July 24, 2026 · 8 min read · casino math · comps · card counting

What your play is worth to the casino

Somewhere in the pit there is a number attached to you. It is not what you have won tonight, and it is not what you have lost. It is what the casino expects to win from you, computed from four inputs a floorperson can observe or assume, and it governs everything the property will do for you: the room rate, the buffet, the line pass, the call from a host you have never met.

The number is called theoretical win, or theo, and the formula behind it is not a secret. It sits in a UNLV Gaming Studies Research Center guide, in expired casino patents, in vendor white papers and in the revenue-recognition footnotes of public 10-K filings. Read those together and the comp stops looking like generosity and starts looking like what it is: a rebate, priced in advance, out of money the house has already calculated you will lose.

The formula

Robert C. Hannum (1953-2018) was Professor of Risk Analysis and Gaming in the Reiman School of Finance at the University of Denver, and co-author with Anthony N. Cabot of the textbook Practical Casino Math. His short guide for the UNLV Gaming Studies Research Center states the industry calculation plainly: a casino can determine how much it expects to win from a player using the house advantage, bet size, duration of play and pace of the game. He calls the result player earning potential, player value, player worth or theoretical win, and gives the formula as Earning Potential = Average Bet x Hours Played x Decisions per Hour x House Advantage.

Every term in that product is doing work, and only one of them is about you specifically. Average bet is yours. Hours played is yours. Decisions per hour belongs to the game and the table, and house advantage belongs to the math of the rules. Notice what is absent: any variable at all for outcome. Whether the cards ran hot or cold does not appear, because in this equation it is not information. It is noise around a number the casino already knows.

Hannum's own worked examples show the scale. A player betting $500 a hand for 12 hours at 60 hands an hour in baccarat, against a house advantage he rounds to 1.2%, is worth $4,320. The same $500 a spin for 12 hours at 60 spins an hour in double-zero roulette, at a rounded 5.3%, is worth about $19,000. Identical bet, identical stamina, more than four times the value, entirely because of which layout the chips landed on. Hannum's rounding is illustrative rather than precise here, and his own house-advantage table lists 5.26% for double-zero roulette and 1.06% to 1.24% for baccarat.

Rated on theory, not on results

The consequence of the formula is the part players find counterintuitive: the comp desk is largely indifferent to whether you won. Hannum is explicit that many casinos set comp policies by giving the player back a set percentage of their earning potential. He hedges honestly — policies based on theoretical loss are the most popular, he writes, but rebates on actual losses and dead chip programs are used at some properties too. The dominant practice, though, prices the customer off expectation.

This is not casino perversity. It is the only stable way to run the program. Actual results are wildly volatile over any single trip; expectation is not. If comps tracked wins and losses, the property would be paying its best offers to whoever ran worst last weekend, which describes luck rather than a customer. Rating on theo means the offer describes the size and duration of the action, which is the thing the casino actually sells.

It also means the rating is only as good as its sample. Theo divided by rated days produces average daily theoretical, the metric that sorts a database into tiers, and it stabilises slowly. A single session can misprice a player badly in either direction, which is why hosts talk about trips rather than hands, and why the number attached to you is not really a measurement of tonight. It is an estimate of the kind of customer tonight suggests you are.

How much of it comes back

The reinvestment rate — the share of theo returned as comps — is where the published figures stop agreeing with each other, which is itself the most useful finding. There is no industry standard number. There is a range, it is wide, it moves with the market, and every source characterising it is describing practice rather than reporting a survey.

Take the estimates in date order and treat them as a spread rather than a consensus:

US Patent 6,267,671, filed in 1999 and granted in 2001, defines the comp factor as the percentage of theoretical expected win the casino is willing to return in complimentary goods or services, and gives a typical range of fifteen to forty-five percent. That is a practitioner's characterisation inside a patent specification, now more than a quarter of a century old.
Andrew Cardno, co-founder and chief technology officer of Quick Custom Intelligence, told CDC Gaming Reports in January 2023 that a typical marketing-department reinvestment rate is about 20 percent of the casino's theoretical win from the player.
Cardno also told the same publication the margin varies wildly, from as low as 10 percent to more than 50 percent in a hyper-competitive market. Where you play changes what you are offered more than how you play does.
Hannum notes that while theoretical-loss policies are the most popular, some casinos rebate on actual losses or run dead chip programs instead — so a minority of offers genuinely are results-based.

A worked example at $10 a hand

John Grochowski's walkthrough for the American Casino Guide makes the arithmetic concrete at a normal stake. Take a $10 blackjack player at a full seven-player table, averaging 60 hands an hour: that is $600 an hour through the felt, and $1,800 risked across three hours. If the house assumes a 1.5 percent edge against an average blackjack player, the theoretical loss is 1.5 percent of $1,800 — $27. Apply Cardno's typical 20 percent reinvestment to that and roughly $5.40 comes back over the session. That is a soda and part of a sandwich.

Two of those inputs are house choices rather than facts. The assumed edge is one: Grochowski's example uses 1.5 percent, while Hannum's table puts a basic-strategy player at 0.50 percent, an average player at 2.00 percent and a poor player at 4.00 percent. Rate the same $1,800 at 0.50 percent and the theo is $9; rate it at 4.00 percent and it is $72, for identical play. The pace is the other. The hands-per-hour figures below come from Jim Kilby's Casino Operations Management, reproduced by the Wizard of Odds, and they say a full table is slower than Grochowski's round 60:

Heads-up against the dealer: 209 hands per hour.
Two players: 139 hands per hour.
Three players: 105 hands per hour.
Four players: 84 hands per hour.
Five players: 70 hands per hour.
Six players: 60 hands per hour.
Seven players: 52 hands per hour.

Where the number actually comes from

For all the precision of the formula, the table-games inputs are estimates made by a person walking a pit. The 1999 patent describes the practice without embarrassment: a casino employee, usually a pit person, surveys the player's wagering activity periodically, making handwritten assessments of average wager on paper slips or cards. Those assessments become the average bet in the equation, and the equation produces the number that decides your offers.

The patent is equally frank about what that implies. Casinos, it says, are dependent upon human assessments of both gross wager and house advantage, and as a result they approximate these variables. It also spells out the substitution at the heart of table rating: Gross Wager = Average Wager x Time x Decisions Constant. Handle is not measured at a blackjack table the way coin-in is measured in a slot machine. It is reconstructed from an eyeballed bet, a clocked duration and an assumed pace.

The slot side is better instrumented and still leaky. Gene Johnson, executive vice president at Victor-Strategies, told CDC Gaming Reports that 30 percent or more of slot play is uncarded, because of card-reader error or players simply not using a card, and that table play has historically been difficult to track precisely because it relies on pit supervisor observation. Grochowski's figures for what that tracking pays are similarly modest: slot-club reward rates run from as little as 0.1 percent of coin-in to a rare 1 percent, with the biggest cluster around 0.2 percent.

Why a counter's rated theo is wrong twice

Now put a card counter into the formula and watch it break. Theo multiplies a rated average bet by an assumed house advantage, and for the counter both terms are wrong. Hannum's house-advantage table lists blackjack card counting at -1.00 percent against 2.00 percent for an average player — an illustrative table value, since the guide gives no rules, deck count or counting system for it, but the sign is the point. The rating math cannot express a negative house advantage, so it books the counter as an ordinary customer generating ordinary expected losses.

The second error is the interesting one, because it is the error that gets noticed. A counter's edge comes from bet spread: small when the deck is poor, large when it is rich. But the rated average bet is exactly what the pit is periodically writing down, and a spread is, arithmetically, a rising average bet accompanied by an inconsistent one. The same observation that feeds the comp system feeds the surveillance question. The pit is not watching your count. It is watching the input to a formula, and a spread moves that input in a way flat betting does not.

Table occupancy compounds it. Kilby's numbers put heads-up play at 209 hands an hour against 52 at a full table — four times the decisions, and therefore four times the theo per hour at the same average bet. The player who wants an empty table for the count is also the player whose rated hourly value quadruples on the same rating sheet. Industry documents from the late 1990s show operators already treating proficient counting as a threat to theoretical win; a 1998 patent for a card-scanning shoe argued exactly that, though it was a product filing with a commercial motive to make the case, and its numbers should be read as a sales argument of its era rather than measured evidence.

The buffet has a price on it

The clearest proof that a comp is a rebate rather than a gift is in the accounting. Caesars Entertainment's Form 10-K for the fiscal year ended December 31, 2025 discloses revenues including complimentaries and loyalty point redemptions totalling $1.3 billion, against casino revenue of $6.617 billion and net revenues of $11.486 billion. The filing states that the retail value of complimentary food, beverage, hotel rooms and other services is recognised as a reduction of revenues for the department which issued the complimentary, and revenue for the department that redeemed it. Money moves between departments. It does not appear from nowhere.

The loyalty accounting is more explicit still. Caesars treats Reward Credits earned on gaming as a performance obligation, and because gaming activity has a wide range of highly variable selling prices, it uses the residual approach: the credits are valued first, and the leftover transaction price is allocated to the gaming activity. A significant portion of what you wagered is deferred and recognised only when the credits are redeemed. In the company's own books, the points are not a bonus on top of your action. They are carved out of it before the gaming revenue is counted.

None of that makes comps worthless. A rebate is real money and, at the margins, a genuinely good one can matter. But the arithmetic runs one way: a set percentage of a negative expectation is still negative. Hannum's guide is also careful to warn against the related confusion that hold percentage equals house edge — it does not, and the two are computed on different bases. The offer is priced off the theory of your loss, not off any measurement of it.

The same discipline, applied to the other side of the table

The useful takeaway is not cynicism about comps. It is that the casino is running a calculation on your play, in public, with a published formula and inputs you can identify. Average bet, hours, decisions per hour, house advantage. Two of those you control completely, one you choose when you pick a table, and the fourth you influence with how well you play the hands. That is a transaction with visible terms, which is rarer than it sounds.

It is also the same habit of mind the games themselves reward. Working out that a $27 theoretical loss returns about $5.40 in comps is the identical exercise as working out whether a hand is worth splitting, or what the count implies about the next few decisions: identify the variables, find the real numbers, do the arithmetic, and let the result be whatever it is. Nobody needs to memorise a rating sheet. But a player who can read the house's math on them is already doing the thing that makes the difference at the table.

The pit's formula is watching your average bet, which is exactly the number a counter has to move. Learning how and why that number moves is the whole skill — and the trainer is free.

Learn the count →

Share this post — the link unfurls with its own card.