BY VLADISLAV HVECKOVICS, CIO & CO-FOUNDER OF SOFTGAMINGS
Given today’s stage of technology development, it’s only logical to start using more sophisticated tools for your player’s data analysis. Advanced data analysis is based on the premise that given the complexities and volume of data, a human selecting attributes for promotions and creating rewards might do a much worse job than a computer. A computer cananalysehundreds of parameters without bias, mistakes or omissions. So if you invest into such data analysis, you really invest in the future of your venture.
If you still feel the paragraph above is too abstract, here is a short list of things that data analysis can do better than a human: selecting a list of players for a particular promotion based on hundreds of parameters, creating clusters or segments of players with their own promotion chains on the fly, predicting players attrition before the players even make a decision to move, and so on.
So, let’s say you decide to try this out, how can you start and what strategies really work?
Firstly, a dedicated team or even one person is the way to go. If your team does this type of work as a 2ndor 3rdpriority, you most likely won’t get the results you will be satisfied with. The team or the person should be both proficient enough in using the statistical methods and have a good understanding of the sportsbook product and players’ psyche.
Inexperienced and new teams start overcomplicating things straight away by trying to build an autonomous system using the most expensive and advanced tools on the market. I would suggest starting simple; we have seen successful operations built on downloading an extract from data to a CSV file and running analysis on that file, generating parameters or extracts in another CSV file which can then beanalysedmanually. Once you see that your models work or adjust accordingly, and you have run the process a couple of times you will be ready to make improvements.
Secondly, you will need to choose a technology. I strongly suggest starting with either Python or R. And in choosing between those two I would suggest Python, just looking at the user base of the language. Something like Python SciKit should be more than enough for you to start.
Thirdly, we suggest building success criteria and methods for checking the usefulness of the statistical methods straight away. You can use regression analysis, decision trees or clustering; the science behind them might seem incomprehensible for the uninitiated, but in the end if all this science works no better than random or even the manual segmentation based on simple 4-5 variables, it should be improved. So some sort of control group is a must.
Once you have the foundations – a dedicated team, tools and data, the experimenting starts. As you have good criteria for measuring success in relation to your previous methods (or even compared to doing nothing), you will have a good feeling as to what works and how well it does. And once the management sees the results attributable to your work, there will be appetite for more.
Finally, once you show top management how successful the data analysis can be, there will be time for automation, purchasing more advanced software or building your own tools, and hiring more people for data analysis and running promos.