Flat 1-unit betting feels disciplined. In practice, it leaves long-run growth on the table by treating a 60% edge bet the same as a 52% edge bet. Kelly fixes that — but full Kelly is so aggressive it will blow up most real bankrolls. The middle path, fractional Kelly, is where the actual edge lives.
If you've spent any time around sports betting strategy content, you've seen two opposed religions. The first is flat-unit betting: 1u on every play, regardless of confidence. The second is "max bet what feels right" — which is just gambling with extra steps. Both ignore the math that determines how a bankroll actually grows over time. That math is the Kelly Criterion, and once you understand it, the “1u everything” rule starts to look like the safety blanket it is.
The Kelly Criterion is a formula that tells you, given a known edge over the market, what fraction of your bankroll to bet to maximize the long-run geometric growth rate. It was developed by John Kelly Jr. at Bell Labs in 1956, originally for information theory, and was quickly adopted by sharp gamblers because it answered a question nobody else had answered cleanly: given that I have an edge, how much should I bet?
The formula for a binary bet (win or lose) with American odds looks like this:
The output, f*, is a fraction between 0 and 1. If f* is 0.04, Kelly says to bet 4% of your bankroll. If f* is negative, Kelly says don't bet at all — the math is telling you the line is bad. The intuition is that the bet size should scale with two things: how big your edge is over the market, and how much the odds pay if you win. Both inputs matter. A 10% edge on a +200 underdog gets a larger bet than a 10% edge on a -200 favorite, because the payoff math is different.
Say you're betting an NHL moneyline at +130. Your model says the true win probability is 50%. The market's implied probability at +130 is about 43.5%. You have a 6.5-point edge.
Plug into Kelly:
If your bankroll is $10,000, full Kelly tells you to put $1,150 on this game. That's a big bet. It's mathematically optimal — but only if your probability estimate is exactly right, your bankroll is uncorrelated with anything else, and you have infinite tolerance for short-term drawdown. None of those things is true for a real bettor. Which is why nobody serious bets full Kelly.
The Kelly formula has one assumption that breaks the moment you try to apply it: that your probability estimate is exact. If you say a team has a 50% win probability and they actually have a 48% win probability, Kelly is telling you to massively overbet. The formula is hypersensitive to errors in your input. A 2-point miss in your probability estimate can mean the difference between “bet 12% of bankroll” and “bet zero / don't bet at all.”
In practice, every sports betting model has noisy probability estimates. Even the best public models are calibrated within a few points, not a fraction of a point. That noise propagates straight into Kelly's output. The result is that betting full Kelly on noisy inputs creates massive bet sizes during what feel like obvious edges — and then the inevitable bad-luck stretch wipes you out faster than a flat bettor with the same skill would lose.
The math behind this is brutal. Full Kelly is the maximum-growth path in theory. It also produces drawdowns that no human bettor will sit through. The expected worst drawdown under full Kelly is around 50% of bankroll — and that's just the expected case. Real drawdowns regularly exceed that. Most bettors who try full Kelly either tilt and overbet further to recover, or quit during the drawdown and miss the recovery. Either way the math wins; the bettor loses.
Theory is one thing. The actual drawdown distribution under each Kelly fraction is what matters at 3am after a 7-bet losing streak. Across thousands of Monte Carlo simulations of a bettor with a real 4% edge across 500 bets, the comparison looks like this:
Full Kelly: ~120% expected long-run growth · max drawdown frequently 40–55% · ~5% of simulations bust the bankroll outright
Half Kelly: ~95% expected growth · max drawdown typically 20–28% · bust rate near zero
Quarter Kelly: ~58% expected growth · max drawdown 11–15% · drawdowns rarely cause emotional capitulation
Flat 1u (1% of bankroll): ~35–45% expected growth (depends on edge dispersion) · max drawdown 8–12% · simplest to execute
Two things to notice. First, going from full Kelly to half Kelly costs you about 20% of expected growth (95 vs 120) but cuts max drawdown by more than half. Going from half Kelly to quarter Kelly costs another 40% of growth but the drawdown curve barely improves. The sweet spot for almost everyone is between quarter and half Kelly — meaningful upside compared to flat sizing, drawdowns small enough to survive psychologically.
Second, the bust rate on full Kelly is real. A bettor with a real edge running pure Kelly has a measurable chance of zeroing their bankroll inside 500 bets — not because the math was wrong, but because the variance distribution under full Kelly has a long left tail that includes ruin. The math says "long-run optimal growth"; the math also says "long-run" means thousands of bets, and you have to survive the first 500. Fractional Kelly is the mechanism that makes survival overwhelmingly likely.
The Kelly formula assumes your probability estimate is exactly correct. In real-world sports betting, every estimate has noise. A model that says "Brewers 48.3%" might mean the true probability is anywhere from 45% to 52% — the model is unbiased but uncertain. Full Kelly handles unbiased noise badly: even when the average estimate is right, individual overestimates produce bets too large, and the overweighted bets dominate the drawdown distribution.
The practical rule: scale your Kelly fraction down by how noisy you think your estimates are. A bettor with a sharp model on a sport they've studied for years and consistent positive CLV across 1,000+ bets can credibly run 1/2 Kelly. A bettor relying on a public service's picks or a model with thin sample backing should run 1/4 Kelly or less. The harder it is to defend the precision of your win-probability estimate, the more aggressively you should fractionalize.
One useful heuristic: if you're not measuring CLV on your own bets, you can't credibly claim your estimates are sharp enough for half Kelly. CLV is the evidence that your estimates beat the market. Without that evidence, quarter Kelly is the right default — not because you're definitely worse, but because you don't yet know.
Fractional Kelly means betting some fraction — usually 1/4 or 1/2 — of what full Kelly says. If Kelly says bet 11.5%, half Kelly says bet 5.75%, quarter Kelly says bet 2.9%. The bet sizes shrink, but the relative ordering across bets stays exactly the same. Bigger edge = bigger bet, smaller edge = smaller bet, same as full Kelly. You just dial down the absolute volume.
The reason this works is that growth rate falls off slowly as you reduce the Kelly fraction, but drawdown risk falls off fast. Half Kelly produces roughly 75% of the long-run growth of full Kelly while cutting expected maximum drawdown from ~50% to ~20-25%. Quarter Kelly produces roughly 44% of full Kelly's growth but drops drawdown risk to under 15%. You give up some upside; you make the math survivable.
For a bettor whose probability estimates have noise — which is every bettor who isn't a quant fund — fractional Kelly also self-corrects for the noise. If your probability estimates are off by a few points on average, full Kelly overbets and gets murdered. Half Kelly's overbet is half as bad and the drawdown stays inside what a real bankroll can absorb.
Flat 1u betting has one virtue: it's simple. You decide a unit (typically 1% of bankroll) and bet 1u on every play. No math, no math errors, no tilt math. The catch is that flat-unit betting treats every bet the same, regardless of how much edge you actually have.
That's leaving long-run growth on the table. A bet with a 10% edge should be sized larger than a bet with a 3% edge, because the long-run profit math says so. Kelly captures that. Flat 1u betting does not. Over a year of betting, the gap between fractional Kelly and flat 1u — at the same skill level — typically comes out 30 to 60% more bankroll growth for fractional Kelly. That's not noise. That's a structural advantage.
The trade-off is variance. Fractional Kelly produces more variance than flat 1u sizing because bet sizes change game to game. A bettor who can't emotionally tolerate that variance should stick with flat 1u — discipline beats theoretical optimality every time. But a bettor who can size bets to their edge will, over thousands of bets at the same skill, end up with measurably more money.
Our official plays come pre-sized. The model's tier-sizing — 1u, 1.5u, 2u, 3u depending on edge strength — is essentially a discrete fractional Kelly approximation. We don't ask members to do the math. The model computes the edge against the closing line, maps that to a Kelly fraction, applies a fractional Kelly multiplier (typically 0.25 to 0.5 depending on sport and signal stability), and emits the unit size directly on the play card.
What that means in practice for a member:
For a member, “1u” should be 1% of their personal bankroll. That's not a rule, that's just the conservative default. A more aggressive bettor with high confidence in our long-run track record could use a larger unit; a more conservative bettor could use a smaller one. The relative sizing across plays — 1u vs. 2u vs. 3u — is what carries the Kelly math, and that ordering stays the same regardless of what dollar value the bettor assigns to a unit.
If you're starting out, flat 1u betting at 1% of bankroll is fine. It's simple, it's safe, it's hard to screw up. Once you have a tracked record of 200+ bets and you trust your model — or your service's model — graduate to fractional Kelly. The bankroll growth difference compounds.
A few practical rules:
Discipline always beats optimization. But if you have the discipline to follow a sizing rule and you understand where your edge actually comes from, fractional Kelly is the sizing rule that grows real bankrolls fastest with survivable variance.
The Kelly Criterion is a bet-sizing formula that calculates the fraction of bankroll to wager given a known edge and known odds, with the goal of maximizing long-run bankroll growth. The basic formula: f* = (b×p − q) / b, where p is your win probability, q is 1−p, and b is the decimal odds minus 1. Plug your numbers into the Kelly Calculator for the math.
Full Kelly bets exactly the fraction the formula returns. Fractional Kelly bets some smaller multiple — typically 1/4 or 1/2 of that fraction. Full Kelly is mathematically optimal but only if your probability estimate is exactly right; any noise produces overbets and large drawdowns. Fractional Kelly trades a small amount of long-run growth for dramatic drawdown reduction, which is why disciplined bettors use it almost universally.
Start with quarter Kelly (1/4) if you're using picks from a service or your model is unproven. Move to half Kelly (1/2) only after you've tracked 500+ bets with positive CLV — that's the evidence your probability estimates are sharper than the market. Most professional sports bettors operate between 1/4 and 1/2 Kelly indefinitely.
Yes — in two ways. If your probability estimates are biased (consistently wrong in a specific direction), Kelly will systematically overbet against your real edge, and you'll lose. If your estimates are unbiased but you use full Kelly with noisy estimates, you'll occasionally bust from variance even with real long-run edge. Fractional Kelly + accurate, CLV-validated estimates are the combination that produces actual growth.
Flat unit betting risks the same amount on every bet regardless of edge size. Kelly sizes by edge: bigger edges get bigger bets. Over a year of betting at the same skill level, fractional Kelly typically produces 30–60% more bankroll growth than flat 1u sizing because it correctly weights high-confidence opportunities. The cost is more variance — bet sizes change game to game.
Yes — the 1u/1.5u/2u/3u tier system on every play is a discrete fractional-Kelly approximation. The model computes the edge against the closing line, maps it to a Kelly fraction, applies a 0.25–0.5 fractional multiplier (depending on sport and signal stability), and emits the unit size directly. Members never have to do the math themselves — the recommended size is already Kelly-sized.
For informational and entertainment purposes only. Past performance does not guarantee future results. Sports betting involves risk — never bet more than you can afford to lose. Please gamble responsibly.