Authors’ Note: This article explores how customer behavior and demographics are suboptimal when it comes to looking at player behavior. This is called “DvB conflict” and is fundamental to how marketing can be refocused on events that relate to how the customer experiences the gaming offering. One example we will consider is the effect of luck—have they been lucky or unlucky? This fundamental gaming experience differs hugely from the average daily theoretical measurement, which measures the expected value of the game play. Furthermore, the expected value of the game play completely (and deliberately) misses the effect of luck. Finally, we will use obfuscated data to show how the variation in gaming experience on games is likely to vary from trip to trip, and discuss how the player might react to this random effect.
Understanding the gambling experience can be likened to fishing, and this analogy may help explain some of the behavioral aspects of the gambling experience. In recreational fishing, the fisherman invests time (and money) in hopes of catching a fish. But on most fishing trips, the time and money invested in fishing is, at least in our experience, nothing akin to the return that is received. However, the moment of actually catching a fish makes it all worthwhile. In fact, it is for this moment that many fisherman fish, including one of the authors.
Beatrice Cook’s observation in Till Fish Do Us Part (1949), “All fishermen are liars; it’s an occupational disease with them like housemaid’s knee or editor’s ulcers,” is a reflection of the response of fisherman to fishing. The hypothesis is that fishermen remember, talk about and revel in those few moments when they actually catch a fish, and relegate the hours of missed opportunities to a back corner of their consciousness.
When comparing fishing to the gaming experience, it is true that patrons spend considerable time (and money) gaming and that this investment, for the most part, does not result in a winning experience. However, every so often (depending on the mathematical model of the game), the player does win. When applying our fishing analogy, this winning experience is like catching a fish: The player is excited and accomplished, and the win pushes to the forefront in remembering their experience. It is the memory of these “big fish”—or maybe even “the one that got away”—that brings the fisherman to try again.
Expected Catch Size
Theoretical win can also be illustrated using our fishing analogy. Consider this example: If a fisherman catches one fish on average for every six hours of fishing, then then the expected value of his total catch is 1/6 of a fish per hour. However, in reality, the fishing experience is dramatically different. Essentially, the fisherman either catches a fish or does not—there is no 1/6 fish. Furthermore, when a fisherman does catch a fish, he is likely to think he is on a run and will continue to fish longer. In other words, catching a fish (winning) affects the amount of hours spent fishing (gambling), which in turn affects the expected catch size (theoretical win).
Demographics of Fisherman
Looking into the demographic vs. behavioral (DvB)1 data, we can extend our fishing analogy even further, using it to ask some very basic questions. Does it make a difference how old the fisherman is? Does it make a difference what gender the fisherman is? The answer is the same and quite simple: Once somebody has chosen to go fishing, what matters is whether they catch fish or come up empty. This leads to fundamental behavioral questions such as, does the person like to fish? What kind of fishing do they like? Where do they go fishing? How often do they fish? How long do they spend fishing? These questions are core to the behavior of the fisherman.
Connecting this thought process back to gaming, we can see that looking at a player based on their daily theoretical win misses the key to understanding their behavior. In other words, were they lucky or unlucky? The average daily theoretical (ADT) win of a player is an outcome metric, not an optimization metric, as players cannot experience their ADT; they only experience winning or losing.
However, in the casino business, we are fortunate to know a great deal about the buying behaviors of our customers. Other retailers can only dream of tracking 50, 60 or even 90 percent of all customer purchases, but in the casino business it’s rare that rated play (play from customers using a rewards card) doesn’t reach these levels. What’s more, the depth of detail from the player tracking systems is astounding. The results of every customer’s gaming session, from time played to the coin in, the theoretical win and the actual win, are available to be analyzed on a daily basis—and often in real time. In addition, new tracking systems (like the Leap Forward product and the latest release of IGT Advantage) are emerging that can track the result of every single handle pull.
Let’s pause for a moment to reflect on the power of that last sentence. Currently, player tracking systems track the results of every gaming session. If instead we can track every handle pull, we can better model and understand the customer experience and more accurately forecast future customer spend.
Let’s take, for example, two customers who both have the same data in today’s player tracking systems. These two customers played the same game with 10 percent hold. Both had $20 in free play, $1,000 in coin in (thus a net theoretical win of $100 – $20 = $80) and a net actual win of $0 (i.e., a casino loss of $20, but net of free play, they broke even). In addition, both customers had the same rate of play and both played for 1 hour. Now, if we had the ability to track every handle pull, we could look at their experience game by game—which we do in Figure 1. For this example, we will look at these results in 5-minute increments.
As you can see in Figure 1, Player A and Player B had very different experiences. Player A started with $20 in free play, ran it up to $500 after 15 minutes, then slowly played it back to $0 by the 1-hour mark and went home. This player never had to pull a single dollar bill out of his wallet. Now, Player B started off with the same $20 in free play, but immediately went on a terrible losing streak. Player B pulled money out of his wallet, to the tune of $1,000 after 25 minutes of play. Finally, his luck turned, and by the end of the 1-hour session he got back to breaking even.
With this new data, we can easily see that these two players are not necessarily the same at all, despite having the exact same data in today’s player tracking systems. We now know that Player B has the ability and the desire to gamble up to at least $1,000, but we have no idea what Player A will do once he has to make the decision whether or not to go into his wallet. With this data, we would know to certainly continue to monitor Player A, but we would immediately get much more aggressive in our marketing to Player B.
But returning to the behavioral data currently available, let’s examine the heavy reliance that casino businesses place on theoretical win. For this example, we’ll focus on slot machines. As you are likely aware, theoretical hold is an estimate of what a player’s value will be in “the long run.” That is, given the amount of money cycled through the machine (the coin-in or money played) and the theoretical hold of the machine, we can calculate how much money we expect to win off a customer from these two metrics. More precisely, (theoretical win = theoretical machine hold x money played). The primary reason for using theoretical hold is that we are in the gambling business. If we used actual hold, then any time a customer won (which happens on about a third of gaming visits), we’d assign no (or negative) value to that visit and would be unable to allocate any marketing dollars to our winning customers. And since winning customers are a good place to look for return customers, this would be a terrible marketing strategy.
However, theoretical win is not all it’s cracked up to be. We saw in our example that two customers with $80 in theoretical win were actually very different customers. But that was just an example. How different are customers with the same theoretical win, really? The example below speaks for itself (see Figure 2). We took thousands of gaming sessions, each of which had a theoretical win of exactly $100 (rounding down to the nearest dollar), and in Figure 2 plotted the time played (in minutes, along the x-axis) vs. casino actual win (along the y-axis).
We’ve had to remove the amounts to ensure anonymity of the data, but it’s clear that these $100 theo customers actually had very different gaming experiences!
Now let’s move our discussion to demographic data. At first glance, it seems that demographic data is vitally important to better understanding our patrons. Factors like education, age, gender, ethnicity and income levels are, at first glance, a real indication of the likely value of a customer. The challenge is twofold. First, to reiterate what we said in our November 2012 article, “[I]f a bimodal distribution exists, such as we think is the case with gamblers and their gambling dollars per visit, then it is quite simply mathematically dangerous to look at the average.”
Furthermore, demographics are averaging. By their nature, they are looking at things from the perspective of players and drawing the conclusion that others in a similar demographic will have the same behavior. However, we argue that demographics are not useful for predicting gaming behavior due, in part, to a combination of the bimodal distribution of gaming spend, some attempts to build a correlation between demographics and gaming behavior, and that gaming data is full of black swans that make this kind of modeling difficult or inaccurate.
Nonetheless, we have found that once we have discovered an interesting behavioral group, then the demographics of that group are often very useful in helping design marketing programs and further product enhancements. An illustrative example of the logic works like this: ETG players are often younger and tend to be male; however, because somebody is young and male does not mean they will like ETGs.2 So when marketing to ETG players, we need to design promotions that appeal to that demographic, and this targeted marketing is, in our opinion, more likely to succeed. However, when marketing to younger males, it is less effective to assume that they will play ETGs (the majority may prefer some other game). In summary, behavioral analysis is accurate, targeted and can be used very effectively; however, demographic-based marketing is inaccurate and serves a broader purpose.
But what other demographic data is available? In our growing world of social media, we can track Facebook posts, tweets and even customer locations—as long as they are using one of our apps. In addition, there are other services available that track detailed information, like FICA scores. To be sure, we would not want direct access to that data, but we could partner with a secure firm that does have such access. This exciting new data was covered in our February 2010 article, and there can be little doubt that it is on many of our minds right now, as most companies now have a Facebook page, a LinkedIn presence and at least one Twitter feed.
As we mentioned last March, “The key attribute of interaction data is that it happens before and after the transaction,”3 leaving the in-depth analysis for the future and, we argue, taking the risk that social data will become like demographic data. That is, without a direct connection to the transaction and not useful for behavioral-based marketing programs, instead filling the general awareness marketing or brand marketing roles. (Some methods of handling this were covered in the April 2012 issue of CEM.)
The Behavior of Fisherman
At the end of the day, as the industry strives to handle new data streams, the one big thing that has changed and is improving by the day is the availability of data—data that shows the true behavior of our patrons. The fishing analogy is a simple one that can be used to explain the effects of luck in terms that are less mathematical; irrespective of how understanding is gained, the key is that now operators can see product preferences, market baskets and other critical understanding of patron choices, and this understanding can enable us to move far beyond the revenue-based models that have become a staple of the industry. We encourage you to keep reading, thinking and trying these new methods as they take us into an age of understanding of customer behavior.
1 “Where’s the Money Now, Part 1: The Youthful Player,” Cardno Thomas Evans, January 2013, Casino Enterprise Management.
2 “Where’s the Money Now, Part 1: The Youthful Player,” Cardno Thomas Evans, January 2013, Casino Enterprise Management.
3 “Where’s the Money, Part 9: Big Data,” Cardno Thomas, March 2012, Casino Enterprise Management.