In our AI Betting Playbook case study, the model analysed 8,439 bets, returning +398 points. Results are expressed in points (sometimes called units), with every qualifying opportunity recorded at the same stake. That framing is deliberate: the record reflects the quality of the model’s selections, not clever staking. Nobody staked a bet bigger because it felt stronger, and nobody chased a losing run. If the selections did not hold an edge, a points record staked this way would show it plainly.
For your own bankroll, the same principle is the sensible starting point. Set aside a betting bank you are entirely comfortable with, divide it into points (100 points is a common convention, so a £500 bank means £5 per point), and stake the same amount on each qualifying opportunity. Level staking is simple, it makes your results comparable like-for-like with the case study, and it protects you from the most expensive habit in betting, which is increasing stakes to recover losses.
There are other approaches. Percentage-of-bank staking recalculates your stake as a fixed share of your current bank, so stakes shrink during losing runs and grow during winning ones. Some experienced bettors vary stakes with the size of the edge. These are legitimate options with their own trade-offs, and none of them will rescue selections that lack an edge. Whichever you choose, two rules hold: never stake money you cannot afford to lose, and judge your results over hundreds of bets, not a weekend. Value betting profits arrive through volume, with variance along the way, and a staking plan you can hold through a losing run beats an optimal-on-paper one you abandon halfway.
Further reading on the blog: Bankroll Management for the Data Era.