The average profit or loss per bet given the probability of winning and the odds offered.
Expected value is the bedrock of any quantitative betting program. The formula is one line: probability of winning, times what you win if you do, minus probability of losing, times what you lose if you don't. Everything else — Kelly sizing, CLV, hold percentage, bankroll math — is downstream of EV.
The intuition is best worked out at a 50/50 coin flip. If a book offers you +110 on heads (decimal 2.10), your EV per dollar staked is 0.5 × 1.10 − 0.5 × 1.0 = +5 cents. Bet a thousand times and you expect to make $50, give or take variance. The opposite side at -110 has EV of -5 cents per dollar — the book's vig.
Real betting is not a coin flip. The two probabilities are estimates, not facts. The closer your estimate to the truth, the closer your realized return tracks your computed EV. Models that overstate edge run hot for a while and then collapse; models that earn their edge are vindicated by long-run convergence between EV and actual profit.
EV is the metric we live and die by. A pick can win and still have been a bad bet — if the model thought it was 60% and the closing line said it was 65%, the bet had negative EV at the time it was placed, and the win was luck. A pick can lose and have been correct — positive EV bets lose all the time. Tracking EV alongside record is how a serious operation separates skill from variance.
Every published pick is gated by minimum EV. Below the floor, we don't publish even if the model favors the side. EV — not W/L — is the metric we tune the pipeline against.