The formula
With decimal odds o, your probability of winning p and b = o - 1:
f* = (b × p - (1 - p)) / b = (p × o - 1) / (o - 1)
f* is the share of the bankroll to stake. The top of the fraction, p × o - 1, is your edge: what you expect back per unit staked, minus the unit. No edge, no stake.
Where the probability comes from
Kelly needs a probability that is better than the odds' own. Some bettors use a model; many use a sharp book's price with the margin taken out (the no-vig calculator does that) and compare it with a softer book's odds. Either way, the answer is only as good as the estimate.
Why most people use a fraction
Full Kelly maximises long-run growth if your edge is real and exactly known. It also produces large swings: a run of losses can take a big share of the bankroll. Overestimate your probability by a few points and full Kelly overstakes. Half Kelly keeps about three quarters of the growth with far smaller swings, which is why fractional Kelly is the common choice.
Whatever the formula says, only stake money you can afford to lose. Results vary and no stake size removes the risk of a bet.

