RTP is the most quoted and least useful number in gambling
Return to player is the percentage of total stakes a game returns to players over an effectively infinite number of rounds. A 96% game returns 96 units for every 100 staked, and the house edge is the complement: 4%.
It is quoted everywhere, it is disclosed by regulation in many markets (in Britain the technical standards require online games to make available the return to player, the house edge or the probability of winning before the customer gambles), and on its own it tells you almost nothing about what will happen to a player or to an operator. Two games with identical RTP can differ by a factor of thirty in how much money they extract per hour, and many times over in how wide the range of session outcomes is.
Four numbers together describe a game. RTP is one of them, and it is the one that matters least for most practical questions.
Number one: return to player, and what it actually promises
RTP is a long-run expectation computed from the game's own mathematics. Testing standards draw the same line: GLI-11 defines a theoretical return, based on mathematical calculations or simulations, and an actual return, based on the metering of a fielded machine. It is not a promise about a session, a day, a month or a player. It is the value the average converges on as the number of rounds tends to infinity.
The computation is straightforward in principle. For every possible outcome, multiply its probability by its payout, sum the results, and divide by the stake.
Take a trivial illustrative game: a single spin costing 1 unit, with a 5% chance of paying 10, a 20% chance of paying 2, and a 75% chance of paying nothing.
- Expected return = (0.05 x 10) + (0.20 x 2) + (0.75 x 0) = 0.50 + 0.40 = 0.90
- RTP = 0.90 / 1.00 = 90%
- House edge = 10%
Real slot models are the same arithmetic over far larger outcome spaces, sometimes hundreds of millions of combinations rather than three: one five-reel video slot studied by researchers had 259,440,000 possible reel combinations, which they analysed with a computer program. That is why maths models are computed by software and checked by simulation rather than by hand.
Two properties of RTP are worth holding onto. It is a property of the game, not of the player or the session, and not visible from the screen: the same study found identical-looking versions of one slot approved in Ontario with payback percentages from 85% to 96.2%. And it says nothing about how the 90% is distributed: a game returning 90% as a steady drip and a game returning 90% almost entirely through a one-in-fifty-thousand jackpot have identical RTP and are completely different products.
Number two: house edge and its relationship to hold
House edge is 100% minus RTP, and the two are used interchangeably in casino contexts. In sports betting the analogous concept is margin or overround, and in operational reporting the relevant figure is hold percentage, which is actual win divided by amount wagered over a real period.
The distinction matters because hold and house edge are not the same thing.
House edge is theoretical and fixed by the game. Hold is measured and varies with what actually happened. Over a very large number of rounds hold converges on the house edge; over a day on a single machine it can be anything, including negative. Check the denominator, too: land-based table games are measured differently, and regulators such as Nevada's measure table hold as win against drop, the cash and credit exchanged for chips, and a table hold percentage is not directly comparable with a house edge.
An operator reporting a hold percentage below the theoretical edge has not been robbed and has not had a broken game. It has had a period where players won more than expectation, which happens. The question a maths-literate operator asks is whether the deviation is within the range the variance of that game predicts, and answering it requires the third number. Nevada builds a version of this into its internal controls: management must investigate monthly table-game fluctuations beyond a set tolerance from the property's own baseline hold.
Number three: volatility
Volatility, or variance, describes how widely actual results spread around the expectation.
A low-volatility game pays small amounts frequently. Sessions cluster near the expected loss, players last a long time on a given bankroll, and the experience is steady. A high-volatility game pays rarely and largely. Most sessions lose more than the expected amount, a small number win enormously, and the distribution has a long right tail.
The formal measure is the standard deviation of the return per unit staked. Rather than the raw figure, a manufacturer's par sheet may give a volatility index, an indication of how much the payback percentage will vary for a given number of games played.
The practical consequence is that volatility, not RTP, determines what a player experiences. Returning to the earlier point: two 96% games, one low volatility and one high, produce completely different session outcome distributions, completely different bankroll requirements, and completely different appeal. A player who dislikes one will frequently dislike the whole category, which is why portfolio design is a volatility question before it is an RTP question.
Volatility also determines how long it takes actual results to converge on theory, which is the operator's problem rather than the player's. A high-volatility game on a small number of positions can run above or below its theoretical hold for a very long time, and an operator that reallocates floor space or renegotiates a supplier deal on a month of data from a high-variance title is reading noise.
Number four: speed, the one everybody forgets
Expected loss is not determined by house edge. It is determined by house edge multiplied by the amount wagered, and the amount wagered is determined by stake multiplied by the number of rounds played.
This is the single most important practical result in gambling mathematics, and it is worth stating as a formula:
Expected loss per hour = average stake x rounds per hour x house edge
Consider two games, with round illustrative figures.
Game A: 1% house edge, 600 rounds per hour, 1 unit stake. Expected loss per hour = 1 x 600 x 0.01 = 6 units.
Game B: 10% house edge, 20 rounds per hour, 1 unit stake. Expected loss per hour = 1 x 20 x 0.10 = 2 units.
Game B has ten times the house edge and costs a third as much to play for an hour. Anyone comparing those two games on RTP alone reaches exactly the wrong conclusion.
This is why event frequency dominates product risk discussions, why a fast low-edge game can out-earn a slow high-edge one, and why regulators intervene on spin speed and autoplay rather than only on return to player. In Britain, an online slot game cycle must last at least 2.5 seconds before the next can start, and other online casino games at least 5 seconds, which caps an online slot at 1,440 spins an hour, and auto-play is banned for online gaming. For scale, researchers timing land-based machines found about 600 spins an hour on mechanical reel slots and about 1,200 on video slots. It is also why a casino floor is laid out the way it is: revenue per position per hour, not house edge, is the operational number.
Hit frequency, and why it is not volatility
Hit frequency is the proportion of rounds producing any win at all. A game with 25% hit frequency pays something on one spin in four.
It is related to volatility but distinct from it, and conflating the two is a common error. A game can pay very often in tiny amounts and still be high volatility if a rare outcome dominates the return. Conversely a game with modest hit frequency and a tight payout distribution is low volatility.
Hit frequency matters commercially because it governs the pace of feedback, which is a large part of how a game feels. It also interacts with a design pattern worth naming: a win that returns less than the stake still counts as a hit. A game can report a high hit frequency while the majority of those hits are net losses, which is the losses-disguised-as-wins effect: in multi-line video slots most of the "wins" are less than the spin wager, and in a laboratory study novice players' arousal responses to them were similar to their responses to real wins. Britain has since banned celebrating a return that is less than or equal to the stake, part of a package of online slots rules announced in 2021. If you are assessing a game, ask for the hit frequency above stake as well as the headline figure, because the two can be very far apart.
Putting the four together
A game specification worth reading gives you all four, and you can reason about the product from them. The three profiles below are illustrative.
- RTP 96%, volatility low, hit frequency 30%, 600 spins per hour. A grinding product. Steady small wins, long sessions on a modest bankroll, revenue accumulated through volume. Expected cost to the player of 24 units per hour at 1 unit stake.
- RTP 96%, volatility very high, hit frequency 18%, 600 spins per hour. Same expected cost per hour, radically different experience: long dry runs, occasional very large outcomes, wide dispersion of session results, and a much larger bankroll required to reach the long run.
- RTP 98%, volatility low, hit frequency 45%, 40 rounds per hour. A table-game profile. Very low cost per hour and very low revenue per position, which is why floor space allocation is not decided on house edge.
Every subsequent lesson in this course builds on these four. The next one shows where they come from: the reel strips, symbol frequencies and paytable that constitute a slot's maths model, and how a target RTP is actually hit.