Understanding expected value

Expected value turns uncertain results into a common unit: the average net result implied by a set of probabilities and payoffs. It does not tell you what will happen on the next spin, bet, or promotion. It tells you what the decision is worth under the assumptions.

That last phrase matters. An exact calculation built on a poor win probability or an incomplete reading of the terms is still a poor estimate. Keep the arithmetic simple enough to inspect, and keep assumptions separate from known conditions.

Probability and net outcomes

A probability is a number from 0 to 1 that describes how often an event is expected to occur in the model. A 25% probability is 0.25. The probabilities for all mutually exclusive outcomes must add to 1, or 100%.

A net outcome measures the change in wealth after accounting for money contributed and money received. In a hypothetical example, if a $50 cash wager returns $125 in total, the net outcome is +$75 because the $50 stake was part of the return. If the wager loses, the net outcome is -$50.

List each possible net outcome before calculating expected value:

Expected value = sum of (probability x net outcome)

Losses enter as negative numbers. Returned deposits and stakes should not be mistaken for profit. Promotional stakes that are not returned need separate treatment.

For a sequence, choose a consistent starting point. A promotion can be measured from before the deposit, from after an already planned wager, or from the moment a credit is issued. Those are different questions. State which one the result answers.

Hypothetical sportsbook example: a cash wager

Hypothetical scenario: A bettor considers a $40 cash wager at decimal odds of 2.20. The bettor estimates the true win probability at 48%.

A winning wager returns:

Total return = stake x decimal odds
Total return = $40 x 2.20 = $88
Winning net outcome = $88 - $40 = +$48

The losing net outcome is -$40. The expected value is:

EV = (0.48 x $48) + (0.52 x -$40)
EV = $23.04 - $20.80
EV = +$2.24

The model gives a positive expected value of $2.24 per $40 wager. One wager cannot produce $2.24. It produces either +$48 or -$40. The average becomes meaningful across repeated decisions with similar prices and accurate probabilities.

Return on investment, or ROI, divides expected net value by the capital at risk:

Expected ROI = expected net value / cash at risk
Expected ROI = $2.24 / $40 = 5.6%

This is an expected ROI, not a promised account return. It also depends heavily on the estimated 48% probability.

Decimal and American odds

Decimal odds include the returned stake. The implied break-even probability, before adjusting for a sportsbook’s margin, is:

Break-even probability = 1 / decimal odds

American odds use positive and negative forms. Positive odds show profit on a $100 stake. Negative odds show the stake needed for $100 of profit.

Convert positive American odds to decimal:

Decimal odds = 1 + (American odds / 100)

Convert negative American odds to decimal:

Decimal odds = 1 + (100 / absolute value of American odds)

Convert decimal odds of 2.00 or higher to American:

American odds = (decimal odds - 1) x 100

Convert decimal odds below 2.00 to American:

American odds = -100 / (decimal odds - 1)

Hypothetical conversion scenario: Odds of +150 convert to 2.50:

1 + (150 / 100) = 2.50
Break-even probability = 1 / 2.50 = 40%

Hypothetical conversion scenario: Odds of -200 convert to 1.50:

1 + (100 / 200) = 1.50
Break-even probability = 1 / 1.50 = 66.67%

Rounding can make a converted price differ slightly from the display. More important, implied probability is not automatically true probability. In a two-outcome market, the implied probabilities from both sides can sum to more than 100% because of pricing margin.

Break-even probability with a promotion

For an ordinary cash wager that wins net profit W and loses stake L, break-even probability is:

Break-even probability = L / (W + L)

At decimal odds D, W = stake x (D - 1) and L = stake, which simplifies to 1 / D.

A promotion can change either payoff. A profit boost increases W. Cashback reduces L. A bonus bet removes the promotional stake from the winning return. Recalculate from net outcomes instead of applying the ordinary odds formula without checking.

Hypothetical sportsbook example: a bonus bet

Hypothetical scenario: A customer already has a $50 bonus bet. Its promotional stake will not be returned. It must be placed at decimal odds of 3.00. The customer estimates a 34% win probability. There is no additional wagering requirement on winnings.

The winning cash amount is:

Cash winnings = bonus stake x (decimal odds - 1)
Cash winnings = $50 x (3.00 - 1)
Cash winnings = $100

The losing cash amount is $0 because the customer does not receive or lose $50 of personal cash at this stage. From the point at which the bonus bet already exists:

EV = (0.34 x $100) + (0.66 x $0)
EV = $34

The token’s face value is $50, but the modeled cash value is $34. ROI needs care here. Dividing by a zero cash stake is meaningless. You could report a 68% conversion rate against face value:

Expected conversion rate = $34 / $50 = 68%

That is not the ROI of the full promotion. If the customer had to risk cash to earn the token, include the qualifying wager and use the total cash exposed as the denominator.

The break-even win probability for using an expiring bonus bet is effectively above 0% when the alternatives are use or receive nothing, assuming no other cost or harmful activity. That does not mean any qualifying selection is equally good. The expected conversion still changes with price and true probability, and the original acquisition step may have negative value.

Hypothetical sportsbook example: the full sequence

Hypothetical scenario: A customer must place a $100 cash wager at decimal odds of 2.00 to receive a $50 bonus bet whether the cash wager wins or loses. The estimated win probability for the qualifying wager is 48%. The later bonus bet has the $34 expected cash value calculated in the preceding hypothetical scenario. The customer would not otherwise place the qualifying wager.

First calculate the cash wager:

Winning net outcome = +$100
Losing net outcome = -$100

Qualifying-wager EV = (0.48 x $100) + (0.52 x -$100)
Qualifying-wager EV = $48 - $52
Qualifying-wager EV = -$4

Then add the expected bonus value:

Full-promotion EV = qualifying-wager EV + bonus-bet EV
Full-promotion EV = -$4 + $34
Full-promotion EV = +$30

Expected ROI against the $100 cash initially at risk is 30%. This result assumes the bonus is issued after either settled outcome, is used before expiration, and has no further restriction. If the credit is issued only after a loss, the branches must be modeled separately because the bonus and first result are linked.

Hypothetical casino example: free spins

Hypothetical scenario: A package contains 100 free spins at $0.20 each on a game modeled at 96% return to player. Spin winnings become withdrawable cash with no wagering requirement or cap.

Total promotional stakes are:

Total spin stakes = 100 x $0.20 = $20
Expected gross winnings = $20 x 96% = $19.20

The package has an expected cash value of $19.20 from the point at which it has already been issued. That does not mean the customer will receive $19.20. Game variance can produce no winnings, a small amount, or an amount well above the mean.

Now suppose the same hypothetical package pays winnings as bonus credit with a 10 times wagering requirement. The $19.20 is no longer a valid cash estimate. It is the expected starting restricted balance. Conversion depends on the eligible game’s edge, the chance that the balance survives, stake limits, and any withdrawal cap. The stages should be modeled separately:

Stage 1: expected spin winnings credited as restricted balance
Stage 2: distribution of that balance after required wagering
Stage 3: amount allowed to convert to withdrawable cash

Using only an average can be especially misleading in Stage 2. A restricted balance may reach zero before the required turnover is complete. Simulation can estimate that risk, but a simulation is only as sound as its rules and game model.

Hypothetical casino example: a deposit match

Hypothetical scenario: A customer deposits $100 and receives $100 in bonus credit. The requirement is 5 times the bonus only, so the wagering target is $500. Eligible play has a modeled 2% house edge. Assume for this simplified example that the full $200 balance always survives long enough to complete the target, no cap applies, and the ending balance becomes cash.

Expected gaming loss is:

Expected gaming loss = $500 x 2% = $10
Expected ending cash = $200 - $10 = $190
Net value relative to deposit = $190 - $100 = +$90

Under those strong assumptions, expected promotion value is $90 and expected ROI on the $100 deposit is 90%.

The survival assumption is the weak point. If variance creates a 15% chance of losing the entire $200 before completion, and the other 85% of paths finish with an average $210 before subtracting the original deposit, calculate by complete outcomes:

Failure net outcome = -$100
Completion net outcome = $210 - $100 = +$110

EV = (0.15 x -$100) + (0.85 x $110)
EV = -$15 + $93.50
EV = +$78.50

The revised EV remains positive but is $11.50 lower. A cap on withdrawals would reduce high completion outcomes further. An expiration risk would add another failure or partial-value branch.

Hypothetical casino example: loss-based cashback

Hypothetical scenario: A casino returns 20% of net slot losses for one day, up to $40. The return is withdrawable cash. A customer plans $100 of eligible play regardless of the offer. Without cashback, the customer models three end-of-day net outcomes: a $60 loss with 30% probability, a $20 loss with 40% probability, and a $50 gain with 30% probability.

First calculate the baseline:

Baseline EV = (0.30 x -$60) + (0.40 x -$20) + (0.30 x $50)
Baseline EV = -$18 - $8 + $15
Baseline EV = -$11

Cashback changes only the loss branches:

$60 loss: 20% cashback = $12, revised net outcome = -$48
$20 loss: 20% cashback = $4, revised net outcome = -$16
$50 gain: no net-loss cashback, revised net outcome = +$50

Now recalculate:

Promotion EV = (0.30 x -$48) + (0.40 x -$16) + (0.30 x $50)
Promotion EV = -$14.40 - $6.40 + $15
Promotion EV = -$5.80

Incremental value of cashback = -$5.80 - (-$11) = $5.20

The cashback improves expected value by $5.20 but does not make the planned play positive EV. The $40 cap never applies in these branches because 20% of the largest modeled loss is $12. If cashback were restricted credit rather than cash, its conversion would need another stage. If the customer increased wagering to pursue the $40 maximum, the original outcome probabilities would no longer describe the new plan.

Sensitivity analysis

Sensitivity analysis changes one assumption at a time to show which inputs control the result. It is more honest than presenting a single estimate with false precision.

Return to the hypothetical $40 sportsbook wager at 2.20. Its break-even probability is:

1 / 2.20 = 45.45%

Compare several probability estimates:

At 44%:
EV = (0.44 x $48) + (0.56 x -$40) = -$1.28

At 46%:
EV = (0.46 x $48) + (0.54 x -$40) = +$0.48

At 48%:
EV = (0.48 x $48) + (0.52 x -$40) = +$2.24

A four-point change moves the decision from negative to positive. If the probability estimate cannot reliably distinguish 44% from 48%, classifying the wager as positive EV should carry substantial uncertainty.

For casino offers, vary return to player, wagering requirement, completion probability, and maximum conversion. For sportsbook offers, vary true win probability, available odds, bonus conversion, qualifying-bet result, and expiration risk. Use conservative, base, and favorable cases rather than one preferred case.

The bonus calculator can organize these inputs. Keep a written copy of the terms beside the model so that the calculation reflects the actual playthrough base, eligible wagers, and payout rules.

Positive EV is not guaranteed profit

Positive EV means the probability-weighted net outcomes average above zero under the stated model. It does not mean the next result will be profitable, the bankroll can tolerate the variance, or the estimate is correct. A small positive EV can be overwhelmed by model error, an overlooked term, a missed deadline, payment friction, or a withdrawal cap.

Repeated positive-EV decisions can also produce long losing runs. Independence does not promise that results will alternate neatly. Correlated bets, similar game features, or repeated exposure to the same uncertain estimate can make concentration worse.

Mathematical value is only one approval test. The offer must also be legal and available to the reader, understandable, affordable, operationally manageable, and compatible with personal limits. The National Council on Problem Gambling’s Internet Responsible Gambling Standards are voluntary guidance rather than universal law, and they emphasize player protection tools and informed decision-making. No EV estimate cancels the need for those protections.

The next chapter, bankroll management basics, addresses the question EV leaves open: how much risk, if any, can a person carry while waiting for uncertain outcomes.

Sources

• National Council on Problem Gambling, “Internet Responsible Gambling Standards”. Accessed September 4, 2026.