Authors’ Note: This is the second article in a subseries to investigate “Big Data.” Locational intelligence is the data that joins brick-and-mortar businesses to social media-centered interaction data explosion. For an introduction to Big Data, please see the March 2012 issue of CEM.
The entertainment industry is firmly in the business of selling experiences—hopefully, exciting experiences. It is simply wrong to compare this to the retail industry, which is in the business of selling stuff. Of course, when shoppers shop in a retail outlet, they gain an experience, but the main goal of shopping is to arrive home with one’s purchases. Entertainment is quite simply different. The main goal is to have an experience, and this is not something that somebody else can do for you—it is a personal experience. This fundamental difference between the industries also results in a fundamental difference in how we can use the data we collect about our customers.
Monitoring Twitter and Facebook1 has shown that there is a huge volume of information being generated by our customers regarding their gaming experiences. In addition, there is an even more significant volume of information being circulated by our customers as they react to other entertainment experiences. Looking at the volume of communication and its seemingly random nature can seem overwhelming, even more so when we consider that people communicate about entertainment at volumes that probably swamp all other aspects of their online communication. This communication largely takes place via social media websites and is about people sharing experiences. In many ways, the entertainment industry is at the center of an information storm, and it seems reasonable to speculate that a huge portion of social network interactions relate to the entertainment industry. For example, with a single YouTube video, a singer like Susan Boyle, a person previously completely unknown to the world, can become an overnight sensation and the center of global discussion.
So how do we start to sift through all of this data and use it to drive business? One of the most important, and easily actionable, data streams for casinos today and into the future is locational data. Let’s take a look at how this data can be used in conjunction with interaction data and transaction data to unlock the value in the Big Data.
Locational data was once strictly the domain of land surveyors and geographers. Today, it is generated by nearly every application on any mobile device. Consider this example: A customer named Andrew is sitting at a slot machine in his favorite casino, Casino Y. He stops playing for a moment and uses his smartphone to search online for a place to dine. Andrew is creating GPS data that records his current location2 and the fact that he is looking for a restaurant. The app provider, to generate advertising revenue, makes this data available to third parties, who in turn use it to place their marketing offers and events. So, Casino Z could be watching for these searches and sending locationally targeted marketing events to Andrew. Figure 1 shows a screen shot of the Foursquare application Andrew used for his search. As you can see, he is shown several location-based offers in his vicinity—most of which are not at Casino Y.3
Figure 1: Location-Based Offers on FoursquareThis interaction data4 is even more interesting because it is managed by another party, in this case Foursquare, which will provide advertising services to almost anybody who is willing to pay. This interaction data is therefore very different than the transaction data that drove our marketing efforts in the last century: Our competition has access to the same resource.
While Figure 1 shows the customer placed on a street map, which is the traditional spatial view, with the introduction of indoor-capable locators, we can now see exactly where we are inside a retail space or casino. One does not need to look much further than Google Maps to see some of the latest work in this indoor space. (See Figure 2.)
Figure 2: Inside Spatial View from Google MapsFigure 2 shows how merchandise is laid out inside a retail store and tracks the customer’s movement as they navigate the retail space. Once again, this data does not belong to the owner of the retail space, but to the individual customer—and is often shared with the app provider, in this case, Google Maps.
Where Are My Customers?
It is well established that proximity to competition is one of the primary drivers of business, and now census data can give us reasonably accurate market share numbers. With the 2010 census data now available, we can see the number of potential gamers by census tract across a market catchment. The knowledge of where the customers are is interesting, but what is fascinating is the relationship that this has to the catchment of your competition. Quite simply, we can now see the effect of competition on our market share in terms of both visits and revenue. Taking this one step further, we can see big data effects in many cases. For example, if the customer uses our Casino X application to search for upcoming shows or events, we can see where that customer was when they did the search. We can now see the relationship between where our customers searched and interacted with our property, and where they were located.
Location stands at the center of this analysis, with its special ability to link seemingly unrelated data sets, such as census information and online searches. To manage this data and these relationships, we need locational storage, locational query and locational analysis tools. These tools form the backbone of the locational analysis systems of the future.
Collecting the Data
To begin to make sense of this new frontier in gaming data, we need to understand what data we can, and cannot, collect about our customers. With the advent of locational services on mobile phones, we can ask our customers to opt-in to sending us locational data, accomplished via a casino app. So, assuming we are able to design an application that engages a significant portion of our customers, we can begin collecting this important new data stream. Many casinos are already doing this, presenting apps that not only describe the various services the casino has to offer, but also interacting with the customer and making them eligible for additional offers when they use the app. Basically, casinos are now doing via smartphone what they’ve done for decades—bribing customers to encourage increased engagement.
Now we have an increased wealth of data about our customers. In addition to all the transactional data we get while they are on our casino floor, we have locational data whenever they use our mobile app. This explosion increases the dimensionality of our data exponentially, as discussed in our last article.5
Mining the Data
Enter the world of predictive modeling. With all this gaming data and locational data, we need to figure out how to extract value from it. Let’s think about what we have to work with, and what we are trying to accomplish.
We know, or should be able to know with some effort, the following types of information:
• Gaming behavior – We can track every transaction on our gaming floor.
• Retail, F&B and entertainment behavior – We can track purchases in our non-gaming outlets.
• Basic demographic information – Age, gender, address and other contact information.
• Appended demographic information – We can purchase additional information about our guests, such as income, magazine preferences, presence of children in household, car purchases, etc.
• Locational data – We know where they are when they use our mobile app.
With this information, we can start to ask questions. Where are our customers when they use our app, and what are the consequences of this action? Do they use the app when they are near the competition, then come to gamble with us? Do they use the app when they are on property? Does using the app increase or decrease their play? And how does all the other data (non-gaming, demographic, etc.) change how the customer’s interaction with the mobile app relates to their gaming behavior?
These questions are all answerable via predictive modeling. To build a predictive model, we need input data, output data and a lot of outcomes. Let’s tackle one of our questions as an example: Does using the app increase or decrease customer play? In a simplified version of this model, we first need to organize the data in a way that the above question can be answered by a computer model.
First we input our data. We have loads of gaming data, but we are trying to determine if the use of the app increases or decreases customer play. So, for each customer we need to collect all the gaming data prior to his or her use of the app. [Note: We won’t get into the structure of how to organize this data—that is beyond the scope of this article.] In addition to the gaming data, we have other behavioral data like retail purchases, and again, we need to collect this information prior to the use of the app. On top of all this we have basic and appended demographic data, which is usually time independent. And, finally, we have the location(s) where the customer used the app.
From this soup of input data, we also need to determine our output data. For this simplified example, let’s attach a “1” to any player whose average daily play increased after using the app and a “0” to any player whose average daily play did not increase. This gives us what is commonly called a base modeling table. From this table, people who are experts in predictive models can weed out which metrics are predictive of a customer increasing his or her play after using the app, which metrics are irrelevant, and which weights to apply to the relevant metrics.
Now the fun starts! It’s time to apply the model. When doing this, we need to keep in mind that when a casino rolls out an app, it often takes time for customers to adopt usage of the app. In the beginning, a small fraction of customers, perhaps 1 or 2 percent, will actually use the app. Even as adoption grows, this percentage will likely remain below 50 percent for years to come. Nonetheless, for our current customers, we can learn which ones are worth encouraging (strongly, via offers) increased app usage. We can also determine when to use the app to push offers based on whether the customer is close to a competitor, close to our casino, or whatever our models tell us is relevant.
In addition, we have non-customers that we want to take from our competitors. The predictive modeling described above can assist with this, too. For non-customers, we can tweak our models to remove gaming and on-property non-gaming data, and then understand how demographic and locational data can combine to tell us how to best leverage mobile tools like Foursquare—getting the right offer to the right person at the right time. These models can be taken further by combining broad sweeps of Internet-trawled data with companies like PredictivEdge6 to build predictive analysis on this data.
Finding the Money
People need to eat, people need to socialize, and people need people. We need to look no further than the movies to see how people still choose to go to the cinema and enjoy a night out rather than stay home. As we have illustrated, location can tie our customer interactions together with the broad world of social data. Clearly, location is a truly differentiating feature of a brick-and-mortar business, and it is time to move your brick-and-mortar data into the GPS world.
1 “Gaming Interactions: The Invisible Force of Social Networks,” Singh and Cardno, CEM, February 2010.
2 Inside data is less accurate, as it is often based on WIFI locational services.
3 Screen shot taken from Foursquare, February 2012.
4 “Gaming Interactions: The Invisible Force of Social Networks,” Singh and Cardno, CEM, February 2010.
5 “Where’s the Money? Part 9,” Cardno and Thomas, CEM, March 2012.
6 Reference used with permission of Bill Thompson. www.predictivedge.com.