Sharp bettors don't pick winners. They find mispriced odds. The single biggest mindset shift between a recreational bettor and a profitable one isn't about which teams to bet — it's about understanding positive expected value (+EV) and applying it to every wager. This guide covers the +EV formula, three worked examples, the variance trap, and why a 48% bet can beat a 60% one.
Ask a casual bettor why they took the Brewers tonight and you'll get a story: the Brewers are hot, the starter is good, the matchup favors them. Ask a sharp bettor the same question and you'll get a number: the model has the Brewers at 48.3%, the market is paying like they're 41.8%, and the gap is 6.5 percentage points of edge. The casual answer is about the team. The sharp answer is about the price. Whether the Brewers actually win any single game is almost irrelevant to whether the bet was correct.
That sounds backwards. Of course it matters whether the bet wins — the dollars only move if the Brewers cover. But what doesn't move with the dollars is whether the decision was right. The decision is right when the price you paid is too generous relative to the true probability. Repeated across hundreds of bets, the math sorts the rest out for you. Repeated without the math, the math sorts you out instead.
When a sportsbook posts the Brewers at +139, that price isn't a prediction. It's an offer. It says: "We will pay you $1.39 for every $1 you risk if the Brewers win." Embedded in that offer is an implied probability — the win rate at which the bet breaks even over the long run. For positive American odds the math is simple:
Implied probability = 100 / (odds + 100)
So +139 implies a 100 / (139 + 100) = 41.8% break-even win rate. For negative American odds the formula flips:
Implied probability = |odds| / (|odds| + 100)
So -150 implies 150 / 250 = 60% break-even. If you bet a team at -150 and you think they'll win 60% of the time, you've taken a coin flip in dollar terms — you'll break even forever. If you think they'll win 65%, you've taken a +EV bet — over enough trials you make money. If you think they'll win 55%, you've taken a -EV bet — over enough trials you give money to the book, no matter how often you win.
This is the whole game. Find the gap between your estimate of true probability and the market's implied probability, bet the side where your number is higher, and the long run rewards you. The team doesn't have to win the game. The price just has to be wrong.
Once you know your win probability and the price you're getting, expected value falls out of two numbers and a subtraction. The general formula:
EV = (Pwin × Profit) − (Ploss × Stake)
Where Profit is what the bet pays you per dollar risked if it wins (the decimal-odds-minus-1 number), and Ploss is simply 1 − Pwin. If EV is positive, the bet is profitable in the long run. If it's negative, you're paying the book to entertain you.
For a $1 bet at +120, you risk $1 to win $1.20. So if your win probability is 50%: EV = (0.50 × $1.20) − (0.50 × $1) = $0.60 − $0.50 = +$0.10 per dollar. The bet returns ten cents on the dollar over an infinite series of identical bets. That's the headline number sharp bettors care about. Want the math without doing it by hand? Plug your numbers into our EV Calculator and it returns EV in dollars, EV as a percentage of stake, and the edge in percentage points.
The formula has a quietly important property: at the price where your implied probability exactly equals the market's, EV is zero. That's the break-even price. Bet above it (i.e., get a better price than fair) and you're +EV. Bet below it and you're −EV. Calculating that break-even price is how you decide whether a posted line is worth taking or worth passing.
On May 19, 2026, our MLB model published Chicago White Sox +128 against Seattle as the official play of the day. The model's win probability for Chicago was 48.3%. The market's implied probability at +128 was 100 / 228 = 43.9%. The gap — what we'd call the edge — was 4.4 percentage points.
Model win probability: 48.3%
Market price: +128 → implied 43.9%
Edge: +4.4 percentage points
Fair-value price at 48.3%: (100 / 0.483) − 100 = +107
+EV window: anywhere from +107 up to +128 (the current price)
Expected value per $1 risked: (0.483 × 1.28) − (0.517 × 1.00) = +$0.10
Two things matter about that table. First, the +EV window is a range, not a single price. If the line moves from +128 down to +120, the bet still has edge (smaller, but real). If it moves down to +107, the edge is exactly zero — you're paying fair value, with no profit and no loss in expectation. If it moves below +107, the bet has crossed into negative expected value and you should pass.
Second — and this is the part that's hard for recreational bettors to internalize — the bet was correct at the moment of placement, regardless of whether Chicago actually won the game. If they lose 8-2, the +4.4pp of edge was still real when we placed the bet. The dollars went the wrong way that night. But the decision was right, and the math doesn't care about a single night. It cares about whether you systematically take 48.3% probabilities at prices that imply 43.9%, because over a few hundred of those, the variance washes out and the edge lands as profit.
The White Sox example was a dog. The pattern works identically on favorites and across sports — the only thing that changes is the sign convention on American odds. Two more cases from real slates:
Our NHL model put the Carolina Hurricanes at 62.5% to beat the Washington Capitals on a Tuesday in March. The market hung Carolina at −145 across most books, which implies 145 / 245 = 59.2%. Edge: +3.3 percentage points. Running the EV formula:
Model win probability: 62.5%
Market price: −145 → profit of $0.69 on each $1 risked, breakeven at 59.2%
EV per $1: (0.625 × $0.69) − (0.375 × $1.00) = $0.431 − $0.375 = +$0.056
Translation: +5.6% expected return per dollar risked. On a $200 stake, that's $11.20 of expected value built into the bet at the moment of placement.
A 3.3-point edge on a hockey favorite is meaningful but not extreme. Carolina lost that night 4–3 in overtime. The bet was correct. The variance was not. Across 100 similar setups, this kind of edge produces roughly $560 of expected profit on $20,000 risked — modest in any single instance, decisive at scale.
Player props are where mispricings live longest because books don't move the lines as aggressively as they move game totals. Suppose a book offers an NBA scoring prop at Player X over 22.5 points +110. Implied probability: 100 / 210 = 47.6%. Your model (factoring in projected pace, recent shot volume, opposing defensive rating, and a teammate sitting out) lands the player at 53% to go over.
Model probability: 53.0%
Market price: +110 → profit of $1.10 per $1 risked, breakeven at 47.6%
EV per $1: (0.530 × $1.10) − (0.470 × $1.00) = $0.583 − $0.470 = +$0.113
Translation: 11.3% expected return per dollar — substantially fatter than the typical moneyline edge, because the prop market has wider hold and slower line response. Sharp bettors hunt props for exactly this reason.
Three examples, three sports, three price ranges — the EV math runs identically. Plug any of these into the EV Calculator and you'll get the same answers. The hard part isn't the formula. The hard part is producing the probability estimate that goes into it.
If the betting market were perfectly efficient, no mispricings would exist. Sportsbooks would price every game exactly at true probability minus their vig, and no one could beat them. The market isn't efficient, for several reasons that compound:
None of these are mysteries. They're documented, repeatable structural features of the betting market. A model that quantifies any one of them on a sport-specific basis produces a meaningful probability gap on a meaningful fraction of games. The job isn't to discover hidden information. The job is to estimate true probability slightly better than the consensus market, on the dimensions where the consensus is weakest.
Even when you find a +EV bet, the price you pay determines how much of that edge ends up in your bankroll. Suppose your model says the Brewers should win 48.3% of the time, and you have three sportsbooks open in your browser. One has Brewers +132, one has +128, one has +124. They all imply roughly 43% probability — they all look like the same bet — but the EV per dollar at each is meaningfully different:
That's a 47% swing in expected profit between the best line and the worst line on the same bet. Stretch it across an MLB season — say 250 bets at $100 risked each — and the difference between always shopping for the best price and always taking the first line offered is roughly $1,000 of expected profit per year on the same exact bets.
This is why every serious bettor we know maintains active accounts at three to five books. The few minutes it takes to check three prices before placing is the highest-ROI work in the entire workflow. The model finds the bet; line shopping decides how much you keep.
This is the part of +EV betting that breaks more bettors than any modeling mistake. The math says you make money in the long run. The short run is a different animal entirely.
Consider a bettor with a real, durable 4% edge — meaning their estimates beat the market's implied probability by an average of 4 percentage points. Over 1,000 bets at $100 each, that bettor's expected profit is around $4,000. But the standard deviation on that 1,000-bet run is roughly $3,200. Translation: in any given 1,000-bet sample, an outcome anywhere from +$800 to +$7,200 is well within one standard deviation of expected. Run that same edge for only 100 bets and the standard deviation is large enough that you can easily be losing money — with real edge.
The implication is uncomfortable: you cannot tell whether your edge is real from a hundred-bet sample. Variance dominates that horizon. Most bettors who quit do so 50–200 bets into a winning strategy, because they extrapolate from a losing run that's mathematically routine. The bettors who survive past that variance window are the ones who size correctly (see Kelly Criterion) and who track CLV instead of P/L — because CLV converges to truth faster than dollars do.
The honest answer to "am I a winning bettor?" is "I don't know from this sample." The answer becomes statistically defensible somewhere between 500 and 1,000 graded bets. Below that, the only thing that tells you whether the process is right is whether you systematically beat the closing line.
The math we've described — finding gaps between fair value and market price — produces a side effect that's easy to measure and impossible to fake. If your fair-value estimates are systematically better than the market's, you'll consistently get prices that the market later regards as too generous. Which means you'll consistently beat the closing line.
That's closing line value, and we wrote a separate piece on why it's the only sports betting stat that converges on a real answer over a couple hundred bets. CLV is the downstream evidence. The +EV mindset is the upstream cause. A bettor who has the discipline to estimate fair value and only bet when the price is generously above it will, over a few hundred bets, post positive CLV. A bettor who's picking winners by gut feel won't, no matter how often they win in the short run.
The reason CLV is the metric that matters is precisely the reason this article matters. Winning and losing individual bets is variance. The price you take vs. the price the market closes at is a measurement of process. CLV is the truth, +EV is the strategy, and one produces the other.
Here's the math that does the most damage to recreational intuition. Consider two strategies:
Strategy A: Bet 60% favorites at -150. You win 60% of the time. EV per $1 risked = (0.60 × 0.667) − (0.40 × 1.00) = $0.00. You break even forever.
Strategy B: Bet 48% underdogs at +120. You win 48% of the time. EV per $1 risked = (0.48 × 1.20) − (0.52 × 1.00) = $0.056. You make 5.6 cents per dollar risked over the long run.
Strategy A is a more accurate "picker" — it wins 60% of its games. Strategy B is statistically a "loser" — it loses 52% of its games. But Strategy B is the profitable one. The 60%-winner is paying so much juice that the win rate exactly cancels the price. The 48%-loser is getting paid such a generous price that the edge accumulates faster than the losses subtract.
If you walked into a card game with two seats open — one where you'd win 60% of hands but lose money, and one where you'd win 48% of hands but profit — you'd take the second seat every time. Sports betting works exactly the same way. The seat is the price.
Every model we run — the MLB starting pitcher leverage system, the NBA pace-adjusted efficiency model, the NHL goalie workload edge — produces the same shape of output: a probability estimate for each side of each game. Those estimates are then compared against the public market price at the moment of publication, the implied probability is computed, and the gap is reported as edge. A play only becomes an official pick when the edge clears our minimum threshold (with a buffer baked in for vig and model uncertainty).
What gets published is the version of the bet where the math says you have a real, measurable advantage. What doesn't get published — and we kill more candidates than we publish — is every game where the model agrees with the market, or where the edge isn't large enough to survive ordinary line movement and noise. That's not a limitation. That's the discipline. Every play we ship is a price we believe is wrong, sized to capture as much of that wrongness as the math allows. The rest, you can read about in the bankroll management piece.
If you're already running this kind of process — comparing your fair-value estimates against the market, sizing by edge, shopping lines, tracking CLV — you don't need us. If you're not, or if you want a quantitative second opinion on every slate without doing the modeling yourself, that's what we're for.
+EV (positive expected value) means a bet's true win probability, multiplied by what it pays, exceeds the loss probability multiplied by the stake. Over a large enough sample, +EV bets are profitable; −EV bets aren't. The math: EV = (win probability × profit on a win) − (loss probability × amount risked). If that number is positive, the bet is +EV regardless of whether it wins or loses individually.
Multiply your estimated win probability by the profit on a winning bet, then subtract the loss probability times the amount risked. For a $100 bet at +120 with a 50% win probability: (0.50 × $120) − (0.50 × $100) = +$10 EV per bet. The bet is profitable in the long run by $10 per $100 risked. Our EV Calculator runs this for any inputs.
Yes — if your win-probability estimates are systematically more accurate than the market's. The math guarantees long-run profit; the practical constraint is that estimating true probability is hard, books take a cut (vig), and variance can produce losing stretches of 50–200 bets even with real edge. Most bettors quit during variance, mistaking it for the strategy being broken.
Mathematically, any positive edge is +EV. Practically, sharp bettors typically require at least 2–3 percentage points of edge above the market's implied probability before placing. That buffer absorbs estimation noise, line movement, and the book's vig. Below 2pp, the bet is technically +EV but the margin is too thin to survive real-world friction.
Absolutely — and most +EV bets that lose individually are still correctly placed. A 55% bet at +100 has +5% EV and a 45% chance of losing each time. Across 100 such bets you'd expect roughly 55 wins, but any individual bet flips a near-coin. Judging a bet by whether it won is the most common mistake in sports betting. Judge it by whether the price was wrong at the moment of placement.
You need some way to estimate true win probability more accurately than the market on at least some games. That can be a quantitative model, a deep edge in one specific niche (one team you follow obsessively, one matchup type), or the de-vigged consensus from sharp books used against soft retail prices. Without any of those, you're guessing — and guessing doesn't produce systematic +EV.