Key takeaways
- Odds of 1.35 imply a 74.1% probability, so a 68% estimate sits below the break-even line.
- A bet has value only when your estimated probability exceeds 1/odds, not simply when it is high.
- The minimum odds for a 68% estimate are about 1.47; anything lower carries negative expected value.
- A short sample proves nothing, so track closing odds and sample size as well as results.
The short answer: the price tells you how much chance you need
A 68% probability is not enough because odds of 1.35 require at least 74.1%. What matters is the gap between your estimate and the price, not how large the number looks on its own.
A high chance and a good bet are different things. A favourite that wins seven times in ten can still be a poor bet if the market charges too much for it. Intuition often misleads here: 68% sounds like "very likely", so the bet feels safe. But safety is not the question. The question is whether the price matches the chance. The basics are covered in our guide to probability, odds and EV, which also has a calculator.
Implied probability: what the odds actually claim
Implied probability is 1 divided by the odds. At 1.35 that is 1 / 1.35 = 0.74074, or 74.07%. In other words, the bet only breaks even if the event happens about 74 times in 100. The odds are effectively a price tag: they state how much chance the market charges for a given payout.
The odds also contain the bookmaker's margin. Implied probabilities across all outcomes add up to more than 100%, so the raw figure is slightly inflated. That is why Slyx removes each bookmaker's margin (a power de-vig) and works with a fair price rather than the raw quote. Details are on the methodology page.
How much margin hides in the odds
An illustrative example: a two-outcome market priced at 1.35 and 3.20. The implied probabilities are 1 / 1.35 = 74.07% and 1 / 3.20 = 31.25%. They sum to 105.32%, so the margin is roughly 5.3 percentage points. The simplest adjustment is to divide by the total: 74.07 / 105.32 = 70.33% and 31.25 / 105.32 = 29.67%. Slyx uses a different, power method that spreads the margin slightly differently, so the figures differ a little, but the principle is the same. The example teaches something important: even the fair price of this favourite would be around 70%, still above a 68% estimate. So compare your estimate with the fair price, not with the quoted odds, and only then with what you can actually get.
A worked example, step by step
The numbers below are purely illustrative (modelled); they are not a real pick or measured performance.
Setup
You estimate p = 0.68 and the offered odds are 1.35.
Arithmetic
- 1Implied probability: 1 / 1.35 = 74.074%.
- 2Edge in percentage points: 100 × (0.68 − 0.74074) = −6.074 pp.
- 3Expected value: 0.68 × 1.35 − 1 = 0.918 − 1 = −0.082, or −8.2% per unit staked.
- 4Kelly criterion: max(0, (0.68 × 1.35 − 1) / (1.35 − 1)) = max(0, −0.082 / 0.35) = 0. Kelly says do not bet.
- 5Minimum odds for a 68% estimate: 1 / 0.68 = 1.4706. Only from roughly 1.47 upwards would expected value be zero or positive.
Result: even if the favourite really wins 68% of the time, you would lose about 8.2% of each unit staked over the long run. This calculation applies to binary markets where win or lose are the only outcomes; markets with refunds or split settlement need different treatment.
A second illustrative example: same odds, different estimate
Now suppose (again purely illustratively) that your estimate is p = 0.80 at the same odds of 1.35. Expected value is 0.80 × 1.35 − 1 = 1.08 − 1 = +0.08, or +8% per unit. The edge is 100 × (0.80 − 0.74074) = +5.926 pp. Full Kelly would be 0.08 / 0.35 = 0.2286, more than 22% of a bankroll, which is absurdly aggressive in practice because every estimate carries error. It does show the point, though: the odds stayed the same, only the estimate changed, and a losing bet became a good one. So the question is never "are 1.35 good odds?" but "what is the real chance?".
Why low odds leave so little room for error
At odds of 1.35 a winning bet earns only 0.35 of a unit, while a losing bet costs the whole unit. One loss therefore wipes out almost three wins (1 / 0.35 ≈ 2.86). That is exactly why the break-even point sits so high, at 74.07%.
The second consequence is sensitivity. Expected value changes by the odds times the change in probability, so at 1.35 an error of one percentage point in your estimate moves EV by roughly 1.35%. If you want a cushion, say EV of at least 1%, the estimate must be at least 1.01 / 1.35 = 74.8%. For a 3% cushion it is 1.03 / 1.35 = 76.3%. A 68% estimate is far from both, and a small improvement in the model would not change that.
The third consequence is psychological. A run of wins at low odds feels calm, but a single losing streak can erase a long stretch of profit. That is why sample size matters and why feeling is a poor guide.
How Slyx handles this
Slyx covers moneyline and totals on .5 lines in the NHL, MLB and MLS. The fair probability is a weighted median across bookmakers; sharp books (Pinnacle, Betfair, Matchbook, Smarkets, Circa) weigh 3x, and a market needs at least two complete bookmakers. The final probability blends the market fair price with a Poisson model of recent form; the model weight is 0.25 in the regular season, 0.10 in preseason and 0 when no model is available. The market is the anchor.
The "play from" minimum odds come from that blended probability: 1 / probability, exactly the logic in the example above. A pick must also clear an edge gate: expected value of at least 1% when a sharp book is present and at least 3% without one. Picks with EV ≤ 0 are dropped. If fewer qualify, fewer are published; the limit is never filled artificially. See the analysis card article for how a card looks.
The model does not include confirmed lineups or individual starting-pitcher models. Before publication an AI research step therefore checks injuries and news with cited sources; its verdict can only keep, downgrade or veto a pick and never changes the market, odds or probability.
What this does not tell you: the limits of an estimate
Everything above depends on the estimate p being right. That is the weakest link in any analysis. The difference between 68% and 72% can come from a single piece of information the model does not have.
A few honest caveats:
- A short sample proves nothing. In the last 30 days (to 15 October 2026) Slyx had 150 settled picks, a 54.7% hit rate and a flat-stake ROI of 16%. In the last 90 days, however, there were 295 picks, a 50.5% hit rate and an ROI of only 1.3%. MLB alone had 229 picks and an ROI of −7% over 90 days. Slyx has not demonstrated independent profitability, and these figures are not proof of it.
- Closing odds are a better yardstick than results. Average CLV was 1.1% over 160 picks with a closing price in the last 90 days; that is also a small sample.
- One match result says nothing about the quality of an estimate. A 68% estimate is supposed to be wrong about a third of the time.
- The examples in this article are binary and simplified. Real markets settle in different ways, and a real estimate is always uncertain.
The quality of probabilities can be measured with, for example, the Brier score, (p − outcome)². Slyx stores predictions before kickoff and evaluates them on later data.
A checklist before any bet
- 1Compute 1 / odds and compare it with your own estimate.
- 2Compute EV = p × odds − 1. If it is negative, stop.
- 3Find the minimum odds 1 / p and compare it with several bookmakers' prices.
- 4Ask whether your estimate rests only on form and ignores lineups or injuries.
- 5Record losses too, and note whether the price moves for or against you after your call.
- 6Set a limit in advance; see responsible betting.
Results for every published pick, losses included, are public in the history and on the results page. The sports analysis checklist covers further ways to check an analysis.
Reader questions
Does negative expected value mean the favourite will lose?
No. Negative EV only says the price is too low for the chance. A favourite with a 68% probability wins most matches, but at odds of 1.35 the long-run loss would be about 8.2% per unit staked. It is an average over many repetitions, not a forecast for one match.
How do I find the minimum odds for my estimate?
Divide one by your probability estimate. At 68% that is 1 / 0.68 = 1.47. Lower odds mean negative EV. Slyx shows this threshold as "play from", calculated from the blended market and model probability. Remember the result is only as good as the estimate behind it.
Why not just compare my estimate with one bookmaker's odds?
A single quote contains margin and may sit slightly off the market. You get a better fair price by removing the margin and comparing several bookmakers, especially sharp ones. Slyx uses a weighted median and requires at least two complete bookmakers. You can try the same steps yourself in the calculator in the learning section.
Sources and methodology
- Wikipedia: Kelly criterion
- Wikipedia: Brier score
- Slyx methodology and model limitations
- Complete Slyx results including losses
Responsible use
For adults 18+ only. This is educational content. Outcomes are uncertain; gambling can cause financial loss. Set time and money limits beforehand and never chase losses.