Kelly Says: Don't Bet
The Kelly Criterion is a formula developed by John Kelly at Bell Labs in 1956 for sizing bets to maximise the geometric growth rate of a bankroll. It's used by professional gamblers, quantitative traders, and anyone who takes bankroll management seriously.
The formula for a binary even-money bet is: optimal fraction equals the probability of winning minus the probability of losing, divided by the payout ratio. On European roulette red/black, the probability of winning is 18/37 (approximately 0.4865). The probability of losing is 19/37 (approximately 0.5135). Payout ratio is 1. The formula gives: 0.4865 minus 0.5135 equals negative 0.0270.
Negative 2.70%. As Wikipedia's Kelly Criterion article explains, when the edge is negative, the formula gives a negative result, indicating that the gambler should take the other side of the bet. Kelly is telling you to be the casino, not to bet at all.
Some sophisticated players discover Kelly and assume that any positive fraction of it constitutes a conservative, mathematically grounded approach to roulette. Fractional Kelly of a negative number is still negative. Multiplying negative 2.7% by any positive fraction gets you a bet that costs money. That's fine. Recreation is a valid reason to depart from Kelly's recommendation. But you should depart from it knowingly, not in ignorance of what the formula actually says.
The one case where Kelly gives a different answer is French roulette with La Partage on even-money bets. The effective probabilities become 18.5/37 on both sides, because zero returns half your stake. Kelly gives exactly zero: bet nothing, break even in expectation. This is mathematically correct. La Partage halves the house edge but doesn't eliminate it. Kelly knows this.
Risk of Ruin: Two Different Questions
The relevant concept for finite-session play is risk of ruin, specifically path-dependent risk of ruin. This is different from the naive end-of-session calculation.
End-of-session loss probability asks: are you down more than B at the end of N spins? Path-dependent risk of ruin asks: did your bankroll touch zero at any point during those N spins? For continuous play, path-dependent roughly doubles the simple end-session probability. This distinction matters because a player who calculates their session odds based on end-of-session results is underestimating their actual ruin risk by approximately half.
For European single-zero roulette, red and black, £500 per spin, a four-hour session at 60 spins per hour is 240 spins. According to calculations based on the reflection principle of Brownian motion, as confirmed by GamblingCalc's risk-of-ruin methodology, a path-dependent ruin probability below 0.5% over that session requires approximately £25,000 as session bankroll. That's the 50-times-stake rule of thumb: 50 times your stake as session bankroll gives you roughly half a percent ruin probability on even-money bets over a four-hour session.
Drop to 20 times stake, £10,000, and the path-dependent ruin probability climbs to approximately 38%. That is not a conservative bankroll. That is accepting a meaningful chance of going broke before the session ends.
Inside Bets: A Different Risk Profile
For straight-up bets at the same £500 per spin, the picture changes substantially.
The standard deviation per spin on a straight-up bet is approximately 5.84 times your stake, versus approximately 0.999 times for even-money bets. You need approximately six times the bankroll to achieve the same ruin probability as an even-money player. At £500 per spin on straight-up bets, a £25,000 bankroll gives you approximately 63% path-dependent ruin probability over a four-hour session. You need around £140,000 to get that number below 3%.
This is why serious players who prefer straight-up bets bring very large bankrolls and aren't being irrational when they do. A 35:1 win at £500 returns £17,500: a meaningful number. But the volatility is six times higher, and without the bankroll to match, you're running a real risk of a session exit before you've had time to find out whether the evening was going to be worth it.
The Short-Session Illusion
Here is the trap most recreational high-rollers need to hear most clearly.
Over a short session, the variance is large relative to the expected loss, so positive outcomes are entirely plausible. The expected loss for 30 spins at £500 per spin on even-money bets is only £405. The standard deviation is approximately £2,739. The player can easily finish £5,000 up without that being particularly improbable. This creates the illusion that the game is "running well," or that a system is working.
Over long runs, the expected loss accumulates linearly, proportional to the number of spins, while variance grows only as the square root of spins. The signal-to-noise ratio of the house edge improves continuously. Short sessions can show positive results on even-money bets roughly 37% of the time. This is not luck, or not just luck: it's variance doing what variance does. Players who book those wins as proof of discipline and conclude their bankroll requirements are therefore lower have made a survivorship error: they are alive to have this conversation precisely because variance went their way.
Over a three-session weekend at £500 per spin, 240 spins each: total action £360,000, expected loss approximately £9,730, combined standard deviation approximately £13,419. The combined 95% confidence interval runs from approximately -£35,967 to +£16,508. A player who refuses to accept the possibility of a -£35,000 weekend has not understood their own bankroll requirements.
What the Pit Boss Is Watching
The casino isn't primarily watching for cheating at a roulette table. It's watching for tilt, credit-extension risk, and problem-gambling indicators.
Chip stack tells: a player who started with organised, colour-coded chips and by mid-session has a disordered pile of mixed-value chips is showing a near-universal sign of depleting bankroll and deteriorating discipline. A bet size that increases by a factor of three or more from the opening bet is often chasing, not ascending confidence.
A player switching from outside bets to inside bets mid-session is sometimes rational, wanting the upside to recover. To an experienced pit boss it reads as tilt, because composed players don't change strategy in response to losses. The table history, visible on a tablet at larger venues, makes the escalation pattern obvious.
At UK casino level, the response to these signals may be a drink, a light comp, or a discreet conversation. VIP hosts get a call. The UKGC's AML and social responsibility guidance requires operators to take welfare seriously; the pit boss's intervention is not personal but regulatory.
Tipping: What Actually Happens
Under UKGC LCCP Condition 10.1.1, every licensed UK casino must operate a tronc system for tips. When you slide a chip to the croupier, it goes into a pool shared across all licensed staff on that shift. The gesture is genuine, and the staff appreciate it. But the idea that you're individually rewarding the dealer who was attentive to you at a difficult moment is a social fiction. The chip goes into the communal pot. This is UK law, not house policy.
At colour-up, £5 to £25 is a normal range for a £500-unit session. More if the session has gone well and the table has been attentive. As discussion on Reddit's casino tipping threads confirms, UK casino tipping is low-expectation relative to the American norm, and no one expects a percentage calculation.
Key numbers
| Stake per spin | Session bankroll for path-RoR below 0.5% (4 hours, even-money) | Same for straight-up bets |
|---|---|---|
| £100 | ~£5,000 | ~£30,000 |
| £250 | ~£12,500 | ~£75,000 |
| £500 | ~£25,000 | ~£140,000 |
| £1,000 | ~£50,000 | ~£280,000 |
| B/stake ratio | Bankroll at £500/spin | Path-dependent RoR (4 hours, even-money) |
|---|---|---|
| 20x | £10,000 | ~38% |
| 30x | £15,000 | ~13% |
| 40x | £20,000 | ~3% |
| 50x | £25,000 | ~0.5% |
| 4-hour session at £500/spin even-money (European) | |
|---|---|
| Total action | £120,000 |
| Expected loss | ~£3,243 |
| Session standard deviation | ~£7,743 |
| 95% confidence interval | -£18,420 to +£11,933 |
Sources: Kelly criterion Wikipedia; UKGC LCCP 10.1.1; Smithsonian on Crockford founding