries and suspensions. The model doesn't adjust for absences, so weigh them yourself.

7. The AI layer

For every match in the next few days, an AI analyst (Anthropic's Claude) reads all of that data, then writes a short preview and explains the tip in plain words. It doesn't pick the tip or set the confidence score: those come from the model. It only works from that data: it doesn't invent team news. You can also ask it questions about this week's matches in Ask the AI.

8. Keeping score

Every tip is logged automatically before kickoff, at the best price we showed, and settled after full time. Wins and losses both count. The results are on our track record page, and The Brain shows whether our chances come true as often as we say.

Bet responsibly

No model wins every bet. Only bet what you can afford to lose, set limits, and stop if it stops being fun. Find free support.