In the late 20th century the Soboba band of Luiseño Indians established a gaming casino in Southern California. The property operates approximately 2,000 gaming machines, 20 table games, two food outlets, three bars and a gift shop, and also has a 3,500-seat outdoor arena. The property draws millions of customer visits yearly from across Southern California. The property operates the IGT Advantage gaming system for gaming and customer management.
On the map in Figure 1, we can see that the Soboba Casino property could draw from a wide geographic area, with the significant nearby areas being Riverside and San Bernardino. One important observation to note from this map is that the visitors making the journey to Soboba area likely to be making a significant investment of time.
So what does this positioning mean in today’s industry environment?
According to Tom Davenport, “In virtually every industry, many former strategic alternatives are no longer viable or likely to be successful. Today, there are few regulated monopolies, or companies with unique geographical access. Proprietary technologies are rapidly copied by competitors, and breakthrough innovation in either products or services is rare. Most of the competitive strategies organizations are employing today involve optimization of key business processes. Instead of serving all customers, they want to serve optimal customers—those with the highest level of profitability and lifetime value.”1
In gaming, this shift has been profound. No longer can we just build a casino and be confident that customers will flood in. Today, most properties are in highly competitive markets where opportunities for growth are limited. Now, as the gaming industry looks for growth, stories of how analytics can enable revenue improvement will become the poster children for achieving success. Let’s look to the success story at Soboba and its transformation into an analytics-focused company.
The Southern California market is no exception to recent market trends, and the last eight years have seen few opportunities for new properties following the expansion that marked the early 2000s. Soboba is no exception to the trends, either. The red line in Figure 2 shows Soboba’s revenues in 2013, the year before it shifted focus to analytics using VizExplorer. The results of this shift at Soboba are shown by the blue line, which represents its 2014 revenue numbers. [Note: The graph in Figure 2 has had its numbers obfuscated.]
The difference is remarkable.
If we take a closer look at the October comparison in Figure 2, we can see that the After VizEx numbers remain nearly flat compared to the massive drop seen in the Before VizEx numbers. And if we take another look at the February numbers, the gap is huge. While the exact dollar amounts have been obfuscated, we can say these results are remarkable, especially considering that the word on the street is that the general market is flat. These wonderful results are a testament to how, done correctly, operationalized analytics can drive substantial benefits.
Primer on Market Basket
The basic question in market basket analysis is, if a customer likes this product, then will they also like this associated product? For example, if a customer likes blackjack, will they also like craps? Or if a customer is playing on a $25 blackjack table, will they also play on a $50 blackjack table? Or if a customer likes The Hangover game, will they also like Wheel of Fortune? Now this concept is quite simple, but the mathematics is complex. There are many advanced algorithms that have been applied—which we authors have written about previously—to try to solve this problem. However, it turns out that a data noise filter does the job.
Basically, if we make a measurement of a number and behavior and it gives a very clear and simple result, then we can consider data to be clean. For example, if we ask for an analysis of the games that a customer prefers, then we should really ignore games that the customer only played a few times. In short, we need an analytics method with which we can filter out these outliers. Or, in other words, filter out the noise.
Let’s consider an example. If Charlie spends 80 percent of her time on Buffalo, Wheel of Fortune and Hangover and 20 percent of her time on about 10 other titles, we can say that she prefers those first three games. The 10 other titles can be considered to be noise in the data and in many cases should be ignored. This data filter removing the 10 extra titles is called a preference filter.
Let’s take a more in-depth look now at how this comes into play when trying to optimize the gaming floor.
Magnet games are leaders in slot performance. They are highly correlated to their surrounding games, and they also outperform the surrounding products (see our discussion of losers, leaders, loners and laggards in the December 2012 issue of CEM). These games are wonderful additions to a gaming floor, as they both attract players to a game and they might also create “spill,” driving additional play on surrounding games. If a game has a high spill effect, then we have truly found a driver that can be used to optimize the gaming floor. Furthermore, we can measure this spill effect as a “lift” on surrounding games by looking at reverse cannibalization. To do this, we need to understand these two new game metrics of spill and lift, which we can measure by performing a directional analysis.
Understanding if a player has game loyalty provides predictive knowledge about many aspects of game optimization. If we intend to move a game, players who have game loyalty need special consideration, and possibly specific communication, about the movement of the game. Also, games with loyalty may present an opportunity to become the centerpiece of a mini casino strategy, as we’ve discussed in previous CEM articles about Penny Alley, Jackpot Wharf2 and Paradise Fishing.3
One of the major differences between loyalty to a game and preference for a game is the amount of time it takes to measure the two. Preference for a game can be measured on a single visit—with the understanding that this preference can change over time. Loyalty to a game cannot be measured on a single visit, or even a handful of visits. However, loyalty is no doubt measurable. For example, we can look at the effects of physical movement of the game and measure the number of players who are prepared to seek out the game in its new location. This brings us to the very interesting challenge of calculating the probability that a player is truly loyal to a game if they exhibit game preference.
In our publications on Jackpot Wharf and Paradise Fishing, we showed how the preference filter was applied to create market baskets and optimize the rooms. This technique has evolved in its application to where it is, in the view of us authors, impossible to see the nuances of the data without this method. To understand the insight, we will lay out two different views of the data. In this first view we followed the following steps.
View One: Preference Play on Target ALL Other Play
Select the games in question, in this case bank 0403 (see Figure 3).
Find all the players who play on games in this bank with preference.
Show where else these players play (with no preference filter).
The bank in question is 0403 and is marked with red pointers in Figure 3. Notice that the play around the gaming machines is very hot. Also notice that the treemap shows that games are high in the hierarchy. Said another way, by looking at the slot floor map we can see that customers who prefer to play bank 0403 also (unsurprisingly) have a great deal of play in the surrounding games. By looking deeper at the treemap and the hierarchy displayed there, we see that in fact the play on these surrounding games is hotter than nearly anywhere else on the floor.
The concept of hierarchical analysis that we used above is not a simple one, so let’s spend a moment here discussing it in further detail. Suppose that we look at a slot floor map and discover three games that are “hot.” Let’s put some numbers around that concept and say that the three games are each doing $250 win per day, on a floor with a $175 win per unit per day average. If we ignore the concept of hierarchy, all we can say is that these games are doing well compared to the floor average. But if we bring in a hierarchy such as manufacturer, denom and game type, we may find drastically different results.
Scenario A—The three games are all IGT 25-cent video poker. Looking deeper at the other IGT 25-cent video poker games, we find that it is a very popular game configuration, and in fact across the floor all 25-cent IGT video poker games perform at $275 win per unit per day. In this scenario, our three “hot” games are in fact quite cold when compared with the relevant hierarchy of manufacturer, denom and game type.
Scenario B—The three games are all different, and within each of their respective classes of manufacturer, denom and game type, these games perform as one of the top five games in each class. These then are in fact three hot games!
We can see then the importance of looking not only at absolute data, but also looking deeper at data relative to a meaningful hierarchy (as shown in View One here and View Two below).
View Two: Preference Play on Target Preference Filtered Other Play
The process to select the second view is very similar and also follows three steps, but the difference is that the final step only shows other preference play.
Select the games in question, in this case bank 0403.
Find all the players who play on games in this bank with preference.
Show where else these players play (with preference filter).
As we can see in Figure 4, now the results are very different. First, when compared with other preference games, the games are now far down the hierarchy. Second, the associated hot games are far away. This analysis shows how the market basket is a combination of location and product preference. In other words, one without the other is flying blind. To understand the market basket, I need to understand the other games the players like to play and where those games are located.
Understanding what customer behaviors are causing the spill and why is crucial. Examine the same players’ preference play and identifying occupancy in preferred areas, players’ worth and visits. Let’s assume these players have high ADT at $300 and low visits at once per month. These players are 10 percent of the database and make up 25 percent of theoretical win. Here exists potential to entice more trips.
Floor mix is not just for balancing denomination, manufacturer and floor to revenue share anymore. Floor mix can be assessed based on player preference by identifying what attributes the preferred games have and how many you have in the preferred areas of the floor. Preferred games in this area are penny video slots with multiple progressive meters and a 50L/250C bet structure. There are three preferred titles and 30 high ADT/low visit players per machine that occupy those games 60 percent of the time. So each customer can only occupy one of the three games 2 percent of the time. These are hot games to these high valued customers.
With this data, operators can better understand the players’ experience. The customer is exhibiting three basic behaviors, enjoyment when they can play that game they prefer, frustration because they need to go in search of another game they prefer and waiting for the preferred game. If they are waiting, they are not playing. If they are enjoying themselves 2 percent of the time this is a good experience. The other 98 percent of the time is not a good experience.
By assessing the floor mix in this way, making decisions to increase the games these customers prefer results in more time on the preferred game by each customer. Increasing these games and letting the player know there are more of their favorite games on the floor will likely increase trips and spend if increasing game count is not an option because they are end of life or there is limited floor space. Since attributes of these games have been identified, select other games on the floor with similar attributes and suggestive sell. Since the market basket is a combination of location and product, placement of similar games near or around the preferred games is ideal.
Bringing it All Together
Like the results from Jackpot Wharf in 2012, the final results at Soboba are amazing. What’s more, it is clear that we are now able to look past optimization of the gaming floor based on game performance. We authors would argue that it is irresponsible to make game product, price or place decisions without first understanding player preference and market basket.
1 Davenport, http://www.babsonknowledge.org/analytics.pdf
2 CEM January 2012, Cardno, Thomas, Evans: Jackpot Wharf, Part 1
CEM February 2012, Cardno, Thomas, Evans: Jackpot Wharf, Part 2
CEM December 2010, Cardno, Singh, Thomas, Evans: Penny Alley, Part 1
CEM January 2011, Cardno, Singh, Thomas, Evans: Penny Alley, Part 2
CEM February 2011, Cardno, Thomas, Evans: Penny Alley, Part 3
CEM March 2011, Cardno, Thomas, Evans: Penny Alley, Part 4
3 CEM June 2012
Andrew Cardno has more than 16 years of experience in analytics ranging from modeling health-care drive times to gaming floor analytics. He presents on the future of analytics and is living in the U.S. and works with worldwide corporations. He serves as the chief technology officer of VizExplorer. om.
Dr. Ralph Thomas, chief data scientist and vice president, gaming division, VizExplorer. During his 10 years in the casino industry, Thomas has focused on maximizing profitability by applying statistical analysis to company databases. Previously, Thomas spent 15 years in academia, as both a student and a lecturer of mathematics.
Alicia Hawkins is a financial planning and analysis analyst at Soboba Casino, specializing in dynamic slot floor operations and financial performance. Alicia has 10 years in retail and business management focusing on financial and operational analytics. Certified in slot performance analysis and proficient in data reporting systems, Hawkins leads the adoption of real-time visual data analysis and its practical application.