How we make our football predictions

Step 1: where does every prediction start?

With the market. We cover the Premier League, Champions League, La Liga, Serie A and Bundesliga. For each match we gather odds from 20+ bookmakers, remove their margin and turn the prices into fair probabilities. Combined bookmaker prices are a strong estimate of what is likely to happen, so we treat them as the baseline rather than pretending to know better from scratch. It also means every pick can be compared directly with the prices on offer. Our how to read odds guide shows how the margin is removed, with a worked example.

Step 2: what does the AI research agent check?

Between 12 and 60 hours before kickoff, an AI research agent works through the same written runbook for every match. It covers both teams and records “no news found” rather than skipping an item:

• Injuries and suspensions, and how important each missing player is.
• Likely lineups and rotation, including fixtures just before or after.
• Form in context: recent results and who they came against.
• Head-to-head meetings, counted only when a pattern repeats.
• Team news, from club sites and established outlets.
• Conditions such as weather, travel and fixture congestion.

Step 3: how much can the research change the odds?

Only a little, and only with evidence. Every fact needs a cited source; without one it is ignored. The agent may move the market probabilities only within capped limits, so a dramatic story can’t override what the market already knows. For example, if a first-choice striker is ruled out with an injury reported by the club, the agent can trim that team’s win probability and the chance of goals, but only by a limited amount, because the market has often already reacted to the news. If no selection meets our rules, the answer is no pick, and the match keeps its preview and market view.

Step 4: who checks the analysis?

A separate verifier agent checks every claim and every source before a pick is published. Drafts that fail are rejected and sent back; if a match’s analysis is rejected twice, it gets no pick at all. We would rather publish fewer picks than unsupported ones. Tipsters who post a pick on every match have no way to say “we don’t know”; we do.

Why are some picks withdrawn?

About two hours before kickoff, when lineups are confirmed, a lineup check runs. If a key player the pick depends on is unexpectedly out, the pick is withdrawn. Withdrawals stay visible on the site and in the record: we never quietly remove a pick, and we never withdraw one simply because the odds moved. If a better selection is still available, a new pick may be published alongside the withdrawn one.

What does “confidence” mean?

The percentage on each pick is our estimated probability that it wins, not a promise. A pick rated 75%, the level for our bankers, should lose about one time in four over a long run if our estimates are good. The same numbers drive every hub, from today’s predictions to over 2.5 and double chance. Our daily 2-odds slip and weekend combo are built only from published picks and show their combined probability, explained in the 2-odds guide.

Why do we publish our losses?

Because a record you can’t check is worthless. Every pick is timestamped with the odds at the moment it was published and graded after full time. Picks can’t be edited or deleted. The full record is on our results page and can be downloaded as a CSV. It is still young, so judge it over hundreds of picks, not a handful. We are not a bookmaker and don’t take bets; read more about us.

No method removes the uncertainty of football: deflections, red cards and late goals decide matches every week. No bet is ever sure: only stake what you can afford to lose, you must be 18+ to bet, and our responsible gambling page lists tools and support.