Home Where’s The Money Now? Part 13 of 18: Crazy Campaign Relationship Management

Where’s The Money Now? Part 13 of 18: Crazy Campaign Relationship Management

Authors’ Note: This article will look at the specialized marketing tools that the casino database marketer needs, building on the ideas introduced in Part 12 of this series. We look deeper into test and control and why a casino has a need for these methods. We explore this by illustrating the needs of a mythical retailer and how it would be considered crazy if it did what gaming database marketers do every day. This article also expands on some of the ideas about customer marketing developed in Professor A. K. Singh’s and our book, The Math That Gaming Made.

In the world of science, a “control” is the way of measuring the results of an experiment. Formally, a scientific control is an experiment to minimize the impact of all variables except one (the independent variable). Scientific control is a critical part of the scientific method, and it is well established as one of the only ways of measuring the impact of a variable on an outcome (the dependent variable).

But let’s think about it a little less formally. The word “experiment” conjures up an image of scientists in white lab coats running about with large arrays of test tubes, and this image can help us understand the idea of control. Consider this illustration: In the laboratory, a scientist sets up two identical potted pine trees, and she then performs the test—increasing the CO2 level on one pine tree. She leaves the other tree with normal CO2 levels as the control. Each week, the scientist can measure the growth of the pine trees and show the difference between the two as being the effect of CO2. But what if CO2 wasn’t the only difference? A confounding event is one that affects the outcome of an experiment, essentially messing up what would otherwise be a nice experiment. For example, say the scientist also varies the amount of water given to the plants, not just CO2. If the plants vary in growth, we do not know if it is the water or the CO2 that caused the growth variations.

In the gaming world, we can think of the changes we make as we evolve our businesses as a series of experiments. For example, adding a new slot machine, running a new kind of bounce-back marketing offer or changing the opening hours of the property’s restaurants are all experiments of one kind or another. But in the gaming world, our experiments are certainly not simple. People do not live in controlled laboratories, and our world is full of distractions and other confounding events.

The Difficulty of Control
Gaining control in marketing, for gaming, is very difficult for at least four reasons: (1) patrons are varied, (2) competition is intense, (3) patrons have choices and (4) the gaming data itself is volatile. Let’s expand on each of these reasons.

First, gaming customers can be highly varied in their cash, visitation and other patterns. Consider the amount wagered, for example. If two patrons are sitting at identical gaming machines, one of those patrons could be wagering 200X as much as the other per game. Or consider blackjack legend Kerry Packer: “It is said that Packer almost broke the MGM Grand, which resulted in the firing of several prominent employees who let him make huge bets.”1 We argue that gaming stands alone as the industry where consumers are the most varied in money.

Customers are also very volatile in their responses. Many factors influence volatility, but one that stands out above all others is competition. In most jurisdictions, customers are inundated with marketing offers, and this multi-offer environment causes a kind of cherry-picking response. The upshot of this cherry-picking behavior is what seems to be a very random response to marketing activities.

Unlike essentials of life such as electricity or food, gaming is optional. Customers choose when to gamble, how much to gamble and where to gamble. This choice, quite simply, makes casino customers volatile in their response to marketing promotions compared to other industries.

Finally, gaming data itself is also highly volatile. A customer shopping for clothes has a goal and a budget, and if he intends to spend $20 for a shirt, he goes into a store and spends $20 on a shirt. Gaming customers have no such behavior. Customers may have a certain amount of free time, they may have a certain “stop-loss” budget set or they may simply give up with time and money to spare if they go on a losing streak and feel unlucky. They also can walk out of the casino with more money in their pockets than when they entered! (In fact, our experience is that this happens about one-third of the time.) These factors all make trying to measure the “spend” of the customer very difficult, and we have written articles about how the industry standard metric of “theo” is highly problematic. (For example, see Figure 2 in “Where Is The Money Now? Part 2” to see why $100 in theo is not very representative of the customer experience.)

When it comes to data analysis, we need to be careful about what we consider a control. Extending last month’s2 discussion on year-over-year (YOY) analysis, let’s examine why this kind of analysis should not be considered to be control in any way. To illustrate why YOY analysis is not control and cannot, on its own, be used for determination of marketing results, consider the figures in Table 1.

At first glance, the revenue figures in Table 1 look fantastic, and executives may be happy with this result. But now let’s take a closer look, considering the confounding factor of market share. Looking at market share, we see that the property’s share is static. Market size is another confounding factor, and when considering the market size increase, it appears that the business is growing at a rate equal to the market. This is very different than saying that business is up 20 percent. Now the executives should still be very happy, as they have been able to capture the right amount of growth compared to their competition, but they had a huge tail wind bolstering their revenue—namely, market growth. Clearly, using year-over-year comparisons without also considering market share and size is a mathematically dangerous way of looking at the business. In fact, any analysis we do needs to consider how we isolate changes in the market.

A Crazy Marketing Program

Imagine this “crazy” marketing program: A retailer, let’s call it Retail A, decides to mail out a cash offer. Basically, this offer says “if you come and buy something small, I will give you $100 cash.” With this kind of offer, Retail A is highly likely to get a tremendous response. In fact, it is quite likely that people will be lining up at the doors to redeem their offer. Customers will be saying that this offer from Retail A is both crazy and very different from usual coupons or volume pricing discounts.

Campaign Relationship Management (CRM) tools have grown up in a world of companies like Retail A, where offers are sent, redemption rates are measured and margins on price are reported on. But gaming is nothing like this. In gaming, we give free play, which is basically cash, and there is no set pricing for the gaming experience, as customers choose how much they gamble and for how long. Nongaming-adapted CRM tools are simply designed for a different task. Let’s compare how a nongaming direct mail business is quite different to the gaming industry (see Table 2.)

Designing the Experiment

In traditional direct marketing test and control, retailers have a few segments of customers containing very large samples. Test and control are therefore quite simple for these retailers:

1. Hold out a small percentage of customers (control).
2. Compare response rates for the customers with an offer (test) to those without an offer and see if the offer drove a significant lift in response.
3. Measure the incremental value in the revenues associated to the offer, less the discounted expense, to see if the program was profitable.

In the casino world, if we were to follow that example, we would end up with unbelievably inconsistent results for the following reasons:

1. Gaming results are highly volatile (as discussed above).
2. Casino segmentation based on recentness, frequency and spend often creates hundreds of segments, the most valuable of which contain very small quantities of players.

Thus, a random sampling of customers is highly likely to produce unbalanced test and control groups. Let’s devise a simple example to see why.

Imagine that we have segmented our business and decided we want to test a $20 free play offer versus a $30 free play offer for one of our mid-tier segments. (Keep in mind, the profitability of a casino’s mid-tier customer segment is equivalent to a high-value customer at a retailer, due to the razor-thin margins faced by retailers.)

This segment contains 400 players with an ADT ranging from $200 to $300 and one to four average monthly visits. This is a typical casino segmentation strategy, and yet the monthly value of customers in this segment ranges from $200 to $1,200. So, it is not at all uncommon if we were to take a 10 percent control sample to end up with results in Table 3 before we even mail the offers.

In this example, due to the wide variation in the customers of our segment and the small size of the control group, the customers in the test group are simply better customers. With 10 percent higher visits per player and a 21 percent higher ADT, these customers are 33 percent better than the control group. Now, how are we supposed to know if a $30 offer is better than a $20 offer if the test group is already set up to outperform the control group?

There are two answers to this question. First, as the sample sizes of segments get smaller, we need to have larger control groups, approaching 50/50 splits in many cases. Secondly, we need to employ “balancing.”

Balancing Test and Control

In the previous section, we saw the need to ensure that the casino test and control experiment was balanced, meaning that the play behaviors of the test group and the control group are the same. Tables 4 and 5 two quick examples: one is balanced, and one is unbalanced.

As we can see, in our unbalanced example the test group is set to outperform the control group by 5 percent, enough to significantly impact the results of the experiment. In the balanced example, all the behavioral data is within 1 percent, so we are ready to conduct our experiment.

Conclusion
Marketing in gaming is, at the very least, very different from marketing in other businesses. This environment drives quite different needs for tools and approaches. We think this has resulted in many trying to bring outside-the-industry tools and approaches to gaming, only to find themselves with some very strange results. While the gaming industry can learn from these outside approaches and adapt to them, it always needs to be done with a clear eye on the specific challenges that gaming presents.

The question is, does your CRM tool allow you to do test and control in the world of casino marketing, and if it does, can you ensure that your tests are balanced?

Footnotes
1 Extracted from http://blackjacklife.com/blackjack-legends-kerry-packer/ on December 2013.
2 CEM December 2013 Cardno Thomas Part 12 of 18 “Where is the Money Now?”

Leave a Comment