This guide explains the core idea of the kelly criterion and how poker players can adapt it to real tables. You will learn the basic formula, a practical mean-over-variance approximation, when to scale back with fractional Kelly, and simple heuristics for cash games and tournaments.

Kelly maximizes long-run expected logarithmic wealth for repeated favorable bets when probabilities are known.

Fractional Kelly, like half-Kelly, reduces volatility while preserving a large share of theoretical growth.

Tournament play needs ICM-aware adjustments because chips do not map linearly to payout equity.

Quick answer: what is the kelly criterion in poker?

The kelly criterion is a formal rule that tells you what fraction of your bankroll to stake when you have a positive edge, because it maximizes long-run expected logarithmic wealth and therefore growth when you can repeat favorable bets.

For poker players the idea maps to sizing decisions when you believe you have a reliable win probability and know the distribution of returns, but real tables add complications like high variance, estimation error, and the nonlinearity of tournament chips compared with cash value. See the World Scientific overview for the theoretical framing and poker-specific discussion for practical adaptation The Kelly Capital Growth Investment Criterion.

Funded Plays Challenges page

See FundedPlays challenges

Short definition

Shortly put, Kelly prescribes the bankroll fraction that maximizes expected log growth when you repeatedly make favorable bets and compound winnings.

How poker differs from simple bets

Poker decisions differ from textbook binary bets because returns vary by pot size and street, edges are estimated from limited samples and opponent reads, and tournament chips do not equal cash in a linear way; these differences mean you often adapt or fractionally scale Kelly recommendations for practical play, especially in tournaments.

Where the Kelly Criterion comes from: theory and intuition

The Kelly idea has roots in information theory and the formal result that maximizing expected logarithmic growth yields an almost-sure long-run outperformance for favorable repeated bets; this link between growth-optimal sizing and information rate is foundational to the criterion Bell System Technical Journal paper. See the Wikipedia overview for a broader summary of the concept.

Intuitively, Kelly works because wealth compounds multiplicatively: staking a fixed fraction of your bankroll turns a series of edges into an exponential growth rate, and the logarithm converts multiplicative outcomes into additive increments that are straightforward to optimize over repeated plays.

Foundations in information theory and optimal growth

Formal proofs show that when outcomes repeat independently and you know the true probabilities, the Kelly fraction maximizes the expected log of wealth and dominates other fixed-fraction strategies in the long run, which makes it the canonical growth-optimal rule for favorable games The Kelly Capital Growth Investment Criterion.

Binary bet derivation at a high level

At a high level the binary-bet derivation balances the proportional gain on wins with the proportional loss on defeats to maximize the expected log multiplier per round; that tradeoff is where the simple closed-form formula for single bets comes from. See advanced derivations and extensions such as Kelly Criterion Extension: Advanced Gambling Strategy.

The basic Kelly formula and the binary bet case

The standard single-bet formula states that for a binary wager with decimal odds multiplier b and win probability p, the growth-optimal fraction is f* = (b p - (1 - p)) / b, which yields the percent of bankroll to risk when you repeatedly face that same bet under stationary conditions The Kelly Capital Growth Investment Criterion.

Define the terms clearly: b is the net odds received on a win expressed as a multiplier of the stake (for example, winning 2 to 1 gives b = 2), and p is your estimated probability of winning the bet; the numerator represents the expected net gain measured against the chance of loss.

Formula f* = (bp - (1 - p))/b explained

When bp equals (1 - p) the numerator is zero and Kelly suggests f* = 0, meaning no bet because expected log-growth is neutral or negative. If the numerator is negative, Kelly prescribes not to take the wager; only when bp exceeds (1 - p) is a positive fraction implied.

As a tiny numeric anchor, if you estimate p = 0.6 and the net multiplier b = 1 (even-money), the formula gives f* = (1*0.6 - 0.4)/1 = 0.2, so Kelly would stake 20% of bankroll on repeated identical wagers under idealized assumptions Kelly Criterion in practical settings.

Practice disciplined sizing in FundedPlays challenge programs

Keep reading for how to adapt this arithmetic to poker situations where bets are sequential and variance is large.

See FundedPlays challenges

A practical approximation: mean over variance and repeated play

For repeated plays with approximately normal returns the Kelly fraction admits a useful approximation: f* ≈ μ / σ², where μ is the mean net return per unit stake and σ² is the return variance; this compact form is often easier to compute when outcomes are frequent and roughly independent Optimal gambling systems for favorable games.

This μ/σ² expression emphasizes that higher variance reduces your recommended fraction for a given edge and that estimation error in either quantity changes the computed fraction significantly.

How should poker players use the Kelly Criterion in practice?

Use the Kelly Criterion to guide proportional bet sizing when you have reliable estimates of edge and variance, prefer fractional Kelly when estimates are noisy, avoid direct application in ICM-sensitive tournament spots, and always pair theory with conservative bankroll controls.

Use this approximation only when you have many comparable plays and a return distribution that is not heavily skewed, because single large pots or correlated outcomes violate the normal, independent-play assumptions behind the formula.

From discrete bets to μ over σ squared

Under the approximation, μ corresponds to the expected return per stake unit (for example expected big blind profit per big blind risked) and σ² is the variance of that return; mapping poker metrics into those terms is the key practical step to using this formula The Kelly Capital Growth Investment Criterion.

When the approximation is appropriate

The approximation is most appropriate for cash-game decisions with many repeated hands where no single decision dominates the variance and where you can treat each stake as an independent trial approximated by a normal distribution over reasonable horizons.

Fractional Kelly: why most practitioners use less than full Kelly

In real play you rarely trust your inputs perfectly, so practitioners commonly scale Kelly down to a fraction of f* to limit drawdowns and the risk from estimation error; this practice is known as fractional Kelly and is a standard risk-control technique Kelly Criterion in practical settings. See related work on modified approaches Modified Kelly Criteria.

Half-Kelly is a widely used default; theory and empirical work show that half-Kelly typically preserves roughly 75 percent of the full-Kelly log-growth while materially reducing volatility and extreme drawdowns, which is why it is popular among experienced players who face parameter uncertainty The Kelly Capital Growth Investment Criterion.

Half Kelly and growth versus volatility tradeoffs

Reducing to half-Kelly trades off some long-run growth for a meaningful decrease in variance of returns, so if your edge estimates are noisy or likely to change over time a fractional approach often produces better realized outcomes from a risk-adjusted perspective.

Practical rules of thumb

Practical guidance is to use higher fractions only when you have very large, stable datasets and well-tested edges; with small samples, nonstationary opponents, or tournament contexts, conservative fractions and caps make more sense than blind adherence to full-Kelly.

Applying Kelly to cash-game poker: mapping win rate and variance

$Minimalist full frame image of a stack of poker chips with a fraction highlighted in accent color to illustrate the kelly criterion showing ascending arrows for compounding growth and smaller arrows for reduced volatility$

In cash games you can translate your win rate and observed variance into the variables μ and σ² so that proportional sizing from Kelly emerges naturally, and that mapping helps convert theoretical fractions into stake sizes such as percent of stack or number of big blinds The Mathematics of Poker.

When you do the algebra the implied Kelly bankroll scale often ends up on the order of σ² / μ in betting units, meaning that for a given win-rate and variance there is a rule-of-thumb bankroll size at which Kelly-style fractions become practically comfortable.

Translating poker win rate to edge per bet

To map win rate into an edge per comparable stake unit, decide on a consistent unit such as one big blind or one effective stack percent, compute average profit per unit and its sample variance from tracked sessions or hand histories, and then apply the μ/σ² formula to estimate f* The Kelly Capital Growth Investment Criterion.

Sizing bets and implications for stack fractions

Once you have f* as a fraction of bankroll, translate that into table sizing by comparing your chosen stake to effective stack sizes and buy-in conventions; in practice many cash players prefer fixed-percentage rules or capped fractions rather than continuously varying stakes to keep sizing simple and predictable.

Why tournaments need a different treatment: chips are not cash

Tournament play breaks the linear assumptions underlying Kelly because chips do not convert to final payouts linearly; instead, independent chip model (ICM) calculations and payout structures define marginal equity, so a direct Kelly fraction applied to expected chip gain can produce misleading choices The Mathematics of Poker.

Because expected chip value differs from expected cash value, tournament sizing should prioritize changes in payout equity and position rather than raw chip expectation, and many players prefer strongly conservative fractions or alternate decision frameworks for shove-or-fold spots.

ICM and nonlinearity of chip utility

ICM converts chip distributions into payout probabilities and shows that gaining chips when you are already big may add less marginal payout equity than the same chip gain early on; this nonlinearity is the central reason Kelly-style arithmetic needs careful adaptation for tournaments.

Adjusting Kelly-style thinking for payout structures

Practical adaptations include using equity-based comparisons, applying much smaller fractions for tournament chips, or avoiding Kelly entirely for critical short-handed or endgame choices where ICM swings dominate raw expectation calculations.

Estimating your edge and variance at the table

Estimating μ and σ² starts with reliable data: hand histories, HUD statistics, tracked sessions, and consistent aggregation by stake and format; larger sample sizes are essential because small samples produce highly unstable estimates that make full-Kelly dangerous Kelly Criterion in practical settings.

Step one, gather comparable hands and normalize outcomes to a common unit such as big blinds; step two, compute the empirical mean profit per unit and the sample variance; step three, apply a fractional multiplier to the resulting f* to reflect estimation confidence and nonstationarity concerns.

Simple Kelly calculator for μ and σ² inputs

Use conservative confidence fraction when samples are small

Use rolling windows and out-of-sample checks where possible to detect shifts in opponent pool or strategy; if your edge drifts or variance increases materially, reduce your fractional Kelly or pause aggressive sizing until you rebuild confidence with more data.

Sources of data and sample size considerations

Good sources include hand history exports from your client, HUD-derived aggregates, and curated session logs; aim for large numbers of hands at a given stake to shrink estimation error because variance per hand can be substantial and misleading on short horizons The Kelly Capital Growth Investment Criterion. See the Funded Plays blog for related posts and examples.

Practical steps to estimate μ and σ²

A simple sequence is: pick your stake unit, compute profit per unit for each hand or session, calculate the sample mean and variance across comparable plays, then plug those numbers into μ/σ² and finally scale by a fractional factor that matches your confidence level.

Risk management: drawdowns, parameter risk, and overbetting

Overbetting relative to the true Kelly fraction destroys expected log-growth and can lead to long-run bankroll decline, so conservative caps and fractional scaling are essential when edge estimates are uncertain or non-stationary Optimal gambling systems for favorable games.

Fractional Kelly reduces the chance of extreme drawdowns and smooths variance, which makes it a practical risk-management tool; common operational controls include a maximum stake percentage, stop-loss rules per session, and periodic re-estimation of μ and σ².

How overbetting destroys log-growth

Betting more than the growth-optimal fraction increases volatility faster than expected growth, so even if short-run results look attractive, the expected logarithmic rate of return falls and eventual ruin risk rises as mis-sizing compounds across repeated plays.

Using fractional Kelly to limit drawdowns

Set conservative defaults like half-Kelly or smaller when sample sizes are limited, then increase toward full-Kelly only as confidence and evidence accumulate; combine fractional Kelly with clear stop-loss and bankroll rules to manage human error and variance spikes.

Bankroll sizing and the implied Kelly bankroll

The implied Kelly bankroll, derived from the inverse relation to μ/σ², gives a sense of how large a bankroll should be relative to your per-bet variance: the scale typically lands on the order of σ² / μ in betting units, which can be converted into buy-ins or big blind equivalents for practical planning The Kelly Capital Growth Investment Criterion.

That rule-of-thumb lets you see whether your current bankroll is large enough for Kelly-style play or whether you should be using much smaller fractions to avoid ruin from normal variance swings. See how Funded Plays evaluations work for related operational details.

Estimating a sensible bankroll for Kelly play

Compute σ² and μ on your chosen unit, invert the relation to estimate a comfortable bankroll scale, then translate that amount into numbers of buy-ins or big blinds using your usual stake and table rules; if the implied bankroll is impractically large, scale your fraction down.

Practical stacking and session rules

Practical measures include capping per-hand risk to a small percent of bankroll, setting session-level loss limits, and avoiding bet sizes that push more than the capped fraction of your effective stack into play on any single decision.

Common mistakes and practical pitfalls

One frequent error is misestimating edge from tiny samples; small-sample noise can make full-Kelly recommendations wildly aggressive and lead to avoidable drawdowns, so always check sample robustness before increasing fraction Kelly Criterion in practical settings.

Another mistake is treating tournament chips as if they were cash; applying cash-game Kelly arithmetic to tournament choices ignores ICM nonlinearity and can produce suboptimal or dangerous plays in pay-structure sensitive spots The Mathematics of Poker.

Misestimating edge

Always test edges with out-of-sample validation, conservative confidence intervals, and rolling windows; if your edge estimate is highly variable, prefer smaller fractions and slower position-sizing changes.

Ignoring nonstationarity

If the opponent landscape or game dynamics shift, your historical μ and σ² may no longer apply; include regular re-estimation and be ready to reduce sizing until a new stable baseline appears.

Three worked examples: cash hand, multi-street pot, and tournament shove

Example 1, small cash-game pot: suppose your consistent analysis suggests an expected win probability p = 0.55 on a particular raise line and the effective pot odds translate to net multiplier b = 1. If you treat the hand as a repeated, comparable decision, the binary formula gives f* = (b p - (1 - p)) / b = (1*0.55 - 0.45)=0.10, suggesting 10 percent of bankroll under ideal assumptions; in practice you would likely scale that down by a fractional factor to reflect sample size and opponent dynamics The Kelly Capital Growth Investment Criterion.

Takeaway: convert your poker read into p and b, calculate f*, then apply a conservative scale factor when uncertainty is present rather than betting the raw output of the formula.

Example 2, multi-street pot: sequential betting changes the effective variance and requires reassessment after each street; treat each decision as a distinct bet with updated p estimates and recompute f* or reduce to a capped fraction to avoid committing too large a share of bankroll when future streets could reverse the outcome The Mathematics of Poker.

Takeaway: update estimates as new information arrives and avoid locking in full exposure early in a hand without accounting for further variance from later streets.

Example 3, tournament shove: raw chip EV from a shove may be positive, but tournament payout equity under ICM can make the same shove incorrect; reduce any Kelly-based fraction or prefer an ICM-weighted decision framework in critical payout-sensitive spots to protect long-term tournament profitability.

A simple decision framework: when to apply Kelly in your play

Before you size a bet, run this quick checklist: are you in a cash-game context, do you have a stable large-sample estimate of μ and σ², is the bet repeatable under similar conditions, and is the computed f* within your pre-set cap? If the answer to any is no, prefer fractional sizing or alternative decision rules The Kelly Capital Growth Investment Criterion.

When to avoid Kelly: short sample horizons, high ICM sensitivity, correlated multi-street swings that violate independence assumptions, or severe uncertainty about opponent tendencies; in those cases, use conservative heuristics and focus on survival and information gathering.

Checklist before you size a bet

Checklist: sufficient sample size, stable estimated edge, cash versus tournament context, max fraction cap in place, and plan for periodic re-estimation and stop-loss limits.

When to prefer alternative approaches

Prefer alternate approaches like pot-odds-first heuristics, ICM calculators in tournaments, or fixed-percent staking when you lack confidence in your estimated μ and σ².

The kelly criterion is a mathematically grounded method that maximizes long-run expected logarithmic growth when you can repeatedly exploit a reliable edge, but it requires honest estimation of edge and variance and careful adaptation to real-world poker contexts The Kelly Capital Growth Investment Criterion. Visit Funded Plays for related tools and resources.

Minimalist 2D vector split illustration comparing cash game Kelly sizing and tournament ICM caution using chip stacks and simple icons on dark Funded Plays palette the kelly criterion

In practice most players deploy fractional Kelly, conservative caps, and tournament-specific adjustments to manage drawdowns and parameter risk; use the criterion as a disciplined sizing framework but never as a guarantee of profit, and combine it with rigorous data, conservative defaults, and responsible bankroll rules Optimal gambling systems for favorable games.

How is the Kelly fraction calculated for a simple bet?

For a binary wager the Kelly fraction is f* = (b p - (1 - p)) / b, where b is the net odds multiplier and p is your win probability estimate.

Should I use full Kelly at the poker table?

Usually not; because of estimation uncertainty and high variance most players use a fractional Kelly such as half-Kelly to reduce drawdowns and parameter risk.

Can I use Kelly for tournament decisions?

Apply Kelly cautiously in tournaments because chips have nonlinear payout value under ICM; prefer ICM-aware adjustments or more conservative fractions.

Treat the kelly criterion as a disciplined sizing and risk framework rather than a promise of profit. Combine it with conservative fractions, good data, and clear bankroll rules to improve the consistency of your poker sizing decisions.