Kelly Criterion: How Sharp Bettors Size Every Bet

Most bettors size their bets by feel. They bet more when they're confident, less when they're not, and adjust based on recent results. This is exactly backwards from how professional bettors approach bankroll management.

The mathematically correct approach is the Kelly Criterion — a formula developed by John Kelly at Bell Labs in 1956 that tells you exactly how much of your bankroll to bet given a known edge.

What the Kelly Criterion actually is

The Kelly formula is simple:

Kelly % = (bp - q) / b

b = decimal odds - 1  (net odds received)
p = probability of winning
q = probability of losing (1 - p)

In plain language: Kelly tells you the percentage of your bankroll to bet that maximizes long-run growth rate without risking ruin.

A concrete example. You have a player prop at -110 (decimal 1.909). Your model says the true probability of hitting is 57%.

b = 1.909 - 1 = 0.909
p = 0.57
q = 0.43

Kelly % = (0.909 × 0.57 - 0.43) / 0.909
Kelly % = (0.518 - 0.43) / 0.909
Kelly % = 0.088 / 0.909
Kelly % = 9.7%

Full Kelly says bet 9.7% of your bankroll on this play.

Why nobody bets full Kelly

9.7% of bankroll on a single bet sounds insane to most people — and for good reason. Full Kelly is mathematically optimal only if your probability estimate is exactly correct. In practice, every model has uncertainty baked in.

If your true edge is 57% but your model says 60%, full Kelly will over-bet relative to your actual edge. Over a long run, over-betting is catastrophic — it produces higher variance without proportionally higher returns, and it can wipe out a bankroll even on a profitable strategy.

The standard solution is fractional Kelly. Most professional sports bettors use quarter Kelly (25% of the Kelly-recommended bet) as a default. This dramatically reduces variance while preserving most of the long-run growth advantage.

Using the example above: full Kelly = 9.7%, quarter Kelly = 2.4% of bankroll.

A 2.4% bet size on a 10-game night means you're never risking more than 24% of your bankroll even if you bet everything that qualifies. In practice, with a filter for +6% EV and sharp confirmation, you'll rarely have 10 qualifying bets in a single night.

The practical unit system

Most bettors don't think in percentages — they think in units. The simplest way to apply Kelly thinking without doing the math on every bet is a tiered unit system mapped to EV ranges:

1 unit: +6% to +8% EV — solid edge, worth betting, keep size small

2 units: +8% to +12% EV — strong edge, slightly larger allocation

3 units: +12%+ EV — rare, maximum conviction, largest size

A "unit" should be 1-2% of your bankroll. So if you have a $500 bankroll, a unit is $5-10. This keeps individual bet sizes sane and variance manageable.

The tiered system approximates fractional Kelly without requiring a calculator for every pick. The higher the EV, the larger the edge, the larger the bet — which is exactly what Kelly prescribes.

What Kelly gets wrong

Kelly assumes you're making independent bets with no correlation. In practice, NBA player props on the same night can be correlated — if the game goes to overtime, multiple players hit their props. Kelly doesn't account for this.

Kelly also assumes your probability estimates are accurate. If your model is wrong — if what you think is 57% is actually 52% — Kelly will over-bet and destroy your bankroll faster than flat betting.

This is why fractional Kelly is standard: it's a hedge against model error and correlation risk. The more uncertain you are about your edge, the smaller the fraction of Kelly you should use.

How this applies to +EV betting

When you're using a model that generates explicit EV percentages on every pick, the Kelly framework maps directly. You know the model probability, you know the market odds, you can calculate Kelly exactly.

The line-move boost that sharp books apply when DraftKings confirms your pick functions as Bayesian updating — it's additional evidence that your probability estimate is correct, which means you can have slightly more confidence in the Kelly calculation.

At +6% EV with sharp confirmation, quarter Kelly on a $500 bankroll is roughly a $10-15 bet. At +10% EV, it's $18-22. The math is manageable.

The key thing Kelly teaches: don't bet the same amount on every pick. Size to the edge. That's the entire insight.

For more on how we calculate EV on every pick, read the model breakdown here. If you're comparing services before committing, see how we stack up against the alternatives.