96% is not a promise: what RTP tells you about a game, and what it does not

Two games display the same label: RTP 96%. In one, small payouts appear on the screen frequently. In the other, little may happen for a while, while the promise of a rare, large outcome attracts attention. Are they similar mathematically? In one respect they can be. From an individual’s perspective, they can look completely different.

That difference is a useful starting point for discussing RTP. The percentage in a game description looks straightforward, but it is easy to attach promises it never makes. Understanding it requires separating an average, a distribution of outcomes and the experience of a short session. This is a story about game design and randomness that is useful beyond casinos too.

The percentage concerns stakes, not a deposit

RTP describes payouts as a share of total stakes. Theoretical RTP comes from game design; actual RTP from observations. Gambling Commission defines these terms.

Imagine a purely mathematical model: the stake is one unit and the expected payout is 0.96 units. Theoretical RTP is then 96%. For a predetermined set of 1,000 such rounds, expected total payouts are 960 units. The expected difference is 40 units. None of these numbers determines an individual’s outcome.

The word “stakes” matters. An account deposit and the sum of all wagers are different quantities. The same funds can be used in successive rounds. The percentage therefore cannot simply be multiplied by the initial deposit and interpreted as the amount received at the end. In this model, we count the units wagered in successive rounds, rather than transfers into an account.

One average, two very different worlds

Let us construct two imaginary games solely as probability examples. Each round costs one unit. Game A pays 1.2 units with an 80% probability and zero otherwise. Its average payout is 0.8 × 1.2 = 0.96 units.

Game B pays 96 units with a 1% probability and zero otherwise. Its average is 0.01 × 96 = 0.96 units. Both therefore have theoretical RTP of 96%, although they distribute payouts radically differently. These are educational models, not descriptions of particular products.

In the first model, a payout event is frequent. In the second, the vast majority of rounds produce no payout. RTP alone does not reveal that difference. Expected value compresses a distribution into one number, which is precisely why it cannot replace the complete description. A similar problem arises when we try to describe a whole group of people using only their average income.

Volatility describes the spread

Volatility describes the spread of outcomes: large, rare payouts or smaller, frequent ones. Gambling Commission identifies standard deviation as a commonly used measure.

Our A and B models make it intuitively clear why the average is insufficient. Game B’s expected payout relies on a rare event with a large value. A short observation may not encounter it at all. Game A’s outcomes are closer together, although still random. Neither model has a positive expected value after the stake is deducted.

The label “high volatility” is not a complete probability distribution either. Two games described that way may differ in their details. The reader still needs the rules, paytable and feature information. A label is a shorthand, and a shorthand always leaves out part of the picture.

A payout does not always mean a profit

Another misunderstanding lies in the word “win”. Suppose a round costs one unit, and an animation accompanies a payout of 0.4 units. The game history records an event with a payout. The balance after that round is nevertheless 0.6 units lower.

The frequency of any payouts and the probability of an outcome exceeding the stake therefore answer different questions. The chance of ending an entire session with a positive balance is another question again. RTP does not directly provide any of those frequencies. Sounds, colours and a prize counter do not change the arithmetic.

This is particularly interesting from an interface design perspective. The screen can emphasise an event, while the balance shows its financial effect. Understanding the game requires noticing both. An impressive presentation does not add another unit to a payout.

A losing streak does not create a debt owed by the game

In a simple model of independent rounds, earlier outcomes do not increase the probability of the next payout. The game has no obligation to “give back” someone’s money after a set number of unsuccessful attempts. An average describes a model, rather than a personal compensation schedule.

In Great Britain’s remote gambling standards, RTS 7 prohibits adaptive behaviour that changes probabilities during play. Permitted bonus features have their own stated rules. This specific regulation should not be extended to every physical gaming machine and every country.

The distinction matters within the product as well. If a game includes an explicitly described state, such as collecting symbols for a feature, the rules governing that state need to be read. That is a different issue from believing that an ordinary losing streak itself forces a random mechanism to balance the result.

Observing a short session has limitations

Actual RTP for a particular set of rounds can be calculated by dividing payouts by stakes and multiplying by 100%. In our own example, payouts of 730 units against stakes of 1,000 units give 73%. This describes that sample. It does not itself change the model’s theoretical RTP or predict the next session.

The number of observations and the spread of possible outcomes matter. A rare event with a large payout can strongly influence a reading. A few dozen rounds therefore do not reliably describe the entire distribution. More rounds do not guarantee a particular result either, and they expose funds to further losses. The statistical concept of a long run should not be treated as encouragement to continue playing.

What a demo version can actually show

A demonstration mode allows the controls, rules and presentation of features to be explored. A successful short trial does not forecast the outcome of real-money play. A random sample remains a random sample even when the balance is virtual.

Great Britain’s RTS 6 requires an operator’s free version to reflect the rules, probabilities and prize distribution of its corresponding real-money game. The standard also distinguishes specifically labelled supplier demonstrations. This is a requirement of a particular supervisory system, rather than proof that any demo found online complies.

The most important information is often outside the main animation

Game help and the paytable may be less spectacular than the main screen, but they are where the meaning of symbols, feature conditions and product parameters should be sought. Great Britain’s RTS 3 emphasises accessible, understandable rules and information about chances before play begins.

The most interesting question about RTP is therefore: which part of reality does this percentage summarise? The answer helps separate mathematics from intuition. Average, volatility, payout frequency and session balance provide different descriptions of the same mechanism. Recognising those differences turns the apparently simple label “96%” into information we actually understand.