Skip to content

iGaming · Guide

RTP Explained for Operators

Theoretical vs observed RTP, how far observed RTP drifts on small samples, RTP vs hold, and how to judge an RTP configuration change without fooling yourself.

RTP explained

RTP explained for operators: theoretical return, observed return, and why a week of data proves little

RTP is the most quoted number about a slot and one of the most misread. It is a property of the game’s maths over an enormous number of rounds — not a forecast for a player, a day, or a promotion.

RTP · Games · Variance · Beginner

Theoretical RTP is what a game is designed to pay back over the long run; 96% means 4% theoretical hold. Observed RTP is winnings ÷ turnover for the period you measured, and on small samples it can land far from the design. Judge a game or an RTP change on enough rounds, not on one week’s GGR.

Think of it like this

Flip a fair coin 10 times and getting 7 heads is no surprise. Flip it 10,000 times and you will land very close to half. RTP works the same way: a 96% slot pays back about 96 of every 100 bet — averaged over millions of spins. One week of one game is the 10-flip version.

Theoretical RTP: The coin is fair: half and half by design.

Observed RTP: What you actually got in your flips.

Hold: The house’s share: 100 − 96 = 4 of every 100, in the long run.

Sample size: How many times you flipped.

A strange week is usually the coin, not a broken game. Count the spins before you judge.

RTP explained

Two RTPs

Where it comes from — The game’s maths and certified configuration — Winnings paid ÷ turnover in your data

Changes when — A different configuration is deployed — Every round

Good for — Comparing games, pricing, compliance — Monitoring, spotting real faults over large samples[1]

Misused as — A promise to a player — A verdict after a few days

Hold is the other side: hold = GGR ÷ turnover = 1 − observed RTP. A game with 96% theoretical RTP has about 4% theoretical hold. Jackpot contributions are often counted inside the RTP figure, which matters when you compare a jackpot game with a non-jackpot one.

How far observed RTP can drift (illustrative)

Assume a 96% game where the payout of a single spin has a standard deviation of 5× the stake — a medium-volatility slot. The range below is roughly where observed RTP lands 95% of the time. Higher volatility widens every row.

1,000 — about 64% – 128% — Nothing about the game

100,000 — about 92.8% – 99.2% — Rough sense; still noisy

10,000,000 — about 95.7% – 96.3% — Close to the design

The width shrinks with the square root of the number of spins: a hundred times more data gives a range ten times narrower. A single big win can dominate a small sample, which is why one game’s GGR can swing negative on a quiet day without anything being broken.

Try it: 12 players, one 96% slot: Hover or tab across the chart to read the range at any number of spins, and rerun to see a different set of sessions. The shaded band is the same maths as the table above.

Choosing or changing an RTP configuration

Many games are certified in several RTP versions. A lower setting raises theoretical hold per bet, but players may play less, and some markets set minimum RTPs or require the setting to be disclosed. Any change must be an approved configuration, and the terms shown to players must match it.[2][3]

From the field: When a provider offered RTP adjustment support as a selling point to operators, my first question was how anyone would know it had worked. A higher-hold configuration and a good week look the same in a short GGR report. We looked at sample size, game mix and player mix first, and discussed a small conditional rebate to share the risk while the data built up.

Compare before and after on turnover-weighted hold, not on GGR alone — GGR also moves with volume.

Keep the game mix fixed, or compare per game.

Watch activity too. A higher hold that cuts play time can lower GGR.

Decide the evaluation window before the change, not after seeing the numbers.

Before you trust the number, ask

Is this theoretical RTP or observed RTP?

How many rounds is the observed figure based on?

Is jackpot contribution inside or outside the RTP?

Is the configuration certified and approved for this market?

Do player-facing game rules show the RTP actually deployed?

Was the evaluation window fixed before the change?

What this guide does not cover

The range table assumes a standard deviation of 5× stake per spin and a normal approximation; real games differ, and high-volatility or jackpot games need far more rounds. Minimum RTP and disclosure rules vary by jurisdiction. RTP describes a game, not a way to predict or improve any player’s result.

References

Bracketed numbers in the text point to the primary sources below. Field notes come from my own work; rules differ by jurisdiction, and these are UK and industry-standard references.

[1] UK Gambling Commission, “Testing strategy for compliance with RTS — Live RTP monitoring,” 2023. [Online]. Available: https://www.gamblingcommission.gov.uk/manual/testing-strategy-for-compliance-with-remote-gambling-and-software-technical/5-live-rtp-monitoring [Accessed: 10 Nov. 2026]. Licensees compare actual RTP with the advertised figure, for under- and overpayment, at a frequency set by volume of play.

[2] UK Gambling Commission, “Remote gambling and software technical standards — RTS 3: Rules, game descriptions and the likelihood of winning,” 2024. [Online]. Available: https://www.gamblingcommission.gov.uk/standards/remote-gambling-and-software-technical-standards/rts-3-rules-game-descriptions-and-the-likelihood-of-winning [Accessed: 10 Nov. 2026]. Game information must show players how the game works and its house edge or return to player (RTP).

[3] Gaming Laboratories International (GLI), “GLI-19: Standards for Interactive Gaming Systems, v3.0,” 2020. [Online]. Available: https://gaminglabs.com/wp-content/uploads/2024/06/GLI-19-Interactive-Gaming-Systems-v3.0.pdf [Accessed: 10 Nov. 2026]. The test-lab standard games are certified against, including payout percentages and RTP display (section 4.7).