Bankroll,
treat it like inventory.
Your bankroll is the only variable you control at the table. The dealer is fixed. The math is fixed. Bet sizing decides whether variance ends the session or just colours it.
The math problem bet sizing solves
Casino games are negative-EV, but the rate at which you lose is dominated by variance for any session shorter than tens of thousands of spins. Variance is what bet sizing controls. Bet too big and the inevitable losing streak clears your bankroll before EV has a chance to show up at all. Bet small enough and you stay in the game long enough for an upside spin to land.
You don’t bet to win. You bet to stay alive long enough to see variance.
Unit size, anchored to session bankroll
A unit is the standard bet for the session. Pick a percent of bankroll, not a dollar value. If your bankroll changes next session, the unit changes with it. The percent depends on game volatility.
| Game type | Unit size (% of session BR) | Why |
|---|---|---|
| Blackjack, baccarat, single-zero roulette | 1–2% | Low variance plus low house edge. You can size bigger without runaway risk. |
| Low-volatility slots | 0.5–1% | Hit rate is high enough that 200–500 spin sessions are realistic at this size. |
| High-volatility slots | 0.25–0.5% | Standard deviation can swallow 50+ unit losses in a stretch. Smaller units buy survival. |
| Bonus buys | 0.5–2% per buy | You are paying for compressed variance. Treat each buy as one outcome, not many. |
| Crash, dice (custom-multiplier) | 0.25–1% | Multipliers above 5× compound losses fast at high stakes; size small or your max-loss hit rate spikes. |
Worked example bankroll = $1,000 game = high-volatility slot (Hand of Anubis, NoLimit territory) unit % = 0.4% unit size = $1,000 × 0.4% = $4 per spin spin budget = $1,000 / $4 = 250 spins before any wins That 250-spin floor is the survival number. Below it, you do not have enough rope for the slot's hit-rate distribution to land a feature. At $20 per spin, you would have 50 spins. Math says no.
Risk of ruin, in pictures
Risk of Ruin (RoR) is the probability that bankroll hits zero before reaching some target. At negative-EV games, RoR is unavoidable in the long run. The question is how fast it shows up. The smaller your unit relative to bankroll, the longer it takes.
| Bet as % of bankroll | Bust risk in 100 spins | Bust risk in 1,000 spins |
|---|---|---|
| 0.25% | ~1% | ~10% |
| 0.5% | ~5% | ~28% |
| 1.0% | ~12% | ~52% |
| 2.0% | ~24% | ~76% |
| 5.0% | ~48% | ~95% |
| 10% | ~74% | ~99.5% |
Approximation for high-volatility slots at 96% RTP, with ~22% hit rate. Real slots will vary; the shape of the curve does not.
Reading the table the right way: at 5% per spin, a 1,000- spin session has a 95% chance of clearing you out before it ends. The promise of "playing for an hour" with 5% bets is fiction at high volatility. You either run hot immediately or you bust.
Kelly fraction, the +EV case
The Kelly criterion gives the optimal bet size when you have an actual edge. Casino games are negative-EV, so Kelly says zero. But two scenarios in crypto-casino play are genuinely +EV: cashable bonus clearing on the right slot, and rakeback / cashback structures.
Kelly formula (binary outcome) edge = p_win × payout − p_loss variance = p_win × payout² + p_loss Kelly% = edge / variance Half-Kelly is the practical bet size. Full Kelly is statistically optimal but the drawdown ride is brutal, and "I miscalculated my edge by 1%" sends you bust. Almost everyone uses 1/4 to 1/2 Kelly. Example: a +2% rakeback structure on slots edge per $1 wagered = +$0.02 variance per $1 ≈ 5 (high-volatility slot) full Kelly% = 0.02 / 5 = 0.4% of bankroll per spin half Kelly = 0.2% of bankroll per spin Match this back to the unit-size table: the half-Kelly answer is 0.2%, right inside the high-volatility slot recommendation.
Kelly is rarely useful on its own at a casino. It is useful as a sanity check on unit-size guesses. If a bonus structure pushes your half-Kelly above 1% of bankroll, the offer is exceptionally generous (or you miscalculated the edge).
Stop rules that actually work
Hard. Pre-committed before the session starts. If the number flashes, you stop. Tilt is the failure mode the rule exists for; the rule does not negotiate when tilt arrives.
At 2× starting bankroll, withdraw half. The remaining half is house money. This converts a streak into a realised result instead of a story about a streak that evaporated.
The recurring losing pattern in casino streamers is not a single bad session. It is a sequence: small loss, raise stakes to chase, big loss, raise stakes harder, ruin. Pre-committed stops break the chain at step one. They feel arbitrary in the moment because they are arbitrary, that is the point. You do not negotiate with arithmetic.
The three rules in one breath
Responsible gambling resources, AU: gamblinghelponline.org.au. UK: gamcare.org.uk. US: ncpgambling.org. If the number on the screen has stopped feeling like math, walk.