BY 
I need to open with a confession, because it's the whole reason to hear me out on this topic. I've blown up three bankrolls. Three.
Not one of them died because my picks were unusually bad. They died from sizing. Doubling after losses, betting money that had a job elsewhere. The classics. CAPTRACKER tracks over 900 handicappers with every pick auto-settled against ESPN data, and the gap between profitable-looking and actually profitable almost always runs through stake discipline.
Step one, define the bankroll
Your bankroll is money set aside for betting and nothing else, and its total loss should change nothing about your life. Not rent. Not the savings you'd rather not touch. A sealed pool, period.
That rule sounds like finger-wagging, but it's really about performance. Bettors risking money they need make systematically worse decisions. They chase, and they hedge at horrible prices. My first blowup was exactly this, a bankroll that was secretly also my car repair fund, and you can guess how calmly I handled a bad week.
If naming a number feels hard, start smaller than feels impressive. You can always grow a bankroll that exists. Torching one to prove a point sets you back years.
Units, the language of honest records
A unit is a fixed percentage of your bankroll, and the standard range is 1% to 2%. A $1,000 bankroll at 1.5% makes one unit $15. Small, I know. That's the point.
Units do two jobs. They force your stakes to scale with your actual capital instead of your mood, which is the exact failure mode that killed my second roll. And they make records comparable across bettors, which is why the CAPTRACKER leaderboard ranks handicappers by units won and ROI. A +40u year means the same thing whether the capper bets $10 a unit or $1,000.
One habit that helped me more than any spreadsheet: write your unit size down somewhere you'll see it before you bet, and don't renegotiate with yourself mid-slate. The renegotiation only ever goes one direction. It's never smaller.
The drawdown math nobody wants to sit with
Losing streaks are a scheduled event. A genuinely good 55% bettor will run into a 7-loss streak regularly over a few thousand bets. That's arithmetic, and arithmetic doesn't care that you're due.
Now the recovery math, and read this part twice. Lose 33% of your bankroll and you need to gain 50% just to get back to even. Lose 50% and you need a full 100% gain to recover. The cost of climbing out grows faster than the hole that put you there.
At 2% flat stakes, a 7-loss streak costs about 13% of your roll. Annoying, survivable. At 10% stakes, the same streak takes over half your bankroll, and now you need to double what's left just to break even.
Same picks and the same skill, one busted account. Oversizing turns ordinary variance into ruin, and I've got three dead bankrolls' worth of receipts on that.
Flat, percentage, or Kelly
Flat staking means every bet is one unit, with the unit recalculated occasionally as the roll moves, weekly or monthly. Simple and sturdy. It's the right default for almost everyone, including you, and definitely including me.
Percentage staking recomputes the stake from your current bankroll on every bet. It throttles itself in a drawdown, which is genuinely nice, at the price of more bookkeeping. If you're not sure which of the two fits you, that uncertainty is itself the answer, and the answer is flat.
Kelly sizing stakes in proportion to your edge divided by the odds. It's elegant on paper and dangerous in the wild, because full Kelly assumes you know your edge precisely, and you don't. Nobody does. If you touch Kelly at all, treat quarter-Kelly as a ceiling, and only with a verified long-sample record behind the edge estimate.
The rules that did the real work for me
- Cap variable sizing at 1 to 3 units. If your confidence scale runs past 3u, it's measuring your emotions instead of your information.
- Never raise your stake to chase. The double-it-back-tonight instinct is the single most reliable account killer in this hobby, and it's the one that got my third bankroll.
- Parlays get fractional units. Compounding legs compounds the book's hold, so size them like the long shots they are.
- Audit monthly, in units, segmented by sport and market. Hold anyone you tail to the same standard, meaning verified units over a real sample. Every pick on the daily feed is timestamped and locked before game time, then settled by ESPN data, and the whole process is documented on the methodology page.
What discipline actually buys you
Notice what none of this requires. You don't need better picks or a smarter model, because sizing is the one edge that never depends on beating the market's information. That's exactly why so many people skip it. It feels too boring to be the answer, and I skipped it three times myself.
Bankroll management won't create an edge. What it does is keep an edge alive long enough to matter. A modest verified edge, sized at 1% to 2% a bet, compounds quietly in the background of your life.
A brilliant record sized recklessly is one cold fortnight from zero. I know because I've been the guy holding that record, twice, wondering where the money went. Protect the roll. The roll is the whole game.
And if you take one sentence from a three-time loser, take this one: the picks were never the problem.