"67% accuracy." Put that on a Telegram channel banner and subscribers arrive by themselves. It sounds like two of every three bets winning — how could that possibly lose money?
Easily. A record of short-priced favourites can genuinely win two times in three and still lose money on every unit staked, because the prices it wins at are too short to pay for the third bet that loses. The accuracy claim can be completely true and completely worthless at the same time.
This article is the arithmetic of why — and it is the single most useful piece of maths a bettor can learn.
Profit is a product of two numbers, not one
The result of any betting record is decided by exactly two quantities:
- How often you win (the win rate), and
- How much you get paid when you do (the average odds).
Multiply them, and you have your return per unit staked (with equal stakes and no refunds):
Return = win rate × average odds of the winning bets
If the product is above 1.00, the record is profitable. Below 1.00, it loses. Careful with the second number: it is the average price of the bets that won, not of all bets. When a record mixes short and long prices, using the overall average can give the wrong answer — sometimes even the wrong sign. A win rate quoted without its average odds is exactly as informative as "this car does 6,000 RPM" — impressive-sounding, and meaningless without the gear.
The break-even table every bettor should memorise
Because profit is a product, every level of odds has a break-even win rate: one divided by the odds (for a record at roughly the same odds). Below it you lose, above it you win — regardless of how the raw percentage feels.
| Average odds | Win rate needed to break even |
|---|---|
| 1.25 | 80.0% |
| 1.48 | 67.6% |
| 1.60 | 62.5% |
| 2.00 | 50.0% |
| 2.50 | 40.0% |
| 3.10 | 32.3% |
| 5.00 | 20.0% |
Take a record whose winning bets paid 1.48 on average: break-even is 67.6%. Win 67.0% and you are just short. That gap — six tenths of a percentage point — sounds like nothing. Compounded over thousands of bets, it is a steady leak: 0.67 × 1.48 = 0.992, a loss of 0.8 cents on every unit staked, forever.
Meanwhile a bettor winning only 35% of the time at average odds of 3.10 returns 0.35 × 3.10 = 1.085 — a healthy 8.5% profit while losing almost two of every three bets. If you showed both records to a casual observer and revealed only the win rates, they would rank them exactly backwards.
Why high win rates are free (and therefore worthless)
Here is the uncomfortable part: a 67% win rate at short odds requires no skill whatsoever.
Football markets are efficient enough that the favourite simply wins most of the time. In our own dataset of top-league matches, blindly picking the market favourite on every match is correct roughly 55% of the time — with zero analysis, zero models and zero insider knowledge. Lean on heavier favourites only, and the raw percentage climbs into the sixties and seventies by itself.
Want 80%? Sell double-chance picks. Want 90%? Heavy favourites on Asian handicap lines where half the "wins" are refunded stakes. The win rate is a dial the seller can set almost arbitrarily by choosing what kind of odds to tip — and every notch upward on the dial pushes the average odds down and the break-even bar up to meet it.
That is why the bookmaker doesn't fear your win rate. The margin baked into every price means the default outcome — for a random picker, for a favourite-backer, for a 67% accuracy machine — is a small, steady loss. Accuracy is not the game. Beating the price is the game.
The five-second test
From now on, every time you see a win rate, perform this test before feeling anything:
- Find the average odds of the winning bets. If no odds are published at all, that is your answer already — the number is hiding its other half for a reason. (If only the overall average is published, the test below is a rough guide, not a verdict.)
- Multiply. Win rate × average winning odds.
- Compare with 1.00.
"67% at 1.48" fails (0.99). "20% winners at odds of 26" from the ROI-chart heroes we broke down previously needs the odds to be genuine and the sample to be more than a lucky quarter — variance at odds of 26 takes thousands of bets to wash out. "55% at 2.10" passes (1.155) — and would be extraordinary, which is precisely why records like it deserve the hardest scrutiny, not the least.
What actually separates skill from packaging
If win rate can be manufactured and short samples can lie, what can't be faked? The answer is the benchmark professionals use on themselves: closing line value — whether the price you took beats the final, sharpest price the market settled on before kick-off. It converges on the truth far faster than profit does, and it cannot be gamed by pick selection, because it re-prices every single pick against the best available estimate of reality.
We explain it in plain language in our CLV explainer, and it is the metric our entire public track record is built on — every pick logged before kick-off, settled against the close, losses published forever.
One last honest note, because it is the entire point of this site: when we ran these same tests on our own models across tens of thousands of matches, we could not beat the sharp market's prices either. Nobody selling you a subscription on the strength of a win rate has done what that number implies. The two-number test takes five seconds. Run it on everything — including us.