The number Kelly needs is not on the felt
Walk up to a blackjack table and the layout will show you payout odds: a winning hand pays even money, a blackjack usually pays 3 to 2, insurance pays 2 to 1. None of those numbers is what the Kelly criterion asks for. Kelly wants the edge, the difference between the true probability of winning a hand and the probability implied by the payout. A payout of even money only tells you the reward side. The edge is the reward side minus the risk side, and the risk side is the part you have to estimate yourself.
That estimate is the whole difficulty. On a slot machine or a roulette wheel, the probabilities are fixed and published, so the house edge is a known constant. At blackjack, the probability of winning the next hand shifts with every card removed from the shoe. The odds you plug into Kelly are not a printed number; they are a running calculation of your advantage on that specific hand under that specific rule set, against that specific remaining deck composition.
How the Kelly formula works here
The standard Kelly fraction for a simple win-or-lose bet is f = p - q/b, where p is your probability of winning, q is your probability of losing, and b is the net odds you receive on a winning bet, expressed as a multiple of your stake. For an even-money blackjack hand, b equals 1, so the formula collapses to f = p - q. Since q is 1 - p, that becomes f = 2p - 1. A hand you win 50% of the time produces a Kelly fraction of zero: no bet. A hand you win slightly more than half the time produces a Kelly fraction of zero.02, meaning two hundredths of your bankroll on that hand.
That last sentence contains the whole discipline of the method. The odds you use are not the payout, and they are not a guess about whether you feel lucky. They are a probability estimate, and the estimate has to be better than the break-even point before Kelly tells you to bet anything at all. If your estimate of p is wrong, Kelly will happily size a bet on a hand you are losing.
Where the probability estimate comes from
In a game with a fixed deck composition, the probability of winning a hand is determined by the rules and the strategy in use. Published basic-strategy charts derive from a specific rule set: how many decks are in play, whether the dealer stands or hits on soft 17, whether doubling after splitting is allowed, whether surrender is offered. Change any of those rules and the probability of winning a given hand shifts, which shifts the Kelly fraction. The same player at two different tables can face two different correct bet sizes on identical-looking hands.
A player who counts cards is trying to track when the remaining shoe has shifted in their favour. The count is a proxy for the composition of the remaining cards, and the composition is what moves the win probability. When the remaining deck is rich in high cards, the player's probability of winning rises, and the Kelly fraction rises with it. The odds plugged into the formula are therefore not the count itself, but the win probability implied by the count under the specific rules of the table.
No published chart can give you that probability for every count at every table. The player must derive it, typically from simulation data for the exact rule set in play, and then translate the count into an edge. That translation is where most of the practical error enters.
What the house edge does to the calculation
House edge and RTP are complements: they sum to 100% for the same game under the same rule set. For a blackjack hand, the house edge is the operator's mathematical advantage on that wager, expressed as a percentage of the stake. A player who plays basic strategy against a fixed shoe faces a small negative expectation on every hand. The Kelly formula, run honestly on that negative expectation, produces a negative fraction. The correct Kelly bet is zero.
This is the part of the method that most write-ups skip. Kelly is not a way to manufacture an edge. It is a way to size a bet once an edge exists. If the probability estimate says the house still holds an advantage, the formula does not produce a small optimistic number. It produces a number below zero, and the only correct action is not to bet. The method cannot turn a negative-expectation game into a winning one; it can only tell you how much of your bankroll to risk when the expectation is genuinely positive.
For a card counter, the expectation crosses from negative to positive only when the composition of the remaining shoe has shifted enough to overcome the built-in house edge. The Kelly fraction is then the size of the advantage on that hand, scaled by the variance of the outcome. A tiny positive edge produces a tiny bet. A larger edge produces a larger bet, but never a bet that risks the bankroll on a single hand, because Kelly's whole purpose is to avoid the ruin that comes from overbetting a real but modest advantage.
Why a single hand is not the long run
RTP and house edge are long-run statistical averages, not per-session guarantees. The same is true of a card counter's advantage. A hand with a positive expectation of half a percent does not pay half a percent; it pays the full payout or nothing, with the probabilities tilted slightly in the player's favour. Over ten hands, the actual result can be anywhere. Over ten thousand hands, the result clusters around the expectation, but only if the player's probability estimates were accurate in the first place.
That gap between the single hand and the long run is why Kelly produces fractions rather than flat bets. The formula is explicitly designed around variance. It asks how much of the bankroll can be risked on a positive-expectation bet without making the eventual ruin of the bankroll likely, even through a long run of losing hands. A player who ignores the variance and bets a fixed large amount on every positive count is not using Kelly; they are using a progression with extra steps.
What changes under the player's control
Several genuine decisions move the numbers that feed into Kelly. None of them change the house edge on a given bet, but they change which bets the player faces and how many of them occur per hour.
| Decision | What it changes | What it does not change |
|---|---|---|
| Table selection | The rule set in force: deck count, dealer stand rules, doubling and splitting options, surrender availability | The house edge on any given bet under a fixed rule set |
| Strategy accuracy | The probability of winning each hand, and therefore the Kelly fraction | The payout odds printed on the layout |
| Bet sizing method | How much of the bankroll is exposed per hand, and how ruin risk scales | The expectation of the hand itself |
| Session length and hands per hour | Total amount wagered, and therefore total expected loss or gain over the session | The edge per hand |
| Stopping rules | When a losing streak ends and the bankroll is preserved | The variance of any individual hand |
Hands per hour deserves more attention than it usually gets. The edge is expressed per wager, but the total exposure is the edge multiplied by the number of wagers. A player who sits at a fast table and plays twice as many hands per hour has doubled their total expected loss, even though the house edge on each hand is unchanged. Kelly sizing controls the size of each bet; it does not control how many bets are made. A Kelly-sized bet played too often still grinds the bankroll down if the edge estimate is wrong or the variance runs long.
Why progression systems are not Kelly
Doubling after a loss, raising after a win, and similar patterns are sometimes presented as bankroll methods in the same breath as Kelly. They are not the same thing. A progression system changes the bet size according to the previous outcome, not according to the current probability of winning. The previous outcome carries no information about the next hand in a shuffled shoe, so the progression is sizing bets on noise.
What a progression does is redistribute the timing of wins and losses. It concentrates the bankroll into a small number of large bets and spreads the rest across many small ones. The total expectation across all bets is unchanged: the house edge applies to every wager at its own rate, and no ordering of bet sizes alters the sum. A progression can produce many small wins and an occasional catastrophic loss, or the reverse, but it cannot change the expected value of the sequence. Kelly, by contrast, sizes each bet from an estimate of the edge on that bet. The two methods look similar only because both produce variable bet sizes. The reason for the variation is entirely different.
No betting pattern changes the built-in edge of a negative-expectation game. No progression system removes the house advantage. The only thing that moves the edge in the player's favour at blackjack is a genuine change in the probability of winning, and the only honest use of Kelly is to size the bet once that change has occurred and been measured.
FAQ
Can I use Kelly without counting cards?
You can run the formula on any probability estimate, but without counting, the estimate is the basic-strategy win probability for a fixed shoe, which is below break-even. The Kelly fraction comes out negative, and the correct bet is zero. The method does not create an edge; it only sizes one that already exists.
What happens if I use half-Kelly instead of full Kelly?
Half-Kelly means betting half the fraction the formula produces. It reduces the growth rate of the bankroll but also reduces the variance and the chance of a deep drawdown. Some players prefer it because their edge estimates are uncertain, and a smaller bet leaves more room for estimation error before the bankroll is damaged.
Does the payout for a blackjack change the Kelly fraction?
Yes, because a blackjack pays more than even money, the formula's b term is not 1 on those hands. The probability of being dealt a blackjack and the payout for it both enter the overall expectation of the game, and a rule set that pays less for a blackjack reduces the player's edge and therefore the Kelly fraction on every hand.
Play responsibly
Gambling should be treated as paid entertainment, never as a way to earn income or recover losses. 1-800-MY-RESET (1-800-697-3738)
18+ or 21+ depending on where you are; follow the minimum age that applies to you.
This article is general information about how these mechanics work. It is not legal advice and not a recommendation to gamble or to use any particular operator. Availability and legality differ by jurisdiction — check the rules that apply where you are.