How much should I bet per game?
Written by Bet selection decides whether you have an edge. Bet sizing decides whether you're still around to collect it. Almost everyone who blows up a bankroll does it with the second one.
The full menu of bankroll strategies
There are more ways to size bets than the two this article recommends. Knowing the full list matters even if you never use most of it, because it's how you recognize which one you're actually doing right now.
What a unit is
A unit is one standard bet, defined as a percentage of your bankroll rather than a dollar amount. On a $2,000 bankroll, a 1% unit is $20. The percentage is the real definition. That's what lets the same rule work at $500 and at $50,000.
Flat betting
Flat betting means every bet is one unit. It sounds unsophisticated, and it's what the majority of disciplined bettors actually do. Its virtue is that it makes your results readable: if every bet is the same size, your record means what it appears to mean, and no single loss can be catastrophic.
The alternative most people fall into isn't a system. It's confidence-based sizing that quietly becomes chasing. Bigger after losses to get even, bigger after wins because you're 'running good'. Both are the same mistake wearing different moods.
Here's the actual reasoning for why flat betting beats confidence-based sizing for almost everyone, not just a rule to take on faith. Sizing a bet correctly by 'how confident you are' requires that your confidence is an accurate readout of your real win probability on that specific bet. That's exactly what the Kelly criterion below needs too, a genuine, tested probability estimate, not a feeling. If you don't have a model or a track record that actually calibrates your gut sense against real outcomes, 'this one feels like a lock' and 'this one feels shaky' aren't measuring your edge. They're measuring your mood, how a broadcast narrative framed the game, or how your last few bets went. Betting more on the ones that feel good adds variance to your results without adding any real information, which is strictly worse than doing nothing. Flat betting sidesteps the whole problem: it makes your bet size independent of a read you can't actually verify, and it isolates the one skill you're trying to measure, picking games, from a sizing skill most bettors don't actually have yet.
The Kelly criterion
Kelly is the mathematically optimal bet size for maximizing long-run bankroll growth, given a known edge:
For a 55% opinion at -110 (b = 0.909): f = (0.909 × 0.55 − 0.45) ÷ 0.909 = 5.5% of bankroll. That is an enormous bet by normal standards, and it's the correct answer only if 55% is exactly right.
When should you actually raise your unit size?
The two strategies above handle this completely differently, and mixing them up is its own mistake. Kelly already has the answer built into the formula: it takes your current bankroll and your calculated edge as direct inputs, so the bet size adjusts automatically every time either one changes. There's no separate decision to make. Raise the bankroll figure you plug in, or update your edge estimate with more data, and the correct size moves with it.
Flat betting doesn't have that built in, which means the bettor has to make the judgment call by hand. The mistake almost everyone makes is upping the unit right after a good week or a hot month, exactly the sample size least likely to mean anything. The better version of this decision waits for a real stretch, a full season, a large enough sample that the result is more likely to reflect actual skill than a run of variance, before recalculating what percentage the bankroll should be. If you can't articulate why now is the right moment beyond 'it's been going well lately,' it isn't the right moment yet.
Sizing for the drawdowns that will happen
At a 55% win rate, five straight losses happen about 1.8% of the time, which across a full season of betting means it will happen to you more than once. At a more realistic 53%, it happens more often than that. Your unit size has to be a number you can lose ten of in a row without changing how you bet.