Authors’ Note: In this latest installment of our series on the use of quantitative analysis to increase understanding of casino games and operations, we’ll explore how different analytical perspectives have completely different data requirements. In fact, a correct view of the data for a slot analysis may be misleading for marketing purposes. The mathematics involved here derives from set theory and database queries. This article will also begin to delve into data warehousing and data schema design, covering slowly changing dimension (SCD) and how these methods apply to downloadable games, something we will expand on in future articles.
When analyzing a gaming floor, it is critical that we first find out what question we are trying to answer and, just as important, who we are answering that question for, as this impacts the data requirements. In fact, there can be more than one correct answer to the same question. For example, the answer to “How is my gaming floor performing?” will change quite a bit depending on who is asking.
To illustrate the different perspectives in your casino, we’ll explore this question in three different ways:
1. From a financial perspective: How much money is my gaming floor making, and what are the contributions from individual games?
2. From the slot management perspective: How are my games performing?
3. From the marketing perspective: What impacts are my marketing programs or incentives having on the gaming floor?
Period Types and SCDs
Within the world of transaction databases, we are typically only concerned with keeping current records of financial transactions. However, in the world of analytical databases, we are far more concerned with putting together a seamless view of the entire environment so that data can be analyzed at a future point in time. For example, if we are looking at gaming floor data from January 2009 and are displaying the results on a map, it is important that the map we are using is correct as of January 2009. There are a number of approaches to doing this effectively. One is the “period type,” which database companies such as Teradata have implemented.
A period type typically has a beginning bound and an ending bound, and both of these must be of the same type (e.g., date, time or timestamp). A “period” represents a duration starting from the beginning bound and going up to the ending bound; it does not include the ending bound.
Using the period type, the process of querying the database is much simpler, because the database handles the storage of the history of the objects.
A more traditional approach is to use a slowly changing dimension (SCD). An SCD applies to a situation in which the attribute for a record can vary over time. Consider, for example, the case of a carded customer, Harold, who used to live in Las Vegas. The initial entry in the player lookup table of Casino Silver Nugget looked like this:
Harold has moved to Henderson now, and the Silver Nugget has to modify the customer table to reflect this change. This is known as the SCD problem.
There are many options for implementation of SCD, the most common being a Type 2 SCD. These variations involve adding columns for a start date and an end date to the record. The Wikipedia definition of Type 2 SCD is as good as any (See Table 1).
SCD for Slots
In the world of slots, the definition of a new record in the SCD is a little more complex than you might think. Following are some questions that should be considered before defining the SCD record:
• Do you want to analyze the effect of a change in a game’s location on its performance?
• Do you want to see the effects of a change in denomination on a game’s performance?
• How much change in the hold percentage is important for analytical purposes? For example, is a 0.01 percent change important?
• Should you track new chip changes to existing games?
• From an analytical perspective, does it matter when new chairs are added to a location?
• From an analytical perspective, does it matter if new signage is added to a location?
The challenge is that the more SCD records that are added, the more fragmented the data becomes. This balance between analytical fragmentation and the questions that can be answered makes the process of deciding what to include a critical part of the warehouse design. It is our recommendation that any changes to theme, location or denomination are tracked and that small adjustments to the hold percentage (which are often corrections) are not tracked.
The alternative is to create a maximum detail SCD that tracks every change as a new record. This alternative requires that additional views of the data are created to build SCDs for the different kinds of analyses. While this maximum detail approach is very appealing, pragmatism normally rules and only one SCD is kept for analytical purposes.
Table 1 shows an example of a slot configuration SCD. Each row represents an attribute in the dimension table. The key to this dimension table is that it combines attributes of the game and the location of the game (See Table 2).
This configuration table is quite a simplification of the machine configuration table as managed by the gaming system; however, it does hold additional information, such as the location of each gaming device. In other words, if a gaming device is moved, it will create a new record in the SCD. This simplification also enables users to answer business questions related to changes in the attributes being tracked. For example, it can show the number of theme changes at this location or, by linking to the transaction table, show the total revenue generated by game type, location or area of the property.
The converse is also true. If the Location ID, for example, is not part of the SCD, finding the answers to locational questions may not be possible. It would be difficult to answer questions like “How did the game’s move affect its performance?” or “How did this game perform in different areas of the floor?”
With the buzz surrounding downloadable games and the ability to place any game at any location with the push of a button, locational analysis will likely emerge as an important dimension in the gaming data. Of course, locational questions like “How are these themes doing in this location?” or “What is the best kind of theme for this location?” will still be difficult to answer. Because downloadable games can have a dramatic number of changes, the management of the associated SCD creates a complex data management problem—one we intend to dig into in future articles.
In the building of data warehouses, the data integrity questions associated with transactional data are relatively easy to answer—we can, for example, add up the actual win numbers and compare them to the total revenue. However, SCDs are much more difficult to manage. For example, if location changes are not captured accurately, as they happen, it is likely this information will be lost forever. If these changes are not accurately tracked, then any analysis that depends on the correct dimensional information may be fundamentally flawed.
Figure 1: Financial Performance Analysis
The Financial Perspective
From the financial perspective, “How is my gaming floor performing?” simply means “How much money is my gaming floor making?” These reports contain gaming machines that have been removed or reconfigured and must reconcile back to the overall volume of money that is generated by the gaming floor.
The queries to generate this data include all transactions for the period and link each transaction to the SCD that is appropriate for the time. As with all things, compromises in the construction of the SCD—for example, adjustments to the hold percentage—often result in variations in the final numbers (See Figure 1).
The Slot Perspective
Now, when moving to the gaming floor analysis from the slot perspective, the current or active floor at the end point of the analysis is of greatest importance. The gaming analyst is looking at different locations on the casino floor, the different games at these locations, and the numbers they are generating on different days of the week. For this gaming floor analysis, it would be quite misleading to include the performance of historic games in location performance numbers.
Figure 2: Slot Performance Analysis
The restriction here is that the transaction data includes only values from machines and their locations at the end date of the visualization. A common issue in the SCD here includes not tracking location changes of games; if these new locations do not create new SCD records, then the location performance numbers could be a blend of the transactions from an old location and the new location. An extreme case of this would be showing the numbers the machine generated while it was placed in the high-limit room on the new location the game was moved to (See Figure 2).
The Marketing Perspective
Enter the marketing department, and the real question behind “How is my gaming floor performing?” becomes quite different: “What impacts are my marketing programs or incentives having on my gaming floor?” The marketing analyst is asking where its XPC dollars got redeemed or where players who responded to its marketing program played on the gaming floor. A marketing person is unlikely to ask to be shown the transactions of only players who played on games that have not been changed by slots.
This quite different question requires a quite different analysis. For this analysis, we are only interested in locations on the floor at the end point of the analysis, and we typically do not wish to restrict the analysis to only games at these locations. The transaction data is shown at the location that it occurred, irrespective of the games that were at the location at the time (See Figure 3).
Figure 3: Marketing Performance Analysis
Three Correct Perspectives!
The correct management of the SCD enables us to answer questions related to how the gaming floor is performing and the impacts of configuration changes and locational moves. We have touched on some of the issues surrounding what is, in many ways, the centerpiece of many gaming floor-related analytical activities, and this centerpiece enables quite different analytical perspectives.
As we have illustrated above, each of the perspectives is correct, depending on who is asking the question, and each perspective will generate different data sets. The complexity here arises from the fact that the data management of the slot configuration SCD normally falls into the hands of the slots department, while the analytical perspectives extend across various departments. As is often the case in the world of business intelligence, this cross-divisional requirement places the data warehouse at the center of some interesting debates.