The Kelly Criterion is a mathematical formula originally developed for information theory that has become the gold standard for optimal stake sizing in probability-based wagering. It calculates the exact percentage of your bankroll to stake on each bet to maximize long-term growth rate while avoiding ruin. For football prediction enthusiasts who have a genuine edge, the Kelly Criterion transforms that edge into the fastest possible bankroll growth.

The Kelly Formula

The Kelly Criterion formula is: f* = (bp - q) / b, where f* is the fraction of bankroll to wager, b is the decimal odds minus 1 (the net odds), p is the probability of winning, and q is the probability of losing (1 - p). For example, if our AI model assigns a 60% probability (p = 0.60) to a home win at decimal odds of 2.00 (b = 1.00), the Kelly stake is: (1.00 x 0.60 - 0.40) / 1.00 = 0.20, or 20% of the bankroll. This is considered full Kelly — most practitioners use fractional Kelly for added safety.

Why Full Kelly Is Too Aggressive

Full Kelly staking maximizes theoretical long-term growth but produces extreme bankroll volatility. A string of losses — which is statistically inevitable even with a genuine edge — can reduce the bankroll by 50% or more before recovery. For this reason, most experienced bettors use fractional Kelly: typically quarter-Kelly (25% of the full Kelly amount) or half-Kelly (50%). Fractional Kelly significantly reduces volatility while sacrificing only a moderate amount of long-term growth rate.

Requirements for Kelly Criterion Use

The Kelly Criterion requires accurate probability estimates — if your probabilities are wrong, the Kelly stake will be wrong too, potentially catastrophically so. Overestimating your edge leads to over-staking, which accelerates losses rather than profits. This is why calibrated AI models like ours are particularly valuable for Kelly-based strategies: well-calibrated probability estimates translate directly into appropriate Kelly stake sizes.

Practical Kelly Implementation

To implement Kelly Criterion with our AI predictions: take the model's probability for each outcome, apply the Kelly formula using the available odds, multiply by your chosen Kelly fraction (we recommend quarter-Kelly for most users), and use the result as your stake percentage. If the formula produces a negative value, this indicates no value exists at the available odds, and the optimal Kelly stake is zero — meaning you should pass on that prediction entirely.