Football Intelligence Prediction Accuracy
Real historical stats, including losing runs — not cherry-picked highlights
Most football prediction sites publish their best periods and hide their losses. Football Intelligence settles every AI pick — wins and losses alike — and this page explains what the numbers actually mean. Statistical models are probability tools, not guarantees. 18+ | Not financial advice.
Why accuracy transparency matters
In the YMYL category of betting tools, accuracy claims without methodology are meaningless. The majority of football tip services cherry-pick winning runs, quote win rates without odds context, or stop publishing when the model underperforms.
Football Intelligence publishes every settled bet. This page explains the honest performance picture: the overall settled ROI is negative (approximately −7%), and even the risk-gate-filtered subset has not yet established a positive edge. The risk gate rejects the historically worst (market × odds) bands; the goal is closing-line value over time, which we are still measuring — it is not proven. We publish this rather than hide it.
How accuracy is measured
Two metrics are tracked:
- Win rate (hit rate): Percentage of settled picks that win. Useful secondary metric but misleading without odds context.
- ROI (return on investment): Net profit divided by total amount staked. This correctly weights both frequency of wins and size of return per winning bet.
All picks are settled at 1 unit staked per pick (flat staking). Only WIN/LOSS records from the ai_picks database table are included in ROI calculations.
Accuracy by market type
| Market | Odds band | Win rate | Status |
|---|---|---|---|
| Over 2.5 / BTTS | 1.50–1.85 | ~63% | Published (LOW) |
| Over 2.5 / BTTS / Win | 1.85–2.10 | ~50% | Published (MEDIUM) |
| Win / Draw | 3.00–3.50 | ~29% | Published (HIGH) |
| Win | 2.10–3.00 | — | Rejected by gate |
| Win / Over 3.5 | >3.50 / any | — | Rejected by gate |
Win rates are approximate and based on 380+ historical settled bets. Past performance does not predict future results.
What probability calibration means
Calibration is distinct from win rate. A Poisson model is well-calibrated if, across all events it assigns a 60% probability to, approximately 60% of those events actually occur. Football Intelligence Poisson model has historically calibrated within approximately ±2% of actual outcome frequencies across main European leagues. This means model probabilities are reliable for detecting mispriced bookmaker odds. Calibration does not mean any individual match is predicted correctly.
Model limitations
- Poisson independence assumption: Home and away goals treated as statistically independent — not perfectly true when match dynamics shift after a goal.
- Early-season small samples: In the first 4–6 matchweeks, Bayesian regularisation defaults toward league average.
- Rotation risk: Squad rotation for cup or European matches is not reflected in advance team-selection data.
- Transfer window: Significant transfers may lag by 1–2 matchweeks before appearing in model attack and defensive ratings.
- Cup and European matches: Home advantage applied as a reduced multiplier for neutral-ground fixtures — a simplification.
How the risk gate was built
After settling 380+ bets, Football Intelligence ran a systematic analysis of which market type and odds band combinations had historically positive ROI. The gate rejects any pick falling into a historically unprofitable band: Win at odds 2.10–3.00, Win above 3.50, and Over 3.5 goals at any odds.
The gate is conservative by design. It will miss some genuine value bets in rejected bands. For the full mathematical derivation, see the methodology page.
Common questions
What football prediction accuracy is realistic?
For match outcome predictions from a calibrated Poisson model, realistic long-run win rates are 40–55% depending on the odds band. Win rate without odds context is not meaningful — ROI is the correct metric. Win rates above 70% claimed by any service should be treated with extreme scepticism.
What is the difference between win rate and ROI?
Win rate is the percentage of bets that return a profit. ROI is net profit divided by total staked. A 60% win rate at odds of 1.50 produces ROI of (0.60 × 1.50 − 1) = −10%, a losing strategy despite most bets winning. Always use ROI as the benchmark.
Which football prediction site shows real accuracy stats?
Most services publish only their best periods or hide losing bets. Football Intelligence settles every AI pick in the database and tracks ROI across all outcomes. The historical overall ROI is approximately −7%. Published picks are the risk-gate-filtered subset (historically less-negative bands); a positive edge is not yet proven — we are still measuring closing-line value.
Check today's AI picks
The AI picks tab shows only picks that have passed the risk gate. Each includes Poisson probability, value edge, Kelly stake recommendation, and Claude AI analysis.
View today's AI picksFree plan · No credit card · 18+