The week that taught me what variance really feels like
I had what I thought was the sharpest week of my NFL season in November 2021. Eight bets across the Sunday slate. Strong analytical case on every one. I went 1-7. Not because the analysis was wrong – closing line value across the eight bets was positive, meaning my prices were better than the market’s closing prices in aggregate – but because variance happens. The week cost me about 11% of my betting bankroll because I’d been stake-sizing too aggressively on “high conviction” picks. The recovery took two months of disciplined play.
That experience reshaped how I think about bankroll management. Most NFL betting advice focuses on picking winners. Bankroll management is the discipline that determines whether picking winners actually makes you money or just makes you feel clever right before you blow up. The math is unforgiving. A 55% bettor (which is genuinely sharp for NFL spreads) still has multi-week losing streaks that look indistinguishable from being a 45% bettor in the short run.
This is the working framework for stake sizing that I’ve used through several NFL seasons. It’s not glamorous. It’s slow. And it’s the single biggest reason I’m still in the market when many of the punters I started alongside aren’t.
Defining a betting bankroll
The first principle: your betting bankroll is money you can afford to lose entirely without affecting any other part of your life. It’s not a transfer from this month’s salary that you’ll “top up” later. It’s not money earmarked for the rent or a car payment. It’s a defined sum sitting in a separate account or wallet, with a clear ledger, and a number you’ve decided is acceptable to lose.
The size of the bankroll depends on your financial situation, but the rule I’d apply: any single year’s betting bankroll should be small enough that losing all of it would be unpleasant but not catastrophic. For most UK punters that probably means anywhere from £200 to £2,000 across a full NFL season. Pick a number, commit to it, and don’t “top up” mid-season if you lose it.
The mental discipline of treating the bankroll as a fixed resource rather than a renewable one changes everything downstream. It forces you to consider stake size as a fraction of finite capital, not as an isolated decision per bet. It forces you to accept losses without trying to chase them. And it forces you to scale your stake to reality rather than to ambition.
The Gambling Commission’s Financial Vulnerability Check threshold of £150 net losses over a rolling 30-day window, in force since 28 February 2025, gives you a useful natural calibration. If your bankroll-management plan involves regularly losing more than £150 over a month, you’re going to trip checks regardless of your personal financial position. Plan your stakes such that even adverse variance keeps you under the threshold unless you’ve actively decided to bet larger and accept the operator scrutiny that follows.
Flat staking vs percentage staking
The two main approaches to stake sizing are flat staking (every bet is the same monetary amount) and percentage staking (every bet is a fixed percentage of the current bankroll). Both have merit. Both have failure modes.
Flat staking is the simpler approach. Pick a unit size – say, £5 per bet on a £500 bankroll, which is 1% per bet – and bet that amount on every position you take. The advantage is mechanical discipline; you can’t talk yourself into doubling up on a “sure thing” because the rule says you bet £5. The disadvantage is that flat staking doesn’t compound. If your bankroll grows from £500 to £1,000, your £5 bets become 0.5% of bankroll, which is more conservative than you should probably be running.
Percentage staking adapts to bankroll size. You decide, say, that each bet will be 1.5% of current bankroll. On a £500 bankroll, that’s £7.50 per bet; on a £1,000 bankroll, that’s £15; on a £300 bankroll (after losses), that’s £4.50. The bet size automatically scales down during losing streaks and up during winning streaks, which dampens variance and accelerates compounding.
The disadvantage of percentage staking is that it requires you to maintain a clean ledger and update your bet size after every settled bet. Most recreational punters don’t, so they end up with hybrid systems that approximate percentage staking without actually implementing it. The result is usually overbetting during downswings and underbetting during upswings, which is the worst of both approaches.
My personal system: flat unit staking with periodic recalibration. I pick a unit size at the start of the season (typically 1-1.5% of starting bankroll), bet that amount throughout, and recalibrate every six weeks based on current bankroll. The recalibration intervals are short enough to capture compounding but long enough to avoid the mid-streak panic adjustments that destroy disciplined systems.
Unit sizing for NFL specifically
NFL has 18 regular-season weeks plus a playoff format with limited games per round. Across a typical season, a recreational punter places perhaps 50-150 bets if they’re betting weekly, more if they’re active on props and live markets. The unit size you choose has to survive the variance distribution of that number of bets.
The rough math: a 53% bettor on NFL spreads (genuinely sharp – not many bettors sustain this) running at 1% stake size across 100 bets has roughly a 5% chance of ending the season in the red despite betting at a positive expected value. The same bettor at 3% stake size has a 12-15% chance of ending in the red, and a meaningful chance of blowing up entirely during a bad month.
For genuinely sharp NFL bettors, 1-2% per bet is the conventional range. For recreational bettors who are honest about whether they’re picking at 50% or 55%, the range should be lower – 0.5-1% per bet – because the variance impact is the same but the expected return doesn’t cushion it.
The temptation to vary unit size based on “conviction” is the biggest single mistake in NFL stake sizing. “This is my best bet of the week” feels like a reason to bet 3% instead of 1%. The math says it isn’t. Your conviction-rating mostly tracks your emotional engagement, not your actual edge. Bets you label “best of the week” hit at roughly the same rate as bets you label “normal” – unless you’re tracking your conviction labels against actual outcomes over a multi-year sample, which almost no recreational bettor does. The disciplined approach is to bet the same unit regardless of perceived conviction, and to refuse to bet at all on positions where the conviction isn’t there.
Kelly criterion in practice
The Kelly criterion is the mathematically optimal stake-sizing formula given a known edge and known odds. The full formula: optimal stake fraction = (edge × decimal odds – 1) / (decimal odds – 1). At its theoretical best, Kelly maximises long-run logarithmic bankroll growth.
The problem with Kelly in NFL betting is the edge estimation. The formula assumes you know your edge precisely. Most punters don’t. The variance in actual edge estimates is so high that full-Kelly stake sizing produces stake recommendations that swing wildly between 0% and 8% of bankroll, which generates uncontrolled variance and frequent psychological catastrophes.
The practical solution is fractional Kelly. Take whatever stake size Kelly recommends and bet 25-50% of it. Quarter-Kelly is the conservative end of the range and produces a smoother bankroll curve at the cost of slower expected growth. Half-Kelly is closer to optimal but more volatile. Full Kelly is for punters who genuinely know their edge with high precision, which is very few people in practice.
The other adjustment: cap the Kelly recommendation at a hard maximum (typically 3-5% of bankroll, regardless of what the formula says). If the formula recommends 7% on a bet, something is probably wrong with either your edge estimate or the price you’re looking at. Caps protect you from your own modelling errors. The implied probability piece covers how to estimate edge against market prices in a way that feeds usefully into stake-sizing decisions.
For most NFL recreational bettors, the honest application of Kelly is: don’t bother with the formula explicitly, but use the underlying principle. Bet more on bets where you have clearer edges and less on bets where the edge is marginal, but keep the variation within a narrow band (say, 0.5-1.5% of bankroll) rather than swinging from 0% to 5%.
Recovery from drawdowns
Every NFL bettor experiences drawdowns. The question isn’t whether you’ll lose; it’s how you respond when you do. The recreational pattern is to increase stake size during losing streaks to “chase” the deficit. The math says this is exactly wrong – a 30% drawdown requires a 43% gain to recover. Doubling stake size to recover faster increases the chance of compounding the loss into a 50% or 60% drawdown, from which recovery becomes nearly impossible.
The disciplined response to drawdowns is the opposite of intuitive. Reduce stake size during losing streaks rather than increasing it. This isn’t because the losses change your underlying edge – they don’t – but because the psychological pressure of mounting losses degrades decision quality, and smaller stakes during downswings let you continue betting without compounding emotional damage.
My personal rule: if I’m down 20% from peak bankroll, I cut unit size by a third for the next 20 bets. If I’m down 30%, I cut by half. If I’m down 40%, I stop betting entirely for the rest of the NFL week, review my analytical process, and only resume when I can identify what’s gone wrong and what to change. These aren’t optimal rules from a pure expected-value standpoint, but they’re the rules that have kept me in the market while watching other punters cycle through bankrolls and then quit.
The wider context matters too. The Q4 2024-25 figure of £596 million in online real-event GGY across the UK and the 290 million online sports bets placed monthly include a vast number of recreational punters in drawdown at any given moment. The patterns that distinguish punters who recover from punters who blow up are not the picks they make. They’re the stake-size decisions they make during the recovery process. Patience compounds. Panic compounds faster, in the wrong direction.
