Building a model for sports betting is possible, but there is a wall in front of the model and its name is the margin. The margin is the commission buried inside the odds, and working it out takes a single addition.
Turning odds into probabilities
The reciprocal of a decimal odd is the implied probability of that outcome: odds of 2.00 mean 50%, odds of 4.00 mean 25%. Add up the implied probabilities of all three outcomes and the result is not 100% — the excess is the bookmaker’s share.
The odds below are the real odds standing on the Patekbahis home page on 5 October 2026, and the calculation is made on them.
| Match | Odds (1 / X / 2) | Total implied probability | Excess | Bookmaker’s share |
|---|---|---|---|---|
| Italy – Turkey | 1.39 / 5.10 / 7.10 | 105.6% | 5.6 points | 5.3% |
| France – Belgium | 1.47 / 4.75 / 6.00 | 105.8% | 5.8 points | 5.4% |
| England – Czechia | 1.11 / 8.20 / 15.00 | 109.0% | 9.0 points | 8.2% |
| Croatia – Spain | 8.60 / 5.90 / 1.24 | 109.2% | 9.2 points | 8.4% |
The difference between the last two columns matters: the excess is how far above 100 the total sits, while the bookmaker’s share is that excess as a proportion of the total — 5.6 divided by 105.6, about 5.3%. The second figure is the real rate deducted from money wagered over time.
What the table says
The margin varies from match to match, and it grows on lopsided ones: 5.6 to 5.8 points on the two even matches, 9.0 to 9.2 points on the two with an overwhelming favourite. On the same site on the same day, in other words, the cost of a bet nearly doubles depending on which match you pick. The value of a model is measured first by whether it clears that gap.
What a model can and cannot do
- It can: assess a large number of matches by the same criterion, compare its own probability estimate with the one implied by the odds, and keep consistent records.
- It cannot: remove the margin. Unless the model finds a gap between its own estimate and the implied probability that is larger than the margin, it has not found a bet.
- The common mistake: treating a hit rate as success. A 70% hit rate at average odds of 1.30 finishes in the red; what matters is not the hit rate but the hit rate multiplied by the odds.
A practical test: work out the margin on a match before taking the bet. On a match with a 9-point margin your model has to beat the implied probability by at least 9 points — a threshold that can be calculated, and is not easy to clear.
On the casino side the equivalent figure is the house edge: the rake at a poker table does the same job in another form, and slot themes and the maths looks at where the RTP is written. Beating the margin by reordering stakes is a misconception as old as the house edge at card tables.
For the term itself and its history, see the entry on vigorish.
Frequently asked questions
How is the margin worked out from the odds?
Take the reciprocal of each odd and add them up. However far the total exceeds 1 is the excess; the bookmaker’s real share is the excess divided by the total. Example: 1.39 / 5.10 / 7.10 adds up to 1.056, so the excess is 5.6 points and the share about 5.3%.
Is the margin the same on every match?
No. On 5 October 2026, on the same site, it was 5.6 to 5.8 points on even matches and 9.0 to 9.2 points where the favourite was overwhelming — so the cost nearly doubles depending on the match you choose.
Does an AI model give an edge in betting?
A model only helps if it finds a gap between its own probability estimate and the one implied by the odds that is larger than the margin. It does not remove the margin, and raising a hit rate is not by itself a profit.