The reason I tried five staking systems before settling on one
My first three NFL seasons of serious betting, I changed staking systems four times. Started with flat staking because everyone said it was sensible. Switched to Martingale (doubling after losses) after a particularly painful week, which lasted exactly two more weeks before I nearly busted out. Tried Fibonacci (a slower progression after losses), which felt smarter but produced the same problem on longer downswings. Then percentage staking, which I’d half-understood. Finally fractional Kelly, which I now use as the foundation of my approach with significant modifications.
Each transition cost me money. Not from picks going wrong, but from being inconsistent with my stake sizing across a season – betting small when I should have been pushing, betting big when I should have been pulling back. The realisation that eventually stuck: there is no perfect staking system, but there are clearly bad ones, and consistency with any reasonable system matters more than the marginal differences between reasonable systems.
This is the comparison I wish I’d had when I started. The advantages, the failure modes, and the staking patterns that genuinely match how NFL betting variance actually works.
Flat staking as the baseline
Flat staking is the default for good reason. Every bet is the same monetary amount, regardless of conviction, regardless of bankroll changes, regardless of recent results. The discipline is mechanical: £5 per bet means £5 per bet, full stop.
The maths is straightforward. If you’re betting £5 per bet at average odds of 1.95 (slightly less than evens, typical for spread bets), break-even hit rate is 51.3%. Above that, you’re profitable; below, you’re losing. Variance over 100 bets at a 53% hit rate produces a typical end-of-season bankroll change between -8% and +30%, with rare outliers in both directions.
What flat staking does well: it eliminates the most common recreational mistakes – sizing up after wins (overconfidence), sizing up after losses (chasing), sizing up on “sure things” (overrating conviction). The fixed unit removes the decision from the moment of the bet, which removes the emotional component that destroys most amateur staking discipline.
What flat staking does poorly: it doesn’t compound. As your bankroll grows, your unit stake becomes a smaller percentage of total capital, which means you’re effectively becoming more conservative over time without any active decision. Over a multi-year horizon this can leave significant expected growth on the table.
The practical fix is periodic recalibration. Every six to eight weeks, recalculate your unit size based on current bankroll. If you started at £5 on a £500 bankroll (1%) and you’re now at £650, your new unit becomes £6.50. This recalibration is slow enough not to introduce mid-streak instability and frequent enough to capture meaningful compounding.
Progressive systems and why they fail
Progressive staking systems – Martingale, Fibonacci, d’Alembert, Labouchere – all share the same underlying idea: adjust stake size based on recent results in a way that supposedly recovers losses or banks wins. They all fail for the same underlying reason: they assume losses are recoverable through stake-sizing, when in reality losses are recoverable only through good picks at fair prices over time.
Martingale doubles the stake after every loss until a win recovers all prior losses plus a unit profit. The math looks tempting until you run the variance simulation. Across 100 bets at 50% hit rate, the probability of hitting a 6-loss streak is around 1 in 64 – once or twice in a typical NFL season. Hit that streak on a Martingale system starting at £5 and you’re betting £160 on the seventh bet, with cumulative exposure of £315 to recover the original £5 loss. Hit a 7-streak and you’re betting £320 on the eighth bet, exposure £635. Most UK sportsbooks would refuse the bet sizes long before the system theoretically recovers, and your bankroll would be gone regardless.
Fibonacci is a slower version of the same idea. After each loss, the next stake is the sum of the previous two stakes. The progression is gentler than Martingale (1, 1, 2, 3, 5, 8, 13…) but the failure mode is identical: a long enough losing streak produces stake sizes that exceed your bankroll or the operator’s limits.
The d’Alembert system increases stake by one unit after a loss and decreases by one unit after a win. It’s the gentlest of the progressive systems and rarely produces catastrophic blow-ups in moderate variance environments. But it also doesn’t actually deliver the promised steady profits in NFL betting because NFL hit rates aren’t 50-50. The asymmetry between the price you pay on a bet (typically 1.91 decimal at standard juice) and the breakeven probability (52.4%) means a 50%-hit-rate bettor loses money on d’Alembert just like on flat staking, but with extra variance.
The unifying problem: every progressive system tries to solve a probability problem with a stake-sizing rule, and probability problems can’t be solved that way. If your underlying picking is at a 51% hit rate against 52.4% breakeven, no staking system saves you. If you’re at 54%, almost any reasonable system works – progressive systems just add variance you don’t need.
Percentage of bankroll staking
Percentage staking is the most defensible non-Kelly system for NFL betting. The principle: each bet is a fixed percentage of current bankroll (typically 1-3%). The percentage stays constant; the absolute pound amount shifts with bankroll changes.
The advantages are real. On winning streaks, your stake size grows naturally, capturing more value as the bankroll grows. On losing streaks, your stake size shrinks naturally, reducing the absolute pound impact of further losses. The system has a built-in safety mechanism – you can’t bust out in a normal variance scenario because your stake is always proportional to what you currently have.
The disadvantage is the operational burden. To bet a true 1% of bankroll each time, you have to update your bankroll after every settled bet and calculate the new stake. Most punters approximate this badly: they bet roughly 1% but stick at a unit size for too long, which means they’re effectively flat-staking during the period and the percentage rebalancing only happens occasionally.
The practical version: bet a percentage of bankroll updated weekly rather than after each bet. This captures most of the percentage-staking benefit while avoiding the operational complexity. It also smooths out the within-week variance that can otherwise cause stake sizes to swing meaningfully between bets.
The 290 million UK online sports bets placed monthly include enormous variance in stake sizes, but the punters who survive multi-year horizons almost universally use some form of percentage- or fractional-Kelly staking. The recreational punters who blow up are almost universally flat-staking aggressively or running progressive systems that haven’t yet caught the bad streak that ends them.
Kelly criterion and fractional Kelly
The Kelly criterion is the staking formula that maximises long-term logarithmic bankroll growth, given an accurate edge estimate. The formula in betting form: optimal stake fraction = (bp – q) / b, where b is the decimal odds minus 1, p is your estimated probability of winning, and q is 1 – p.
Worked example: you think a +6.5 underdog has a 55% chance of covering, and the line is priced at 1.91 decimal. b = 0.91, p = 0.55, q = 0.45. Optimal Kelly fraction = (0.91 × 0.55 – 0.45) / 0.91 = (0.5005 – 0.45) / 0.91 = 0.055, or 5.5% of bankroll.
That 5.5% number is unusable in practice. The reason: your 55% estimate is probably wrong. Your actual edge might be 53% or 51% or 56%. The Kelly formula is exquisitely sensitive to edge estimation – it assumes you know your edge precisely – and recreational bettors don’t.
Fractional Kelly addresses this by betting a fixed fraction of the Kelly recommendation. Quarter-Kelly (25% of the recommended stake) is the conservative end of the spectrum. Half-Kelly (50%) is closer to the theoretical optimum but more variant. In the worked example above, quarter-Kelly would suggest betting 1.4% of bankroll, half-Kelly would suggest 2.75%.
Fractional Kelly produces several useful properties. It scales naturally with bankroll (because it’s a percentage), it sizes bets according to perceived edge (more on high-edge bets, less on low-edge bets), and it’s robust to edge-estimation error (because you’re betting fractionally below the optimum). The trade-off versus simple percentage staking is that you need an edge estimate for every bet you place, which forces explicit analytical thinking before betting.
My personal version: quarter-Kelly with a hard cap of 2% of bankroll per bet. The cap protects against modelling errors that would otherwise produce stake recommendations of 4-5%. The fractional-Kelly base captures most of the bet-sizing benefit without the variance whiplash of full Kelly.
Adapting staking to NFL variance
NFL has specific variance characteristics that affect staking. The 18-week regular season plus playoffs creates roughly 100 betting days for active punters, with most weeks featuring multiple correlated bet decisions. The 2025 NFL regular season’s distribution shift – favourites won outright 65.9% but covered the spread only 47.8% – produced unusual variance in spread-betting outcomes, with multiple multi-week losing streaks across the betting community even for sharp picks.
The lessons that aren’t always obvious:
NFL bets are not fully independent. Bets on the same Sunday slate correlate because league-wide variance patterns (high-scoring weeks, blowout weeks, weather-affected weekends) affect multiple games at once. A staking system that assumes independence can underweight the within-week correlation, leading to oversized exposure to a single Sunday’s outcomes. The practical fix: cap the total stake exposure across a single day at 5-7% of bankroll, regardless of what individual-bet Kelly recommendations suggest.
Long shots create exaggerated variance. Backing a +400 underdog or playing four-leg parlays in moderate size produces variance that doesn’t smooth out across 100 bets the way 1.91-decimal spread variance does. If you’re playing a mix of bet types, weight your stake sizes downward on the high-variance products. The same Kelly fraction applied to a 4.00-odds outcome produces much more bankroll swing than applied to a 1.91 outcome.
Live betting compounds bet count and variance. If you started the season planning 50 bets and you’re now placing 5 live bets per Sunday on top, your effective bet count has doubled. The same flat stake size now represents twice the seasonal exposure. The honest response is to halve your unit size if you take up live betting seriously, not to maintain the original unit and rely on “these are smaller bets”. The live betting piece covers the latency and variance considerations specific to in-play markets.
The Q4 2024-25 figure of £596 million in online real-event GGY across the UK and the 1.1% statutory levy introduced in April 2025 are macro context for how the regulator sees the market. The Financial Vulnerability Check threshold of £150 net losses over 30 days is a more direct constraint – it caps how aggressively you can bet before tripping operator scrutiny.
