Slot hold percentage is easy to define and easy to misuse. In the mathematical sense, theoretical slot hold is the casino-side complement of theoretical RTP. If an approved game configuration has a theoretical RTP of 92%, its theoretical hold is 8%.
That 8% is not a promise that the casino will keep exactly eight cents from each dollar today, from each player, or even from one machine this month. It is a long-run property of the approved game math under the stated wager and configuration. Real results arrive unevenly, so casino operators also track actual hold: what the machine or group of machines actually won relative to measured coin-in over a defined period.
The difference between those two numbers is where most confusion begins.
Theoretical hold and actual hold answer different questions
Theoretical hold asks: What percentage of eligible wagered value is the game designed to retain over its mathematical cycle or sufficiently large sample?
Actual hold asks: What percentage did the casino actually win from the measured coin-in during this reporting period?
Using compatible meters, the operating formulas are:
Theoretical hold = 100% − theoretical RTP
Actual hold = actual win ÷ coin-in
Suppose a machine records $100,000 of coin-in during a week and $7,500 of actual win. Its actual hold for that period is 7.5%. If its approved theoretical hold is 8%, the week finished half a percentage point below theory. That does not establish a fault. It is one observed result from a random game.
A different week might produce $14,000 of win on $100,000 coin-in, or only $1,000. Both can occur without any change to the underlying configuration.
For the player-facing side of the same relationship, see Slot RTP and Slot Machine House Edge.
Coin-in is not the cash a player inserted
This is the most important operational distinction on the page.
Coin-in is accumulated wagering action. A player may insert or transfer $100, win and replay credits several times, and create $500, $1,000, or more of coin-in before cashing out or reaching zero. The same credits can circulate through many wagers.
Physical cash, ticket value, account funding, or buy-in therefore cannot be substituted for coin-in when calculating slot hold.
Consider a player who starts with $100 and makes 200 spins at $2 each before leaving. Total coin-in is $400. If the player leaves with $80, the player’s net loss is $20, but the machine processed $400 of action. A theoretical hold of 5% would attach about $20 of expected loss to that $400 of coin-in. In this simple example, the actual result happens to match theory; most individual sessions will not.
This is why Coin-In matters so much in slot analysis. The casino’s mathematical exposure is created by repeated wagering, not merely by the first bill or ticket presented to the machine.
“Casino win” also needs a defined meter set
Actual hold sounds straightforward until two reports use different definitions of win.
In an operating analytics report, machine win is commonly derived from compatible game meters and may be expressed as wagered value less payouts, with adjustments according to the system and accounting design. Statutory gross-revenue reporting can follow jurisdiction-specific accounting rules that are not identical to a simple player-facing “coin-in minus coin-out” explanation.
That distinction matters around:
- hand-paid jackpots;
- linked-progressive settlements;
- promotional or noncashable credits;
- ticket-in/ticket-out movements;
- canceled credits and attendant-paid amounts;
- meter resets or replacements;
- machine moves and conversions;
- accounting-period cutoffs.
The right question is not “Which formula is universally correct?” It is “Which meters and adjustments does this report define, and are numerator and denominator measuring the same activity?”
Slot Accounting Explained goes deeper into this reconciliation problem.
A three-machine example shows why one period can mislead
Suppose three machines each have $100,000 of weekly coin-in.
| Machine | Theoretical hold | Actual win | Actual hold |
|---|---|---|---|
| A | 8.0% | $7,900 | 7.9% |
| B | 8.0% | $1,500 | 1.5% |
| C | 6.0% | $12,000 | 12.0% |
Machine A landed close to theory. Machine B was very player-favorable for the week. Machine C was very casino-favorable.
Nothing in those three rows, by itself, proves that B is loose, C is tight, or either machine was reconfigured. A large jackpot, a cluster of bonus rounds, ordinary variance, or simply limited sample size can move actual hold dramatically.
If the same pattern persists across a much larger amount of coin-in, operators investigate more seriously. The investigation still begins with data quality: configuration, meters, jackpots, denominator, dates, denomination, progressive participation, and whether the comparison group is truly comparable.
Sample size is more important than the calendar label
A common reporting mistake is to treat “one month” as automatically meaningful. A month on a high-volume bank can represent enormous coin-in. A month on a lightly played game can represent very little action.
The amount of relevant play matters more than the name of the period.
Variance also differs by game. A low-volatility game can cluster around theory faster than a highly volatile game with rare large awards, even when both have the same theoretical hold. Progressive games can look especially strange over short periods because jackpot funding and jackpot hits are not evenly timed.
So when actual hold differs from theoretical hold, ask:
- How much coin-in is in the sample?
- What volatility and jackpot structure does the game have?
- Were any exceptional pays or meter adjustments recorded?
- Is the theoretical percentage for the exact active configuration?
- Are all machines in the comparison group using the same wager model and denomination assumptions?
Without those checks, an actual-hold percentage can be precise and still be uninformative.
Higher theoretical hold does not tell you how fast one player will lose
A 12% theoretical hold game is more expensive per dollar of long-run coin-in than a 6% game, all else equal. But session experience also depends on wager size, pace, volatility, bonus structure, hit frequency, and how long credits are replayed.
Two machines can therefore create very different experiences even if their theoretical hold is identical.
Imagine two 8% theoretical-hold games:
- Game X pays frequent small wins and has relatively modest volatility.
- Game Y pays fewer ordinary wins but has a rare large top award.
Their long-run hold can be the same while the path to that result feels completely different. One player may get a long session on X and a fast loss on Y; another may hit Y’s large award and leave far ahead.
Hold is an expectation, not a session script.
For this reason, a player comparing games should not use hold or RTP alone. Slot Volatility and Time on Device Calculator explain why the shape and pace of returns matter alongside the average percentage.
What an operator should investigate when actual hold looks wrong
A good slot manager does not treat every unfavorable variance report as a mathematical emergency. The first response is structured verification.
A practical sequence is:
Confirm the game identity. Is the report attached to the correct asset number, theme, denomination, and approved configuration?
Confirm the meter period. Do beginning and ending meter dates align with the reporting period? Was the device moved, reset, converted, or replaced?
Confirm exceptional payouts. Did a jackpot, hand pay, or linked award distort the short period?
Confirm the denominator. Is the report using actual coin-in, not physical drop, ticket-in, or a funding measure?
Compare with sufficient history. Is the variance persistent over substantial action, or is it a normal short-run swing?
Escalate unexplained divergence. If a large sample remains inconsistent with expected performance after accounting checks, technical review becomes more important.
This process protects both sides. It prevents staff from dismissing a genuine meter or configuration problem as “variance,” and it prevents ordinary random performance from being misdiagnosed as tampering.
Approved game math is not an attendant-controlled dial
Modern regulated gaming devices are approved and controlled products. Changing game rules, available wagering opportunities, or theoretical hold is not equivalent to an attendant selecting a mood for the floor.
Nevada’s current Regulation 14 treats a change in theoretical hold percentage as a game/device modification subject to its approval framework. Gaming Laboratories International likewise publishes technical standards for gaming devices that regulators and testing laboratories can use as part of certification and control work.
For primary-source examples, see the 2026 Nevada Regulation 14 amendments and GLI-11 Gaming Devices.
Jurisdictions differ in approval details, so those documents should not be treated as universal law. The operational principle is broader: theoretical hold belongs to controlled game math and configuration, while actual hold belongs to observed performance.
Hold can be measured at several useful levels
Casinos may review hold for one machine, a bank, a theme, a denomination, a zone, a manufacturer grouping, or the whole property. Each level answers a different management question.
A machine-level view can help identify an outlier. A bank-level view can show whether adjacent devices are performing similarly. A denomination view can reveal differences in guest behavior. A property-level view is useful for revenue forecasting but can hide individual machine problems.
The danger is averaging incompatible things. If a report combines very different theoretical holds, wager structures, progressive games, and volumes, the blended actual hold may be mathematically correct but operationally weak.
Weighted averages are usually required. A machine with $1 million of coin-in should influence a floor-level hold calculation more than a machine with $10,000 of coin-in.
Translate the percentage back into dollars
Percentages become clearer when converted to expected value.
Suppose a game has:
- theoretical RTP: 94%;
- theoretical hold: 6%;
- monthly coin-in: $500,000.
Its simplified theoretical win is:
$500,000 × 0.06 = $30,000
If actual win is $24,000, actual hold is:
$24,000 ÷ $500,000 = 4.8%
The machine finished $6,000 below simplified theoretical win for that period. That does not mean $6,000 is “missing.” It means the observed result fell below the long-run expectation over that sample. The next step is to decide whether the sample and accounting are good enough to make that variance noteworthy.
You can test the same relationship with the RTP Comparison Tool and Expected Loss Calculator.
The clean definition to keep
For slot math, theoretical hold is the casino’s long-run percentage of coin-in implied by the approved RTP; actual hold is the casino’s observed win divided by compatible coin-in over a defined period.
Keep coin-in separate from buy-in and physical cash. Keep theoretical performance separate from actual variance. Keep statutory revenue reporting separate from a simplified machine-performance ratio. Once those distinctions are in place, “hold percentage” becomes a useful operating measure instead of a vague statement that the casino simply keeps a fixed slice of every player’s money.
Continue with Payback Percentage for the player-side mirror, How Payback Is Configured for configuration context, and Slot Hold and RTP from the Casino Side for the management view.