A 90% win rate can lose money when you back low odds. Winning frequently is only part of the calculation. The other part is how little each win earns compared with the amount lost when a bet fails.
At decimal odds of 1.05, a winning €100 back bet earns €5 before commission. A losing one costs €100. You need twenty such wins to replace one full loss before fees. Low odds describe the payout; they do not make the stake safe.
Implied probability is a break-even calculation
For a back bet at decimal odds o, a win earns o − 1 units per unit staked and a loss costs one unit. Before commission, the break-even probability is p = 1 ÷ o. Multiply by 100 to express it as a percentage.
This is an implied threshold, not proof of the selection’s true chance. A price of 1.10 implies approximately 90.91% before commission. If the true chance were only 90%, the price would be too short to break even over repeated comparable bets.
Across a full market, prices can also reflect the spread, available liquidity and trading conditions. Do not treat a single displayed quote as a measured probability. The relevant odds for your result are the prices at which your bets actually matched.
How many wins does one loss erase?
With equal back stakes, wins needed to offset one loss = 1 ÷ (odds − 1), before commission. This compares net profit on wins with the full losing stake. Including the returned winning stake would make the comparison meaningless.
| Decimal odds | Implied probability | Wins per loss |
|---|---|---|
| 1.05 | 95.24% | 20 |
| 1.10 | 90.91% | 10 |
| 1.20 | 83.33% | 5 |
| 1.25 | 80.00% | 4 |
These ratios assume the same stake on every bet. If the losing stake is larger, one loss erases more wins. Raising the stake after a winning run can therefore make a high overall win rate coexist with a substantial cash loss.
A hypothetical 90% win-rate scenario
Hypothetical numbers, not observed results: each row below assumes 100 separate bets, 90 wins, 10 losses and a fixed €100 stake. Every bet in a row has the stated odds. Commission and other costs are excluded so the payout arithmetic is clear.
| Odds | Profit from 90 wins | Cost of 10 losses | Total P&L |
|---|---|---|---|
| 1.05 | €450 | −€1,000 | −€550 |
| 1.10 | €900 | −€1,000 | −€100 |
| 1.20 | €1,800 | −€1,000 | +€800 |
| 1.25 | €2,250 | −€1,000 | +€1,250 |
The positive rows are not a strategy or a forecast. They assume the same 90% success rate at different prices, which you cannot simply choose to obtain. The table isolates how payouts change the result at a given win rate.
For this equal-stake setup, total gross P&L is stake × [wins × (odds − 1) − losses]. With variable stakes or odds, calculate each result separately. Multiplying an average stake by an average price can hide where the largest losses occurred.
Commission makes the threshold higher
Ordinary winnings commission reduces profitable market results. It does not reduce the stake lost on a losing back bet. For one bet per market, break-even probability = 1 ÷ [1 + (odds − 1) × (1 − commission rate)].
Hypothetical fee example: at 1.10 with an assumed 5% commission, a €100 win earns €9.50. The break-even win rate becomes about 91.32%. Replacing a €100 loss takes approximately 10.53 equal wins, so eleven whole wins are needed to get above zero.
The assumed rate is not Orbit’s stated rate. Check your own terms and read how market commission works before applying a per-bet calculation to positions that share a market.
Why 30 bets do not prove an edge
A short record tells you what happened, but gives a noisy estimate of what might happen next. In a hypothetical sample of 30 bets, 27 wins give a 90% win rate. One different outcome changes that rate by 1 ÷ 30, or about 3.33 percentage points.
Standard error describes the typical sampling wobble around an estimated win rate under a simple model. For independent bets with the same win probability, a rough estimate is √[p × (1 − p) ÷ n], where p is the observed win fraction and n is the number of bets.
Putting 0.90 and 30 into that formula gives about 0.055, or 5.5 percentage points. That is large relative to the gap between a 90% record and the 90.91% break-even threshold at 1.10. It is not a guaranteed range, and a normal approximation is rough with so few losses.
Under the same simplified assumptions, quadrupling the sample halves standard error. Real records can be less informative if bets share events, conditions change or you selected the strategy after seeing its results. More bets do not repair biased record keeping or guarantee future profit.
Review the risk as well as the result
Variance means outcomes fluctuate, including losing clusters. A smooth sequence of small gains can be interrupted by a full-stake loss. Record matched odds, stakes, market identifiers and net P&L; keep all results, including abandoned approaches.
Use the bankroll guide to distinguish stake from tolerable loss. Use the balance guide and real-profit calculation to keep open exposure and deposits from distorting performance.
Frequently asked questions
Can a 90% win rate lose money?
Yes. With equal stakes at odds of 1.10, nine wins earn 0.9 stakes and one loss costs a full stake. The result is a loss of 0.1 stakes before commission.
Are odds of 1.05 safe?
No. A losing back bet still costs the full stake. At 1.05, twenty equal-stake wins are needed to replace one loss before commission.
Do 30 bets prove that a method works?
No. They provide a small, noisy record. The win rate alone also ignores odds, stake sizes, commission and selection bias. A larger sample helps estimation but cannot guarantee an edge.
Does implied probability predict the outcome?
No. One divided by decimal odds gives the break-even threshold before commission. It does not establish the true probability or guarantee that a favourite will win.
Keep your result in view
Export or copy your own exchange statement and broker history, then import them into Stakeboard. View pricing or join the early-access list.
Get early access