← Back

How BetAI365 Works

Our AI model, value framework, and staking logic — fully explained.

The AI Model

BetAI365 uses Google Gemini (2.5 Flash) as its reasoning engine, fed with live data from API-Football: fixtures, team form, head-to-head records, live odds, and injury reports. Every morning at 07:00 UTC the model analyses today's major-league fixtures and produces up to 6 high-confidence picks.

The model is instructed to prefer positive expected value over raw confidence — it will skip a 90%-confidence favourite if the bookmaker's odds leave no edge, and instead flag a 68%-confidence pick where the fair price is significantly better than market odds.

10-Factor Scoring

20%

Team Form

Last 5–10 match results, home vs away form, goal difference trend.

15%

xG Model

Expected goals from shots taken/conceded. Strips out luck from scorelines.

10%

Home Advantage

Venue-specific stats. Home teams win ~46% of top-league matches.

10%

H2H History

Head-to-head record for this fixture, weighted toward recent meetings.

10%

Injury & Availability

Confirmed team news, key-player absences, estimated impact on probability.

15%

Market Value

Comparing our probability to bookmaker implied odds to find positive EV.

10%

Market Movement

Odds drift vs sharp-money shortening signals informed bet activity.

2%

Weather

Rainfall, wind speed — affects high-scoring vs low-scoring outcomes.

3%

Referee

Card and penalty rates per official. Affects Over/BTTS markets most.

5%

Rest & Schedule

Days since last match, upcoming fixture load, cup vs league context.

Expected Value Framework

Every pick includes an EV% computed as: EV% = (p × (odds − 1) − (1 − p)) × 100, where p is the AI's estimated probability and odds is the best available bookmaker price.

No Value< 0%

Skip. The bookmaker has an edge over our model.

Low Value0–3%

Marginal edge. Small stake only.

Good Value3–8%

Clear edge. Standard stake.

Great Value8–15%

Strong edge. Larger stake.

Elite Value> 15%

Rare find. Maximum kelly fraction.

Kelly Staking

Recommended stake is the Kelly Criterion fraction: K% = (p × odds − 1) / (odds − 1). We display this as a percentage of your bankroll. We recommend using ¼ Kelly to reduce variance — full Kelly maximises long-run growth but produces large drawdowns.

Transparency Commitments

  • Every prediction is stored permanently — nothing is deleted after a loss.
  • Win rate and ROI are computed from all settled bets, not cherry-picked periods.
  • The AI model and prompt are disclosed (see admin panel for full prompt).
  • No hindsight editing — pick and odds are locked at generation time.
  • Gambling carries risk. Past performance does not guarantee future results.