Most sports bettors spend all their time picking winners. They argue about lines, research matchups, and look for edges. Then, when they find a bet they like, they wager a random amount — $50, $100, whatever feels right.

This is backwards. Bet sizing is at least as important as pick selection. You can have a genuine edge on every bet you place and still go broke if you size your bets wrong. Conversely, disciplined bet sizing turns a small edge into compounding growth over time.

The Kelly Criterion is the mathematical formula that solves this problem. It tells you exactly what percentage of your bankroll to bet on any given wager to maximize long-term growth. It was developed by physicist John Kelly Jr. in 1956 and has been used by professional gamblers, poker players, and even Wall Street traders ever since.

Here's everything you need to know to use it correctly in sports betting.

What Is the Kelly Criterion?

The Kelly Criterion is a formula that calculates the optimal bet size as a fraction of your total bankroll. It maximizes the long-run growth rate of your capital when you have a genuine edge.

The formula is:

Kelly Criterion Formula

f = (bp - q) / b

f = fraction of bankroll to bet
b = decimal odds minus 1 (profit per $1 wagered)
p = your estimated probability of winning
q = probability of losing (1 - p)

Let's work through a concrete example.

Say you're betting on an NBA game. After doing your sports betting research, you estimate the home team has a 55% chance of winning. The sportsbook is offering -110 (bet $110 to win $100, or equivalently, bet $1 to profit $0.909).

  • p = 0.55 (your estimated win probability)
  • q = 0.45 (loss probability)
  • b = 100/110 = 0.909 (profit per $1 at -110 odds)

Kelly fraction = (0.909 × 0.55 - 0.45) / 0.909 = (0.500 - 0.45) / 0.909 = 0.050 / 0.909 = 5.5%

With a $1,000 bankroll, Kelly says to bet $55 on this game.

Why Kelly Maximizes Long-Term Growth

The Kelly formula is derived from information theory. It finds the bet size that maximizes the expected logarithm of wealth, which is equivalent to maximizing the long-run compound growth rate of your bankroll.

Bet less than Kelly and you're leaving growth on the table. Bet more than Kelly and your bankroll growth slows — and at some point reverses entirely. At twice Kelly (overbetting by 2x), expected growth rate is actually zero. At more than twice Kelly, you are expected to go broke in the long run, even if every individual bet has positive expected value.

The counterintuitive insight: Betting more than Kelly makes you worse off even when you're winning, because the losses when they come are proportionally larger. Overbetting is one of the most common ways bettors with genuine edges still go broke.

The Problem: Estimation Error

Here's the critical issue with full Kelly in practice: the formula is only as good as your probability estimate.

If you estimate 55% probability but the true probability is 52%, full Kelly says to bet 5.5% of your bankroll. But the correct Kelly for 52% probability is much smaller. You are overbetting relative to your actual edge, which means you're taking on more variance than the math justifies.

And here's the uncomfortable truth: most bettors overestimate their edge. It's human nature. You research a game, develop conviction, and that conviction inflates your probability estimate. Even sophisticated modelers have estimation error.

This is why virtually all professional sports bettors use a fraction of Kelly rather than full Kelly.

Fractional Kelly: The Practical Solution

Fractional Kelly means multiplying the Kelly output by some factor less than 1. The most common choices:

  • Half Kelly (0.5 Kelly): Bet half the Kelly recommendation. Reduces variance by 75% while giving up only about 25% of expected growth rate. The most popular choice among professional bettors.
  • Quarter Kelly (0.25 Kelly): Conservative. Very low variance but still meaningfully beats flat betting. Good choice if you're uncertain about your probability estimates.
  • Full Kelly: Theoretically optimal if your probability estimates are perfect. In practice, almost no one uses it because estimation error makes it too volatile.

Using the example above (5.5% full Kelly), half Kelly would be 2.75% of bankroll ($27.50 on a $1,000 stake). Quarter Kelly would be 1.375% ($13.75).

Rule of thumb: If you're confident in your probability estimate, use half Kelly. If there's meaningful uncertainty in your estimate, use quarter Kelly. If you're uncertain whether you have an edge at all, flat bet a fixed small unit until you have more data.

How to Estimate True Probability

The hardest part of using Kelly correctly is getting an accurate probability estimate. Here are the main approaches, from most to least reliable:

1. No-vig market prices

Sharp books like Pinnacle and Novig operate with very low or zero vig. Their prices reflect the true market consensus probability more accurately than recreational books. Strip the vig from a Pinnacle line and you have a solid baseline probability estimate. Use our free betting odds converter to calculate the no-vig probability from any set of odds.

2. Multi-model AI consensus

Running your bet through multiple independent AI models and averaging their probability estimates reduces the noise from any single model's biases. When ChatGPT, Claude, and Gemini all arrive at similar probability estimates for a game, that consensus is more reliable than any single estimate. This is exactly what Juice does automatically from a screenshot.

3. Your own research

Injury reports, recent form, pace matchups, coaching tendencies, weather (for outdoor sports). The more complete your picture of the true probability, the more confident you can be in your Kelly input. Our guide on how to find +EV bets covers the full research workflow.

4. Calibration tracking

Track every bet you make along with your estimated probability. After 200+ bets, compare your estimated probabilities to actual outcomes. If you estimate 60% and win 60% of those bets, you're well-calibrated. If you estimate 60% and only win 52%, you're systematically overconfident and should shade your estimates down before inputting them to Kelly.

Kelly Criterion Examples for Common Bet Types

Example 1: Standard -110 spread bet

You estimate 54% probability. Odds are -110 (b = 0.909).

Full Kelly = (0.909 × 0.54 - 0.46) / 0.909 = (0.491 - 0.46) / 0.909 = 3.4%

Half Kelly = 1.7% of bankroll

Example 2: Underdog moneyline at +150

You estimate 42% probability. Odds are +150 (b = 1.5).

Full Kelly = (1.5 × 0.42 - 0.58) / 1.5 = (0.63 - 0.58) / 1.5 = 3.3%

Half Kelly = 1.65% of bankroll

Example 3: Heavy favorite at -300

You estimate 78% probability. Odds are -300 (b = 0.333).

Full Kelly = (0.333 × 0.78 - 0.22) / 0.333 = (0.260 - 0.22) / 0.333 = 12%

This is a surprisingly large Kelly fraction. Heavy favorites require big bets relative to your edge because the profit per dollar is small. Half Kelly = 6% of bankroll.

Let Juice calculate your edge automatically.

Upload a screenshot of any bet. Juice estimates true probability using three AI models, calculates expected value, and shows you where to get the best line across 18 sportsbooks.

Download on the App Store

Common Kelly Criterion Mistakes

Mistake 1: Treating it as a maximum bet size

Kelly is the optimal bet size. Betting less is always safer. Many bettors think of Kelly as a ceiling ("I shouldn't bet more than Kelly") rather than as an input ("this is the right amount to bet"). The distinction matters: if you consistently bet less than Kelly, you're giving up expected growth rate but reducing variance. That's a reasonable tradeoff.

Mistake 2: Using it on correlated bets

Standard Kelly assumes bets are independent. If you're betting on multiple games from the same slate where results might be correlated (e.g., bad weather affects all NFL games), you need to reduce your bet sizes further to account for correlation risk.

Mistake 3: Not adjusting for your actual bankroll

Kelly is a percentage of your current bankroll, not a fixed dollar amount. If your bankroll grows, your bet sizes should grow proportionally. If your bankroll shrinks, your bet sizes should shrink. This is the "Kelly staking" approach and it's what makes Kelly optimal for long-run growth.

Mistake 4: Applying Kelly to negative EV bets

If a bet has negative expected value, Kelly will output a negative number, which means "don't bet." Kelly can only work with positive EV bets. Use our free sports betting EV calculator to check whether a bet has positive EV before sizing it with Kelly.

Mistake 5: Ignoring bankroll ruin risk

Even with fractional Kelly, a long enough losing streak can cause significant drawdowns. Make sure your bankroll is large enough relative to your individual bet sizes that a 20-30 bet losing streak (which will happen) doesn't wipe you out or cause you to change your strategy out of emotion.

Kelly vs. Flat Betting: Which Is Better?

For bettors with a consistent, well-calibrated edge, Kelly (or fractional Kelly) will outperform flat betting over the long run. The reason is straightforward: Kelly bets larger when the edge is larger and smaller when the edge is smaller, which is optimal allocation of risk capital.

For bettors who are still building their track record or who aren't confident in their probability estimates, flat betting a small fixed unit (1-2% of bankroll) is often the smarter choice. It's simpler to execute, easier to track, and doesn't punish bad probability estimates as severely as Kelly.

The honest answer: most recreational bettors should use flat betting until they have 200+ bets of tracked data showing they're genuinely calibrated. Then graduate to fractional Kelly.

Putting It All Together: The Kelly Workflow

  1. Find a +EV bet. Use the full sports betting research workflow from our +EV betting guide. Confirm EV is positive using our free EV calculator.
  2. Get your best probability estimate. Use no-vig lines, AI model consensus, or your own research. Be honest about your confidence level.
  3. Calculate Kelly fraction. f = (bp - q) / b using decimal odds.
  4. Apply fractional Kelly. Multiply by 0.5 (half Kelly) or 0.25 (quarter Kelly) based on your confidence.
  5. Find the best line. Compare odds across books before placing. A better line means a larger edge and therefore a slightly larger Kelly bet. Our odds comparison guide covers this.
  6. Record your estimated probability. Track it against outcomes to calibrate over time.

Frequently Asked Questions About Kelly Criterion

What if Kelly gives me a very large bet size?

Cap at 5% of bankroll on any single bet, regardless of what Kelly says. Very large Kelly outputs usually indicate overconfident probability estimates. A 20%+ Kelly fraction is almost always a sign that your p value is inflated.

Does Kelly work for parlays?

Technically yes, but parlays are rarely positive EV in the first place. If you've confirmed a parlay has positive EV (uncommon outside of specific arbitrage situations), you can apply Kelly to the combined parlay probability and combined payout. In practice, most bettors are better off avoiding parlays unless they've specifically identified mispriced legs.

How do I handle simultaneous bets?

If you're betting multiple games at once, reduce each individual Kelly fraction proportionally. If you have three bets all at 5% Kelly, don't bet 15% of your bankroll at once — scale each back to stay within safe total exposure (typically 10-15% of bankroll across all simultaneous bets).

Is Kelly still relevant in modern sports betting?

Absolutely. The math hasn't changed. What has changed is that edges are smaller and shorter-lived as markets have become more efficient. This actually makes proper Kelly sizing more important, not less — small edges require careful bankroll management to compound meaningfully over time.