Authors’ Note: This article will define two kinds of metrics: optimization metrics and outcome metrics. We will also explore how metrics that are related to the player experience can be used as the central driver of gaming floor optimization. In future articles, we will explore how we can use mini casinos combined with optimization metrics, such as corrected utilization and corrected theoretical win, to explore revenue optimization.
Ten years ago it could be argued that the best way to make money in a casino was to open one. In those good ol’ days it was a reasonable optimization strategy to simply keep the best-performing machines, and optimization strategies based on this approach were quite successful. In today’s world, however, many markets are saturated and now it is all about beating the competition—and, in many cases, this competition is from non-traditional sources. For example, we now compete with online entertainment.
We have spent a great deal of time thinking about slot optimization and looking for methods to increase gaming revenue by making improvements to the slot floor. Exploring the philosophical side of gaming floor optimization, we can ask “What is slot optimization?” From the slot operator’s perspective, the answers can include: Optimizing win per unit per day, or WPU; replacing poor performers (based on WPU) with new product; and the effort to make sure each machine is driving its fair share of the budget (via WPU).
Digging up one of our previous articles, “An Analyst’s Guide to Slot Floor Optimization,”1 we found a wealth of optimization tactics, including this slot optimization philosophy: “But how do we know when the gaming floor truly is optimized and we can go play golf? Well, when the incremental WPU for each machine is the same!”
This philosophy may well have been described as optimizing incremental WPU. In the months that have passed since we wrote that article, our philosophy has evolved. We still believe that understanding incremental gains is fundamental to optimization efforts, however, we’ve learned that true slot optimization is best defined from the players’ perspective.
The Player Perspective
To truly understand gaming floor optimization, we have to put ourselves in the position of the player. Players do not, at least in our experience, make any decision regarding the win per unit per day of the gaming machine. And since the floor hold percentage is more dependent on the mix of gaming products than the hold of the gaming machines, players are also not interested in the overall hold percentage (although they are likely to respond to different pay tables on video poker). Finally, in the past, denomination or denom was a key driver of player choice. But in this new era of penny games with a wide range of configuration options, denom is no longer the driver of player choice.
From this, player-centric optimization is defined as improving the player experience in ways that drive the most incremental player net revenue. Expansion of this definition results in two kinds of metrics: optimization metrics and outcome metrics.
• Optimization metrics: These are metrics that measure effects the players can observe. Furthermore, it is generally desirable to optimize these metrics to drive incremental revenue. In short, optimization metrics are metrics that the player notices. Utilization is one metric that the player notices—you might hear statements from patrons like “I found my favorite” or “the best games are always occupied when I want them.”
• Outcome metrics: These are metrics that the players do not observe. For example, the theoretical win per unit per day on the gaming machine. It has been shown how the player experience2 differs dramatically from the expected or theoretical outcome, and furthermore, the theoretical win per day is an average from a number of different players. Quite simply, players do not experience the spending of other players. Another outcome metric is the slot floor hold percentage, or what is often incorrectly termed the “price” of our games. In our article “Hold Another Sacred Cow”, we debunked the myth that one can increase gaming revenue by simply tinkering with the overall floor hold percentage.3 Quite simply, the hold percentage of the whole gaming floor is an outcome metric that is more reflective of the mix of gaming product than the player experience.
Digging deeper into our player-centric slot optimization philosophy, the goal is to improve the player experience, thereby driving revenue and beating the competition. One cannot improve the player experience by watching WPU numbers. Rather, we must make changes to our slot floor to improve the player experience, and do so in ways that drive incremental player spend.
Consider this illustrative example: Imagine having an extremely popular slot machine. WPU for this slot machine is four times the floor average in WPU. Looking at some of the optimization options we encountered previously, it is tempting to look at that game and say “great, it’s doing its job” and then focus on improving the WPU of lower-performing products. But now, looking at optimization from the players’ perspective, a game with a WPU of four times the house average is a potential problem and, furthermore, a revenue opportunity.
If this high-performing game is in the high-limit room, it is likely to have low utilization. But if it is in a high-volume area of the floor, it might be a problem. Quite simply, it may have a negative effect on player experience. In an extreme example, if the game is 100 percent utilized, players are in effect always competing to play. Furthermore, these players may choose to play at a competing property where the product is more available. In this situation, the opportunity exists to optimize utilization to enhance the player experience and drive incremental revenue. As such, we can apply utilization to drive incremental revenue by enhancing the player experience.
Of course, that’s all fine and good as a philosophy and for this simple example, but in the real world our patterns are finer grained, our data is large and many dimensioned, our player numerous, and our capital is limited. The next section introduces some methods for handling these real-world challenges.
Slot Change Implementation Metrics
Let’s take this to the next level by looking at optimization metrics in more detail and showing how statistical clustering methods can be applied to this data to provide real-world optimization.
The player cares about time on device and their gaming experience. Put another way, how much does the player expect to lose over an hour of play (SPH)? The factors that drive this metric are:
1. Average Bet – Certainly how much the player spends per play is a big part of SPH. Average bet itself is a complicated function of the configuration of the game via denom, minimum bet, maximum bet, maximum bet to cover all lines, etc.
2. Game Speed – A slow game can provide for longer time on device, whereas a fast game can provide for more exhilarating action. The player can choose which speed suits their play style better, but this factor is vital to understanding game performance in general and SPH in particular.
Once we calculate SPH, we can better understand the “cost” of each of our games from the players’ perspective of time on device.
So now our player has selected a game that provides an SPH that he/she is comfortable with. The next big question likely to come from our players is “Where is my game?” Studying the location of games is a critical part of this and will be covered in a future article.
By cross tabulating the quadrants of SPH of the player versus SPH of the machine (and leaving aside the debate of whether we should be using median or mean or Monte Carlo simulations), four categories of playing experience are created (see Figure 1):
Grinders: These players are spending below their typical SPH. If this is because they cannot find the gambling experience they want, then we have an opportunity.
Losers: These players are being hammered. Unless they are lucky, they are unlikely to have the wallet to continue this gaming experience.
Gamers: These players are spending at low amounts and often represent the majority of the utilization on the gaming floor.
Gamblers: These players probably drive the majority of the revenue in the property. It is critical that we optimize their gaming experience to ensure that they find the gaming devices they want.
Quite simply, we want patrons to have the gaming experience they are looking for. This experience should be neither above nor below their “expectations” and should be on the game they desire at the location they desire.
Applying these player experience concepts to groups of players or individual players is even more interesting. For example, a high-value player may be grinding for a while but then go back to gambling. When building the optimization models around this data, it is critical that we consider these gaming experiences. To illustrate this point, consider what would happen if the optimization model mixed these four player experiences into one number. This WPU number would probably be driven by gamblers but the machines would likely be occupied by gamers. One model at least one of us authors has applied is to maximize gambling and minimize gaming.
Another question players ask is, “Is my game available?” To answer this question, we leverage utilization. The calculation is simple: What percent of the time that a game is available is it being used? Utilization is often calculated on a 24-hour basis, meaning that whether a game is being played at 4 a.m. on a Tuesday or 7 p.m. on a Friday, the amount that play contributes to the overall utilization score is the same. To handle this, consider utilization as one method to simply compare games. If one game has 50 percent utilization and another has 10 percent utilization, it is far more likely that the game with 50 percent utilization is going to be busy when a player wants to play it than the 10 percent game.
Let’s finish this particular discussion with another illustrative example. Suppose there are three games available and we need to decide which game needs more units. In terms of our metrics, we have:
A) 50% utilization and $15 SPH
B) 20% utilization and $50 SPH
C) 5% utilization and $250 SPH
Let’s calculate WPU for each scenario, taking the utilization x 24 hours in a day x the spend per hour. The WPU for Game A is $180, Game B is $240 and Game C is $300. A traditional WPU analysis of these games would leave an analyst thinking that Game A is the weakest and that no more units are required. However, what if it was discovered that incremental spend from the player occurs once utilization crosses a certain threshold, say 30 percent? If that were the case (and significant slot change analytics needs to be done to determine this threshold), then only Game A could provide incremental spend via incremental units. Adding more of Games B and C may only result in diluting the performance of the existing games. (See Figure 2.)
So far we have left utilization as a generic concept. In practice, overall utilization can potentially be as useless as overall floor hold percentage. A player who only plays on Friday nights does not care about the overall utilization of a game; they care about the utilization of a game on Friday nights! As was mentioned above, this may not be an issue in less competitive environments, but as our industry gets more and more competition, and as our supply and demand reach close to equilibrium (or even oversupply), understanding utilization at the proper time periods becomes more and more important. However, once we place a game, we are stuck with it 24 hours a day, until we choose to replace it, though server-based games have long held the promise of helping with this time of day optimization. One powerful method for handling the dimensionality of optimization across different time periods is clustering.
In the past it could be said that operators have assumed that they have one player. Player-independent metrics like WPU and floor hold percentage were used, and the operators might say things like, “Don’t worry about where you place that game, the player will find it!” Lost in this view is the fact that we have thousands upon thousands of players. And while these players may have behaviors that we can average down to statements like “the player spends $100 per visit,” this actually applies to very few of our players. As an exercise, take the average spend of your players, then see what percent of your players actually have an average individual spend within 5 percent of the overall average. Most likely the vast majority of your players do not actually fall within this range.4
Using methods discussed in “Clustering: The Key to Understanding High Dimensional Data,”5 group your players by not only the games they choose to play, but also the time periods they choose to play in. By adding this dimension of time to our clustering methodology, the analysis measures utilization across various player clusters and gives us a better understanding of which games cause our players to answer the question, “Is my game available?” with a frustrated “No!” We can relieve that frustration by getting more of that game, thus improving the player experience.
The concept of player clustering can be a difficult one to grasp, so here is a clarifying example. Instead of the complicated question of, “Is the game available for the player when they want to play?” we ask a more simple question: “Do players make their game decisions based on the look of the game or on the play mechanics?” At least one of the authors performed a simple clustering analysis of slot product and discovered that the answer to the question is both.
There are certain player clusters for which play mechanics is the primary driver of their game choice. We see this when manufacturers offer “clones” of their games, meaning the math and game rules are the same but the symbols and pictures on the cabinet are different. Certain player segments have identified that this play mechanic is the one for them, and they play all different versions of this product, regardless of how the game looks.
However, there is a completely different cluster of players that are attracted to what can only be described as “exotic” games. These are the games we often see at industry conventions, with bright lights, fancy seats and, often, pop-culture references. These exotic game chasers love the look of a new game and could care less that these games come with a wide variety of different play mechanics.
This non-WPU method for slot floor optimization sets out to enhance the player experience and use this enhanced experience to beat the competition. We have been applying these methods with some remarkable success, and as our philosophy of gaming analytics advances, we expect to see further refinements in how to drive this incremental revenue. Of course, the choice is yours. You can continue to optimize using WPU and hope that the methods that worked so well 10 years ago will continue to be effective, or you can switch to a player-centric, and likely more profitable, approach to optimization.
1 “An Analyst’s Guide to Slot Floor Optimization,” November 2010, Casino Enterprise Management.
2 “The Long and Short of It: Slot Games from a Player’s Perspective,” Singh, Cardno, Gewali, April 2010, Casino Enterprise Management.
3 “Hold Another Sacred Cow,” Cardno, Thomas, April 2011, Casino Enterprise Management.
4 “Clustering to Uncover Hidden Behavior: A Case Study of Silverton Casino,” Cardno, Thomas and Evans, April 2011, Casino Enterprise Management.
5 “Clustering: The Key to Understanding High Dimensional Data,” Cardno, Singh and Thomas, February 2011, Casino Enterprise Management.