How Oddazo predictions work

Oddazo turns real match data into probability estimates. This page explains exactly what goes in, what comes out, and how to read the numbers. Not every data point is available for every match — when something is missing we say so and lower our confidence.

1. The data we use

• Fixtures & kickoff times

  From our football data provider. Matches that have started are never predicted.

• Bookmaker odds

  Pre-match prices (currently Bet365 via our data provider), stored with the time they were captured. With the margin removed they give the market's view.

• Goal records

  Season home/away goals scored and conceded when available, plus results we store ourselves every day.

• Recent form

  Last five results, goals for and against.

• Injuries & suspensions

  When the provider reports them for that fixture.

• Head-to-head

  Past meetings, used only when there are at least three.

• Our own team ratings

  Elo-style strength ratings built from results stored in our database (introduced October 2026, still growing).

Not used today

• Expected goals (xG) — our current data plan does not supply it, so no prediction uses it.
• League position and standings — not available on our current data plan.
• Lineups, weather, referee or betting-exchange data.

2. The model

1. We estimate how many goals each team is likely to score from their attack and defence records, adjusted for home advantage and recent form.
2. A Poisson goals model turns those expected goals into a probability for every market we cover: 1X2, Double Chance, Draw No Bet, Over/Under, Both Teams To Score and first-half goals.
3. When team statistics are thin, the estimate leans on bookmaker odds with the margin removed, and the pick is labelled as a market-based estimate.
4. Injuries and head-to-head trends make small adjustments.
5. Confidence is then adjusted using our own track record: if past picks in a market type won less often than predicted, confidence for that market type is reduced.

3. Choosing a pick

For each match the markets are ranked by model probability, data quality and whether real bookmaker odds exist. The best one becomes the match's main pick. In slips and the Bet Builder, very low odds (under 1.10) are skipped because they add risk for almost no return. Matches with too little data get no pick at all — “insufficient data” is better than a guess.

4. Reading the numbers

Example: Over 1.5 goals · odds 1.30 · probability 81% · confidence 71/100. The model expects this outcome about 8 times in 10 over many similar matches, so it should still lose roughly 1 time in 5. A confidence of 71 means the data behind it was fairly complete, not that it is 71% certain.

5. Where AI fits in

Oddazo AI in the Bet Builder and Match Analyzer reads your request (“3 games, low risk, 2–5 odds”) and turns it into filters. Every selection, probability and price it shows comes from the model and database above — it cannot invent statistics, injuries or odds.

6. Checking ourselves

• Every published pick is timestamped and settled automatically. Wins and losses stay on the Results page.
• Picks are never published or changed after kickoff, and every change before kickoff is archived.
• We track calibration (do 70% picks win about 70% of the time?) and closing-line value (did we get better odds than the final price before kickoff?).