Finding an edge is only half the job. The other half is deciding how much to stake on it, and getting that wrong can ruin even a winning model. Stake too little and a genuine edge grows your bankroll at a crawl. Stake too much and a normal losing streak wipes you out before the edge has time to pay. The Kelly criterion is the classic answer to this problem: a formula that sizes each bet in proportion to your edge, aiming to maximise long-term bankroll growth while, in theory, never risking ruin.

The formula for a simple two-outcome bet is easy to state. Take your estimated probability of winning, multiply it by the decimal odds, subtract one, then divide by the odds minus one. The result is the fraction of your bankroll to stake. In plain terms, bigger edges mean bigger stakes, and no edge means no bet. If your model says a team wins 55 percent of the time and the odds are 2.00, the formula suggests staking 10 percent of your bankroll. That already sounds large, and it is: full Kelly is aggressive, and its aggression is the main reason most practitioners modify it.

Full Kelly maximises long-term growth only if your probabilities are exactly right. In betting, they never are. Models are miscalibrated, markets move, and a small overestimate of your edge translates into a large overestimate of the correct stake. With full Kelly, a run of bad luck can halve a bankroll even when the underlying model is profitable, and few bettors have the stomach, or the capital, to ride that out. The mathematics is correct; the assumptions are not realistic for a human with imperfect information and finite patience.

The standard compromise is fractional Kelly: stake a fixed fraction, often a quarter or a half, of what the formula suggests. Half Kelly sacrifices some theoretical growth but dramatically reduces bankroll swings, and it forgives modest errors in your probability estimates. Many professionals go further, capping any single stake at a small percentage of bankroll regardless of what the formula says, and refusing to stake at all unless the edge clears a minimum threshold. These guardrails are not cowardice. They are the difference between a system that survives its first bad month and one that does not.

Kelly only works when fed honest probabilities, so it belongs at the end of a disciplined pipeline, not the start. First, build and validate the model. Second, check calibration, so the probabilities going into the formula mean what they claim. Third, de-vig the market odds to find the genuine edge. Fourth, apply fractional Kelly with a cap, and log every stake, edge estimate, and outcome. Review the log monthly: if your realised returns consistently undershoot what Kelly predicted, your edge estimates are too optimistic, and the fraction should come down further.

Finally, remember that Kelly assumes you can keep betting indefinitely with the same edge, which is rarely true. Bookmakers limit winning accounts, markets sharpen, and edges decay. Kelly is a staking tool, not a promise. Used with calibrated probabilities, a conservative fraction, and a hard cap, it turns a model's edge into steady, survivable growth. Used with inflated probabilities and full stakes, it turns a good model into a short story. Choose the version that keeps you in the game.

The most common Kelly mistake is feeding it probabilities from an uncalibrated model. If your model says 60 percent when the true chance is 52 percent, Kelly will recommend a stake far larger than your real edge justifies, and repeated overstaking is how winning models still ruin bankrolls. Calibrate first, stake second, and never reverse that order. The second mistake is ignoring bankroll definition: Kelly fractions apply to your dedicated betting bankroll, not to your total savings, and mixing the two inflates every stake.

A third mistake is recalculating the bankroll after every bet and letting stakes swing wildly with short-term results. Updating the bankroll weekly or monthly, rather than after each result, smooths staking and keeps decisions calmer. Finally, remember that Kelly assumes you can reinvest indefinitely at the same edge. In practice, edges fade and accounts get limited, so treat Kelly as a guide for sizing today's bets, not as a promise about where your bankroll will be in a year. Conservative fractions, hard caps, and regular reviews turn a sharp formula into a survivable system.