Big Data in Gaming

What is big data

In the data world of last century the huge data was financial in nature and the grain of the data was typically one record per item (or number of items) sold. For example; in a retail world one grain of data was a single product/price/number of items sold.This century the data volumes are amplified to a level that makes lasts centuries data look tiny. This century the each grain of transaction data, is surrounded by thousands of interactions. These interactions exist in web logs, in social media and in fact in nearly everything that we do, extending to remote sensing from satellite imagery.  In addition, the grains themselves are no longer necessarily single numbers representing some aspect of a transaction – often data comes to us in the form of unstructured data (imagine for example trying to store every single webpage on the internet every day), which has forced some data collectors to move away from relational databases and towards other forms of data storage like Hadoop.

In the world of gaming the transaction data includes two main pieces, firstly the gaming operational data and secondly the player tracking data. The operational and player tracking information is related with limitations.  A couple of examples and their consequences:

1. On multi game devices the multi game data is not generally shown by player, and in some systems its not even shown by machine (in other words, if a machine contains a 1cent version and a 2cent version of the same game, many slot systems will report that the machine won $300 yesterday, but may not be able to report that the 1cent version won $250 and the 2cent version won $50).  This is problematic not only for slot operators trying to optimize their slot floors.  Far worse, marketers are often over- or under-valuing customers who play these multi-denom / multi-game machines – which are taking up an ever increasing share of many slot floors.  Imagine a machine that has 2 games – Game A with 10% hold and Game B with 2% hold.  Many player tracking systems won’t know which game the customer is playing, so they simply average the hold percents of the 2 games and assume the machine holds 6%.  What this means is that no matter who is playing this game, we are guaranteed to be assigning the wrong theoretical win to this customer!  A customer with $1,000 of coin in on Game A should have a theoretical win of $100, but instead the system reports $60.  A customer with $1,000 of coin in on Game B should have a theoretical win of $20, but instead the system still reports $60.  So, our player tracking system is going to cause us to either under-invest in this customer by 40%, or over-invest by a whopping 300%! (The authors are fascinated by the fact that these player tracking systems still dominate the industry.  Imagine a slot floor with 70% multi-game / multi-denom.  Now imagine the slogan for one of these systems “correctly tracking your players 30% of the time!”)

2. Session-based player tracking.  Currently most player tracking systems aggregate a customer’s play by slot machine by day.  A significant amount of information is lost by this method of tracking.  Imagine 2 customers, both of whom lost $60 over the course of an hour of play.  The first customer put $60 into the machine and then slowly lost it over the course of the hour.  The second customer had a much more volatile session, losing $1,000 at minute 50, then hitting a jackpot to recoup most of that to finish with a $60 loss.  These are obviously 2 very different sessions, but one wouldn’t know this looking at the player tracking data.  

Fortunately there are efforts to correct this in the industry and replace the current player tracking systems with tools that track each and every wager that the player makes on a slot floor, rather than summarizing by machine, solving both of the aforementioned issues simultaneously.