72.1% of our predictions landed.
We backtested the model across the 2024 and 2025 seasons, scoring 129,776 predictions against the final result of every match. 72.1% of them came in.
Premier League
Man City
1.85Premier LeagueMan City1.85The process
How a prediction comes together
Five stages between raw match data and the predictions on the card.
Pull the data
Form, head to head, season numbers, shots, corners, lineups, injuries and live bookmaker prices come in fresh for every fixture.
Crunch the stats
Raw feeds become rates the AI reads directly: scoring and conceding rates per side, BTTS frequency, clean sheet rates, corner counts.
Score the result markets
Match result (1X2), double chance (1X / X2 / 12) and half-time result. Each fixture gets a probability on every outcome.
Score the goals and corners markets
Both teams to score, total goals on every line from 1.5 to 4.5, team goals per side from 0.5 to 2.5, and corners from 5.5 to 10.5. Every line gets its own probability.
Compare to bookmaker prices
Each probability is checked against the implied price (1 ÷ decimal odds). Where the AI disagrees with the book, the gap is the edge. Predictions sort by probability or by edge.
The inputs
What the AI looks at
Eight blocks of data go in for every fixture. Some matter more for one market than another.
Recent form
The last several matches for each side, with goals, results and underlying stats.
Season averages
Goals scored and conceded per match, clean sheet rate, BTTS frequency and over 2.5 rate across the full season.
Match statistics
Shots, shots on target, possession, fouls, yellow and red cards from recent fixtures.
Goal timing
When sides typically score and concede. First-half versus second-half splits feed straight into the half-time and full-time markets.
Corners and set pieces
Corners taken and conceded by each side at home and on the road.
Head to head
Recent meetings between the two sides and the scoreline patterns that keep showing up.
Lineups and injuries
Confirmed starting elevens, suspended players and known absences. A missing first-choice striker or keeper shifts the call once the team news lands.
League context
The scoring and corner profile of the league the match is in.
And a few more we keep to ourselves
A handful of extra signals feed the model too. The eight above are the ones that shape most of the calls.
Value detection
How we find the edge
A prediction has value when our probability beats the bookmaker's implied one. The bigger the gap, the stronger the call against the market.
Model probability
72%
What the AI scores the prediction
Implied probability
55.5%
From a 1.80 bookmaker price (1 ÷ 1.80)
Edge
+16.5
Plotted as the green badge on the card
Negative edges are hidden. If the bookmaker has the price right or better than our model, the prediction still shows but the green badge stays off.
Coverage
Markets the AI predicts
Every fixture is scored across the lot. Each market gets its own page; the hub shows them mixed.
Timing
When the AI calls a match
A first probability is available around two weeks before kick-off. The AI then refreshes the call through the week as team news, lineups and recent results land. The prediction on the card the day before a fixture has had more inputs than the one issued ten days out, so it is worth checking back closer to kick-off if the call matters to you.
Frequently asked questions
How accurate are the predictions?
72.1% of 129,776 backtested predictions landed across the 2024 and 2025 seasons.
What data does the AI actually use?
Recent form, season averages, match statistics (shots, possession, corners, cards), goal timing patterns, head to head, confirmed lineups, injuries and per-league context, plus a handful of additional signals we don't publish.
When does the AI make a prediction available?
Around two weeks before kick-off. The call then refreshes through the week as team news, lineups and form numbers update, so the probability on the card the day before the match has had more inputs than the one issued ten days out.
What does the edge percentage on a card mean?
The gap between our model probability and the bookmaker's implied probability for the same outcome. If we score a prediction at 70 percent and the price implies 60 percent, the edge is plus ten. That is where the value sits.