Let's Talk Turkey: Applying Retail Market Basket Analysis to Gaming

Imagine a customer pushing a steel grocery cart (with its front-wheel wobbling, of course) through any one of Las Vegas' luxury casinos. But instead of filling it with frozen vegetables, milk and meats, casino customers fill it with sundries, gifts, restaurant foods and beverages, lodging, and wagering products. Casinos have arguably one of the widest varieties of entertainment products and service offerings under one roof. The casino is no longer the center of revenue generation for our products. While games generate the highest margins, customers now expect more. This trend of increasing the diversity of the market basket is not limited to the gaming industry. Amazon.com, for example, has gone from being strictly a bookseller to offering thousands of products and services in scores of categories. This leads us to define the market basket for casinos as being the customer experience defined by time and product consumption choices made by customers during each visit.

Traditional retail stores spend a great deal of time studying the mix of products people buy, and make critical product placement, merchandizing, procurement, staffing and marketing decisions based on what is termed "market basket analysis." Think of last month's Thanksgiving promotions at the grocery store. Stores know that by selling turkey at a loss, they can get customers to fill their baskets with other traditional items such as cranberry sauce, beer, wine, liquor, soft drinks and deserts--in many cases at a high profit margin. The loss on the turkey is more than made up for by the other purchases. This is a good example of a loss-leader promotion.

It is our belief that applying retail market basket analysis techniques to a casino's product and service offerings may also prove beneficial to the bottom line. It has the potential of uncovering cross-selling opportunities, optimizing game mix and placement, managing comps (e.g., how much the casino room rate should be for a certain player), and personalizing offers (e.g., offering beer immediately to poker players in an area of the casino where beer-loving poker players usual play).

Market basket analysis can get very complex very quickly. Think about all of the combinations of items that may be purchased during a single visit to a casino or grocery store. Add to this other variables such as price, quantity and location, and the possibilities become mind-boggling. That is why statistical algorithms are used to identify the patterns within the baskets. A common technique used is association rules. The result of this type of analysis forms an "if this, then that" statement. In other words, if a customer buys Product X, then that customer will likely buy Product Y too. To illustrate this technique, let's start with a simple and familiar example.

Suppose we have some sales information from a roadside stand that sells three products: yams, cranberry sauce and turkey. During the period observed, eight customers each shopped at the stand once. Table 1 summarizes their purchases. (We will assume each customer bought no more than one package of each product and that this sample is representative of all customers.)

This simple set of data can be analyzed manually. For example, it is easy to see that most people--seven out of eight--bought turkey (this is the stand's specialty). Five of those seven turkey buyers also bought yams, and four bought cranberry sauce. With this knowledge, how can we help the proprietor best place the products in the stand? It seems best to place the yams near the turkey, because 63 percent of all customers and 71 percent of turkey-buying customers purchased both of these items. Furthermore, turkey should be placed in the middle because 57 percent of turkey buyers also bought cranberry sauce. This middle placement is also convenient for the two customers who bought all three products--and could possibly encourage this behavior. Only one customer bought only yams and cranberry sauce (13 percent of all customers), so it is not likely to hurt business if these items are placed far apart from each other.

An "if this, then that" association rule that results from this analysis would be: "If a customer purchases turkey, then the customer is likely to also purchase yams." In the example above, we referred to some percentages. Information like this, called "support" and "confidence" factors, is typically provided with each association rule. The support factor for our association rule is the percentage of transactions that included both turkey and yams. We calculate this by dividing the number of transactions that included both turkey and yams by the total number of transactions, which in this case is 63 percent. The confidence factor is the percentage of people who bought turkey and also bought yams. We calculate this by dividing the number of transactions including turkey and yams by the total number of transactions that included turkey, which in this case is 71 percent. These factors give us an indication of the accuracy and usefulness of our rule. Typically, analysts look for rules that generate at least 50 percent support and 50 percent confidence levels.

The full association analysis of this example would include 12 rules (if turkey, then yams; if turkey, then cranberry sauce; if turkey and yams, then cranberry sauce, etc.) with varying confidence and support factors. The equation for calculating the number of association rules is R = 3d - 2d+1 + 1, where "R" equals the total number of association rules and "d" equals the number of distinct products. Looking at this equation, we can see that the number of rules increases exponentially with the number of available products. Let's now apply market basket analysis to a simple casino problem.

First, it is important to distinguish between the grocery stand products we have described above and casino gambling product, especially slot machines. When one purchases yams, for example, the significant characteristics that the consumer is weighing are the price, the quantity and the quality of the product. But for multi-denomination slot machines, for example, the physical characteristics (e.g., video poker versus video reel) will not adequately describe the product. The customer may choose the price, but the price does not affect the quantity or quality of the product offered. In other words, on a penny slot machine, the quality of the game is the same no matter how much is bet, and the quantity purchased for the price remains the samea single game is played without regard to the size of the bet. Downloadable games provide the consumer with even more control over the price and quality of the product. This may result in the same physical game being classified as several different products within the market basket, based on how customers interact with it.

Now let's say we are looking at customer purchases for seven separate individuals' visits to a certain small casino. The casino offers four products: two types of slot machines (as pointed out above, these types may be defined by the customer's interaction with them and not simply their physical characteristics), a hotel, a single restaurant and one table game.

We are analyzing anyone who has made at least one purchase (placed a bet on a slot or table game, spent a night at the hotel, or ate a meal in the restaurant) inside the casino. Table 2 summarizes the purchases used for this example.

In this example, there are 50 possible association rules, but few meet the minimum 50 percent support and 50 percent confidence criteria.

For example, market basket analysis reveals that 57 percent (four out of seven) of all customers analyzed played Slot 2 and stayed in the hotel, while 100 percent (four out of four) of those that played Slot 2 also stayed in the hotel. The association rule would therefore state: If someone plays Slot 2 games, then they will always stay in the hotel. Conversely, the second association rule that meets the minimum support and confidence specifications states: If someone stays in the hotel, then they play usually play Slot 2 games.

Other interesting associations are:

" If someone stays in the hotel, then they always play at least one slot (100 percent of people who stayed in the hotel played either Slot 1 or Slot 2).

" If someone visits the casino, then they always play Slot 1.

Negative associations can also be interesting. For example: If someone plays table games, then they do not stay in the hotel or play Slot 2 games. This negative association, however, is suspect because the support factor for the market basket that includes table games (if table games, then Slot 1) is only 29 percent (two out of seven), which is well below our 50 percent threshold.

Without getting too hung up on cause and effect (e.g., wondering if there is something magical about hotel rooms that make people want to play slots), concentrate on what actually occurred and come up with ways to capitalize on the knowledge. For example:

Casino Marketing Department

* Because slot players seem to also stay in the hotel, offering a multiday slot tournament for Slot 2 games that includes a hotel stay will likely create interest.

* Because 100 percent of the players played Slot 1, redirect marketing dollars to encourage Slot 2 play, especially if its hold percentage (the amount the casino statistically expects to win over the long run) is higher than Slot 1.

Slot Department

* Place Slot 2 machines close to the hotel elevators.

* Provide maps of the locations where players can find Slot 2.

* Be flexible regarding where Slot 1 games are placed, because they are played on each visit. Also, because Slot 1 game placement is already optimized, look at other factors to explain why Slot 1 games are always played.

Table Games Department

* Though not conclusive, the analysis suggests that, in general, room comps may not effectively encourage table game play. Room comps should be given out with more discretion (e.g., looking at the individual's record).

* The table game product itself could be altered to include different side bets, jackpots and game varieties that more closely resemble slot machines.

* It may be possible to increase the casino's room rate for table games players.

Casinos have been characterized as small cities that offer hundreds of products and services to their customers. As we have shown, analyzing so many products requires complex analysis and advanced tools. Market basket analysis is a mature data mining technique, so a great deal of information and tools are already available. Market basket analysis is also intuitive. The results are in plain English, allowing practitioners to make adjustments and interpretations by applying their own industry experience.

But before diving into market basket analysis with both feet, we suggest you start small. For example, a restaurant manager can start by studying which desserts are commonly purchased with particular entrees, then have waiters suggest the corresponding dessert at the end of a diner's particular meal. This can probably be determined with a simple chart similar to the one used in our casino example. Later, other elements can be added. For example, do people order different combinations on the weekends than they do during the week? Do people dining at the bar order different combinations? By performing even these simple analyses, a great deal of actionable information about customers' behavior may be discovered. Combining these observations with the industry experience that got you where you are today will really put you in a position to "talk turkey."

Bart Lewin has more than 25 years of experience in the Engineering and Information Technology field, holding serveral technical and executive technical management positions. He is currently a technical and management consultant. He can be reached at balewin@mac.com.

Dr. Ashok K. Singh has taught statistics, mathematics and operations research courses at New Mexico Tech, Socorro, N.M., and statistics and mathematics courses at University of Nevada, Las Vegas. He has over 75 publications in theoretical and applied statistics.

Andrew Cardno has more than 16 years of experience in business analytics, ranging from modeling healthcare drive times to casino gaming floor analytics. He often presents on the future of analytics across the world and has spent the last seven years living in the United States and working with corporations around the world.

All articles are subject to the terms in our copyright statement.