Home Where’s the Money Now, Part 6: The Human Side of Perspectives of Data: Financial Vs. Slots Vs. Marketing

Where’s the Money Now, Part 6: The Human Side of Perspectives of Data: Financial Vs. Slots Vs. Marketing

Authors’ Note: The perspectives of data show how three different consumers of gaming performance data can have three different sets of numbers and that all these different perspectives are actually correct. In this article, we draw heavily from the analysis shown in our book “The Math That Gaming Made,” co-authored with Dr. A.K. Singh, and expand with canonical exemplars that quantify the actual differences. The article describes how these perspectives illustrate the human side of data, a side that is messy, fraught with arguments and strange business rules that determine when things change. This human side results in examples in which it seems like there are no good answers on how to view the data.

Primer on Dimensional Data1
Most data can be divided into three broad categories. The first category is “things that can be added” called facts. The second category is the attributes of this data called dimensions and the third category covers structural aspects of the data.

Let’s look at a simple example. Consider a customer named “Alex.” This customer spends money, this money can be added and the customer has attributes, such as eye color and an address. Now let’s break this data into three categories.

Facts
Our customer Alex spends money and generates information relating to that spend of money. For example, he generates amount wagered, theoretical win, actual win and jackpot numbers. All of these numbers can be added and are in fact generally straightforward to reconcile. On the accounting side, the numbers are:

Historic: Accounting numbers are kept for years. The requirements for legislative records require that this information is kept in good form and can be reviewed by various government organizations.
Generally correct: This data is very correct; even small variations (say 1 percent), are carefully tracked down and adjusted because a percentage as small as 1 percent represents a significant amount of money.
Verifiable: At the end of the day you can count the money in the machines and determine the actual win. There are multiple meters measuring the input, and the numbers are easy to validate to a very high degree of accuracy.

Dimensions
As Alex drives from his new home to the property, we notice that his trip frequency has increased; in fact, we notice that Alex has doubled his trip frequency. This date of the change of address is not well known to us at the casino and while we might see it, due to mail redirects and other mechanisms, it could be months before the casino notices that his address has actually changed. This data is very human. It is:

Hard to keep a history: In Alex’s case, it could be months or—if Alex moves away—even years before we know we have the incorrect information. We have personally experienced physical mail turning up seeking to reach people who have not lived at the address for years.

Often incorrect: Not only is this information often incorrect, it is also totally unreliable. Consider the case in which the local customers have additional incentives to visit. The authors have seen customers using their friend’s address, their second house address, their mother’s address or even a P.O. Box with a local address. Quite simply, once the mail has left, the operator does not know if this attribute is correct; we only know that sometimes it is returned.

Difficult to verify: Let’s dig deeper into this look at other human dimensions, such as marital status. Even if we could gather this information from Facebook, we would be relying on the human who maintained the page to give the correct information, something that is again fraught with challenges. It is becoming common practice to run two Facebook accounts: one that people share with their workmates and a second for their family. This practice results in people engineering their appearance to fit the requirements for the role.

We can do some work to verify these attributes, but it is a constant battle with the data. We are constantly involved in discussions with good data entry policies on the importance of looking at drivers licenses, but this data is hard to verify.

Structural Aspects of the Data
Structural aspects of the data are hierarchal components that can be laid out into some kind of organization that goes beyond the basic attribution. This data is also very human in nature. Good examples are space and time. Consider space: the structure of the gaming floor, with walls, bathrooms, restaurants and gaming positions. All of these things are related. For example, the location of the bathroom is very important to humans and spatially related to the location of the gaming position. Clearly this data is structural in nature and in this case can be managed in a range of places from AutoCAD files to spatially enabled databases.

Structural data is also very human. It is:

Hard to reconcile the history: It is probably true that anyone who has tried to compare AutoCAD files from different years will understand that this process is an approximation at best.
Normally correct at least for a point in time: Consider the example of the AutoCAD file. Normally there is an up-to-date AutoCAD file that shows the building as it was constructed and is generally accurate.

The Value of the Human Side of Data
It is difficult to verify historically; it is easier with current records. For example, it is quite easy to verify the current floor plan—simply walk onto the casino and take a look. It is the history that is hard. Try thinking back 10 years to remember the structure of the property.

The human side of data, the dimensions and the structure are central to why we make money in a business. Consider the example of address. Approximately 14 percent2 of our customers move each year, and that means we have this huge ongoing flow in our customer bases that reflects a constant churn in the customer base simply due to physical movement in our customers. This huge inflow and outflow of customers is a primary reason why we gain and lose customers. The challenge is that without seeing the human side of the data, our analytics will always be driven by a factor that is not visible in our data, meaning our results at best will be inaccurate and, at worst, misleading.

Perspectives and What They Mean
Once the importance of the human side of data is established, there is one more step: understanding the different perspectives of the data. These perspectives are quite remarkable, as they show how different people within the same organization are quite correct in having different views of the same data. Here we examine the finance, slots and marketing departments and look at how they see gaming data in different ways. We will also see that all of these are actually correct for different reasons.

A Fundamental Question
What is the performance of a slot machine? This question seems quite simple: We look at the gaming machine and count the money that it took in. Now this would be a very short article if that were a good answer. Let’s look at some canonical characters from each department and re-ask the question.
3
Adam from Accounting: How much money did we make on the gaming machines last month? Adam is not interested in the individual machines except that he wants to break down the revenue by machine. When looking at a long period of time Adam is concerned that the totals from his accounting system match the sum of the allocation of revenue from each slot machine.

Samantha from Slots: Sam is very concerned about the amount of money that the slot machines currently on the floor are making. If a slot machine has moved location, Sam will not want to include the performance of that slot from its previous location. For example, if a slot machine moves from high limit to the main floor and the win changed from $800 win per day to $200 win per day, Sam would not average these numbers to say the slot machine is doing $500 win per day. Sam would be interested in the current performance and use the performance of the slot at its historic location only for comparative purposes.

Marty from Marketing: Marty is very concerned about customers. When asking the where did this group of customers spend money, Marty does not want the amounts filtered down because Sam moved a slot machine.

These questions are all reasonable and, in the view of the authors, quite correct.

To answer all of these questions, we need to introduce one more concept: a slowly changing dimension. This slowly changing dimension, or SCD, is a database attribute that has a start and end date. For example,the SCD for our fictional customer Alex, who changed his address, would indicate the following: the end date of the last record would be shown as the last day he was at the last address, and a second record would show the first date at this new address. Table one below shows a simplified form of this address slowly changing dimension.

Enter Slot Configuration
Slot configuration is a very complex setup, and many questions surround what constitutes a change in a slot machine configuration. This article examines only slot machine location moves at a simplified casino where there are only three slot machines in operation. The three slot machines are at locations 010201, 010202 and 010203 (this is a common numbering convention meaning section one, bank two and locations one, two and three). This data is summarized in Table 2.

These three locations all have slot machines. In total there are five slot machines, and we will simply call these slot machines A, B, C, D and E. Moves between locations are described in table three and then shown in the figures below with a vertical line on the slot machine location. For simplicity we have added the performance numbers for each slot machine (see Table 3),and we also assume 100 percent rated play.

Three Pieces of Information to Illuminate the Data
1. Looking at the table, you can test your interpretation by verifying that slot machine A was nowhere to be seen on 3/2/2013.
2. Slot machines A and C have both been in the casino since opening, and slot machine A was moved once.
3. There are a total of three slot machine changes, two of which involved new slot machines.

The Specific Question
What is the performance of my slot locations 010201, 010203 and 010203 from Jan. 1 through March 31?

Adam from Accounting: Performance Analysis
Adam is interested in all of the money. In particular he is looking at the performance over the entire time and irrespective of the location of particular games. This is shown by the green circle covering all the locations and not paying attention to the changes in games at the locations. Adam’s primary concern is the total amount of money and that no money has been lost. Also notice that Adam does look at the performance of all machines individually; this is shown by the thick blue line.

Samantha from Slots: Machine Analysis
Samantha is thinking about the current performance of the gaming floor. This is shown by the green circles only circling the performance of the most recently placed gaming machine. For example, at location three only the performance of Machine A (refer to Table 3) is shown.

Marty from Marketing: Marketing Perspective
Marty from Marketing is thinking about which machines did the customers he marketed to spent their money. The movement of Machine A from location 010201 to 010203 does not affect his analysis. This does create some real questions about how to exactly reconcile these different views, as Marty might be tempted to say that Machine A is in two locations at the same time. For the purposes of this analysis, if the game moves, then the marketing analysis will look at the machine’s entire performance over the time period.

The Human Side and Resolving the Numbers
Now while the debate rages among Adam, Sam and Marty, we can see that all are just different and correct perspectives of the human or dimensional side of gaming data. It is this human side of data that in the experience of the authors drives considerable complexity in building and managing data. It is also this human side that often delivers the value from analytics as it is a representation of how the business is structured or operated.

Schools of Thought in Data Modeling
We briefly touched on an interesting issue in the world of data modeling. In our attempt at making a simple example of three slot machine locations, we combined dimensional data (which machine was where and when) with facts (how much money did the casino take in for each machine/location/time period). In addition, we did this in the presence of a slowly changing dimension, which is in fact highly unusual.There are many schools of thought on how to model data, and within those different schools of thought, there are two major ones that exist in data warehousing. The first is normalized data modeling, sometimes called the Bill Inmon school, and the second is dimensional modeling, sometimes called the Ralph Kimball school.

Normalized data modeling, or Normal Form Modeling4 was invented by Edgar F. Codd in 1970.5 This model quite simply removes duplicates in the data structures. For example, a customer has an address, and this address may be shared between customers (say, a husband and wife). In a fully normalized model, the address would probably be broken out into a separate database table and not duplicated for both the husband and the wife. The normalized data model has many advantages and is less prone to errors. For example, a correction to the address of the husband also will correct the address for the wife in the normalized model.

Dimensional modeling is the use of facts and dimensions, as we have done in this article. This does not necessarily mean that we are advocating this style of database design. This discussion will be left for future writing.

Footnotes:
1 Refer to http://en.wikipedia.org/wiki/Dimensional_modeling for more description of Dimensional Modeling.
2 Refer to http://www.melissadata.com/enews/articles/0705b/1.htm extracted May 2013
3 Diagrams taken from “The Math That Gaming Made”, Cardno Singh and Thomas 2012.
4 Refer to http://en.wikipedia.org/wiki/Database_normalization#Normal_forms for more information on database normalization.
5 Dr. E.F. Codd in his 1972 paper, “Further Normalization of the Data Base Relational Model”.

Leave a Comment